mixed land use and obesity: an empirical comparison of alternative land use measures and geographic...

This article was downloaded by: [University of Chicago Library]On: 18 November 2014, At: 06:15Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,UK

The Professional GeographerPublication details, including instructions forauthors and subscription information:http://www.tandfonline.com/loi/rtpg20

Mixed Land Use and Obesity:An Empirical Comparison ofAlternative Land Use Measuresand Geographic ScalesIkuho Yamada a , Barbara B. Brown b , Ken R. Smith b

, Cathleen D. Zick b , Lori Kowaleski-Jones b & JessieX. Fan ba University of Tokyob University of UtahPublished online: 27 Jun 2011.

To cite this article: Ikuho Yamada , Barbara B. Brown , Ken R. Smith , Cathleen D.Zick , Lori Kowaleski-Jones & Jessie X. Fan (2012) Mixed Land Use and Obesity: AnEmpirical Comparison of Alternative Land Use Measures and Geographic Scales, TheProfessional Geographer, 64:2, 157-177, DOI: 10.1080/00330124.2011.583592

To link to this article: http://dx.doi.org/10.1080/00330124.2011.583592

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all theinformation (the “Content”) contained in the publications on our platform.However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness,or suitability for any purpose of the Content. Any opinions and viewsexpressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of theContent should not be relied upon and should be independently verified withprimary sources of information. Taylor and Francis shall not be liable for anylosses, actions, claims, proceedings, demands, costs, expenses, damages,and other liabilities whatsoever or howsoever caused arising directly or

http://www.tandfonline.com/loi/rtpg20

http://www.tandfonline.com/action/showCitFormats?doi=10.1080/00330124.2011.583592

http://dx.doi.org/10.1080/00330124.2011.583592

indirectly in connection with, in relation to or arising out of the use of theContent.

This article may be used for research, teaching, and private study purposes.Any substantial or systematic reproduction, redistribution, reselling, loan,sub-licensing, systematic supply, or distribution in any form to anyone isexpressly forbidden. Terms & Conditions of access and use can be found athttp://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14

http://www.tandfonline.com/page/terms-and-conditions

Mixed Land Use and Obesity: An Empirical Comparison of

Alternative Land Use Measures and Geographic Scales∗

Ikuho YamadaUniversity of Tokyo

Barbara B. Brown, Ken R. Smith, Cathleen D. Zick, Lori Kowaleski-Jones,and Jessie X. FanUniversity of Utah

Obesity is a growing epidemic in the United States. Walkable neighborhoods, characterized as having the threeDs of walkability (population Density, land use Diversity, and pedestrian-friendly Design), have been identified asa potentially promising factor to prevent obesity for residents. Past studies examining the relationship betweenobesity and walkability vary in geographic scales of neighborhood definitions and methods of measuring thethree Ds. To better understand potential influences of these sometimes arbitrary choices, we test how fourtypes of alternative measures of land use diversity measured at three geographic scales relate to body mass indexfor 4,960 Salt Lake County adults. Generalized estimation equation models demonstrate that optimal diversitymeasures differed by gender and geographic scale and that integrating walkability measures at different scalesimproved the overall performance of models. Key Words: geographic scales of neighborhoods, land usediversity, obesity, walkability.

La obesidad es una epidemia en expansin en los Estados Unidos. Los vecindarios transitables, que se car-acterizan por tener las tres D de la condicion de caminables (Densidad de poblacion, Diversidad de uso delsuelo y Diseno amigable para el peaton), han sido identificados como factor potencialmente prometedor paraprevenir la obesidad de los residentes. Estudios anteriores que examinaron la relacion entre obesidad y lacondicion de caminables varan en cuanto a las escalas geograficas para definicin de las barriadas y los metodosde medicion de las tres D. Para entender mejor las influencias potenciales de lo que a veces son escogenciasarbitrarias, hicimos pruebas para ver como cuatro tipos de medidas alternativas de diversidad de uso del suelo,tomadas en tres escalas geograficas, se relacionan con el ndice de masa corporal de 4.960 adultos del Condadodel Lago Salado. Modelos de ecuaciones de estimacion generalizada demuestran que las medidas ptimas dediversidad diferıan por genero y escala geografica y que integrando medidas de la condicion de caminables adiferentes escalas se mejoraba el desempeno general de los modelos. Palabras clave: escalas geograficasde vecindarios, diversidad de uso del suelo, obesidad, condicion de caminables.

∗This research was supported in part by Grant Number 1R21DK080406–01A1 from the National Institute of Diabetes and Digestive and KidneyDiseases (NIDDK), the National Institutes of Health (NIH), and Grant Number 3R21DK080406–02S1 issued under the American Recoveryand Reinvestment Act (ARRA). The content is solely the responsibility of the authors and does not necessarily represent the official views ofNIDDK or NIH. We would like to thank Jonathan Gallimore and Carol Werner for land use coding assistance, Tim Noyce for land use codinginformation, Linda Keiter for her assistance with the Dun & Bradstreet data, and Larry Frank and Daniel Rodriguez for useful input. We are alsograteful for the anonymous reviewers who provided helpful comments on an earlier version of the article.

The Professional Geographer, 64(2) 2012, pages 157–177 C© Copyright 2012 by Association of American Geographers.Initial submission, September 2009; revised submission, April 2010; final acceptance, June 2010.

Published by Taylor & Francis Group, LLC.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14

158 Volume 64, Number 2, May 2012

T he obesity epidemic is firmly entrenchedin the United States (Mokdad et al.

1999; Blanck et al. 2006; Ogden et al. 2006),with 35 percent of adults considered obese(Hedley et al. 2004; Flegal et al. 2010). Therapid rise in obesity points to contextual causesand has prompted a search for environmen-tal factors that encourage physical activity andprevent obesity (Hill and Peters 1998). Neigh-borhood walkability, the physical environmen-tal supports for walking, has been identified asan especially promising research direction forbetter understanding the rise in obesity in theUnited States. Walking is relatively safe, easy,and affordable. Individuals report that walking,especially in their neighborhoods, is their mostpreferred physical activity (Booth et al. 1997;Giles-Corti and Donovan 2002; Fisher et al.2004; Lee and Moudon 2004). A recent exten-sive review concludes that walkable environ-ments, indeed, promote more walking (Saelensand Handy 2008). A less extensive and consis-tent (Frank et al. 2007) but growing body ofresearch also relates walkable environments tohealthier weights (Ewing et al. 2003; Inagamiet al. 2006; Laraia et al. 2007; Frank et al. 2008;Smith et al. 2008). Fundamental questions re-main, however, concerning the relationshipsbetween human health and walkability and therole of the neighborhood built environment ingeneral. Specifically, what aspects of neighbor-hood environments should be measured, withwhat operational definitions, and at what geo-graphic scale? (O’Campo 2003; Forsyth et al.2006; Hanson 2006; Messer 2007).

Neighborhood walkability is often concep-tualized by the three Ds: population density,pedestrian-friendly design, and land use diver-sity (Cervero and Kockelman 1997). Densityprovides a critical mass of people; pedestrian-friendly street design allows convenient andfairly direct routes; and diversity creates multi-ple attractive destinations for pedestrians. Den-sity is often measured by density of population,housing units, or jobs; pedestrian-friendly de-sign is often measured by street intersectiondensity or sidewalk availability. Diversity, alsoreferred to as mixed land use, is operationalizedin a variety of ways (Song and Rodriguez 2005;Brown et al. 2009), with little consensus on thebest measures. Similarly, researchers adopt arange of geographic scales when defining theextent of the neighborhood. Their choices are

often based on data availability and quantitativeconsiderations rather than theoretical motiva-tions (Messer 2007).

The objective of this study is to enhancethis emerging literature by providing empiricalguidance on the issues of neighborhood scalesand measures of built environment by examin-ing the relationship between body mass index(BMI, defined as weight[kg]/height[m]2) andfour types of mixed land use measures obtainedat three geographic scales that define neigh-borhoods (1 kilometer street-network buffer,census block group, and census tract). Our fo-cus on land use diversity among the three Dsis based on its multifarious operationalizationsmentioned earlier. We build on prior work byBrown and colleagues (2009), one of few stud-ies that conduct comparisons across differenttypes of mixed use measures. We extend thisearlier work by examining a broader range ofmixed use measures and three levels of geo-graphic scales, as well as by exploring the util-ity of integrating multiple scales into a singlemodel. BMI data for this analysis include 4,960licensed drivers in Salt Lake County, Utah.Individual-level BMIs and neighborhood walk-ability measures are related via generalized es-timating equations described in a later section.

Definitions of Neighborhood Scale

Choosing a geographic unit of analysis is along-standing challenge in any spatial researchbecause spatial data and analytical results de-pend on the data aggregation unit or scale,an issue known as the modifiable areal unitproblem (MAUP; Openshaw 1984). Health re-search is no exception. Although research aboutneighborhood effects on health is proliferat-ing, the appropriate geographic scale for mea-suring neighborhoods is still an open question(Hanson 2006; Gauvin et al. 2007; Messer2007; Weiss et al. 2007; Brownson et al. 2009).

When walking is the health behavior of in-terest, the neighborhood should reflect the dis-tance that people can walk to and from home.Two approaches are often used to define neigh-borhoods when measuring walkability. Thefirst and more common is to rely on prede-fined administrative or census boundaries, withcensus tracts and block groups being frequentchoices in the United States (Krieger et al.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14

Mixed Land Use and Obesity 159

2003; King et al. 2005; Frank et al. 2006; In-agami et al. 2006; Rundle et al. 2007; Smithet al. 2008; Zick et al. 2009). Census bound-aries, though, might not necessarily reflectresidents’ walking range within their neigh-borhoods. In addition, walkability measuresconstructed for these boundaries might maskconsiderable heterogeneity within each unit.

The second approach is to create a bufferwith a specific distance around individuals’home locations. A buffer can be a circle basedon the straight-line distance or a polygon cre-ated along a given street network based onthe shortest-path distance (Frank et al. 2005;Cohen et al. 2006; Norman et al. 2006; Berkeet al. 2007; Guo and Bhat 2007; Moudon et al.2007; Oliver, Schuurman, and Hall 2007). Thestreet-network buffer is conceptually more ap-pealing than the straight-line buffer because theformer reflects walking routes imposed by ex-isting streets, although one study found bothtypes of buffers performed similarly in predict-ing walking (Moudon et al. 2007).

Buffer approaches have the advantage of de-lineating more individualized neighborhoods,although geographic information system (GIS)computations can be prohibitive for large datasets. A challenge when using buffers is thechoice of the buffer distance—another MAUPconcern. The buffer distances used in previousstudies vary from 0.1 km (Berke et al. 2007)to 1.6 km (Norman et al. 2006; Forsyth et al.2008). Several studies cite empirical researchto help guide them in defining the limits of atypical walkable distance, but these also varysubstantially, from 0.8 km (Tilt, Unfried, andRoca 2007) to 1 km (Moudon et al. 2007) to1.5 km (King et al. 2005). Acceptable walk-ing distances have been found to vary by in-dividual factors (e.g., age and health status),environmental factors (e.g., route directnessand topography), and destination types andattractiveness (e.g., grocery stores vs. transitstations; Moudon et al. 2006; Canepa 2007).Morency et al. (2009) also demonstrated thatindividuals’ mobility can vary considerably de-pending on their demographic characteristicsand locational settings. These findings illustratepotential complications in the determination ofappropriate buffer distances. They might alsoimply the need for differential buffer distancesacross individuals or locations, but we leave thisissue for future research.

Here we compare alternative measures ofneighborhood walkability constructed for threegeographic scales—census tract, block group,and 1-km street-network buffer—in relationto their association with individual-level BMI.The 1-km buffer is chosen because of its provenusefulness in past research and its compatibilitywith the mixed use measures to be employed.We also explore the possibility that a com-bination of predictors at different geographicscales might provide a superior ability to predictBMI than predictors at one level of scale, giventhat appropriate neighborhood definitions canvary across different variables (Galster 2001;O’Campo 2003).

Measures of Mixed Land Use

The literature examining mixed land use andwalkability presents somewhat conflicting find-ings. One comprehensive review by Saelens andHandy (2008) confirmed that mixed land usesupports physical activity by providing a rangeof destinations within walking distance, such astransit stations and grocery stores. This reviewalso confirmed that measures of mixed use anddistances from home to destinations provideoverlapping alternatives for capturing land usecharacteristics that invite neighborhood walk-ing, especially walking for transportation pur-poses (Saelens and Handy 2008). A recent studythat examined forty-four alternative walkabil-ity measures, however, found that only themeasure of “social land use” (e.g., churches,parks) predicts more walking for transportation(Forsyth et al. 2008). In light of these conflict-ing results, comparative studies of alternativemeasures are needed. This study investigatesfour general types of mixed use measures: sta-tistical summaries of mixed use, areas of walka-ble land uses, distances to specific destinations,and proxy measures.

Statistical Summary IndexesA common summary statistic of mixed land useis an entropy score that measures the extentto which land use categories are equally dis-tributed in an area (Frank et al. 2005). Thismeasure is often found to be positively associ-ated with more physical activity and healthierweights (Frank, Andresen, and Schmid 2004;Mobley et al. 2006; Rundle et al. 2007; Li

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Table 1 Statistical summary indexes of mixed land use

Frank’s entropy score Shannon’s index Simpson’s index

Formulae −6∑

i =1pi ln (pi )

6∑

i =1bi (bi − 1)/{a (a − 1)}−

6∑

i =1

(bi /a

)ln

(bi /a

)/ ln (N )

= −6∑

i =1pi ln (pi )/ ln (N )

Value range (low to highheterogeneity)

0–1 0–positive infinity 1–1/6 (or more generally,1/K ); lower numbers aremore diverse

Neighborhood A (1/2 eachland uses 1 and 3)

1.00 0.69 0.50

Neighborhood B (1/6 eachland uses 1–6)

1.00 1.79 0.17

Note: Land uses: 1 = single-family residential; 2 = multifamily residential; 3 = retail; 4 = office; 5 = education; 6 =entertainment. a = total square feet of land for all six land uses present in a neighborhood; N = number of six land useswith area > 0; K = number of land use categories (in this specification, K = 6); bi (i = 1, . . ., 6) = area of a specificland use, from 1 to 6 above; pi = bi /a = percentage of area of a specific land use.

et al. 2008). The entropy score varies from0 to 1, where 0 indicates maximally homo-geneous land use and 1 indicates maximallyheterogeneous or mixed use. Adapted origi-nally from Shannon’s information theory index(Shannon and Weaver 1949), the entropy scoreis widely used across disciplines to index theevenness of spread across different categories(Krebs 1989). It has been applied to measuresuch things as biodiversity (Ravera 2001) andland use mix (Kockelman 1997; Forsyth 2005).Based on findings of Brown et al. (2009), thisstudy adopts an entropy score using six landuse categories developed by Frank et al. (2006)shown in Table 1.

Alternative summary indexes might provebetter choices than the entropy score, whichhas several limitations (see extended discussionin Brown et al. 2009). For example, Table 1shows how Neighborhood A, equally dividedacross two land uses, and Neighborhood B,equally divided across six land uses, have thesame maximum entropy score, despite the factthat Neighborhood B is more diverse. We thusconsider two alternative diversity indexes usedin other fields that can mitigate this limita-tion of the entropy scores. Shannon’s index(Shannon and Weaver 1949) indicates greaterdiversity with a variety in uses even if some arerare; Simpson’s index (1949) indicates greaterdiversity with evenness of dominant uses(Nagendra 2002). Unlike the entropy score,these two indexes show higher diversity whengreater numbers of land use categories arepresent, as shown in their computations for

Neighborhood B in Table 1. All three pro-posed statistical summaries are subject to otherlimitations, however. Specifically, they do notdistinguish among qualitative differences, nordo they reflect differences in spatial distribu-tions. For example, a fine-grained distributionof many small stores is equivalent to the samearea of a big box store, although the formermight be more likely to induce residents to walkmore.

Walkable Land AreasSome studies examine neighborhood land ar-eas or proportions that are considered to bewalkable but without integrating them intostatistical summaries (Forsyth et al. 2008;Brown et al. 2009). For example, more publicopen space (Giles-Corti et al. 2005) and socialland uses (Forsyth et al. 2008) relate to walk-ing for leisure and transportation, respectively.Brown et al. (2009) demonstrated that the sixland use categories from the preceding Neigh-borhood B example relate to BMI better thanthe corresponding entropy score. In particu-lar, they found that females have lower BMIswhen living in a neighborhood with more en-tertainment and office space, and males havelower BMIs when living in a neighborhood withmore multifamily residential space (Brown et al.2009). Such area measures reflect the extent ofpotentially walkable land uses but, like entropymeasures, they cannot assess whether walkablelands are accessible, well distributed, or attrac-tive to pedestrians.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Destination-Oriented MeasuresThese measures assess the presence, density,or proximity of walkable destinations withina neighborhood. For example, living close togrocery stores, restaurants, and other retailstores relates to more neighborhood walk-ing (Moudon et al. 2007), and living close toemployment establishments is associated withlower weight (Lopez 2007). Although easy toconceptualize and compute, the number of des-tinations studied varies widely, from a singledestination such as parks (Cohen et al. 2007) toover twenty destinations (Moudon et al. 2007;Forsyth et al. 2008). Destinations also vary inspecificity (e.g., retail vs. drug store), and thegeographic clustering of destinations might ormight not be considered in some of the mea-sures used. All of these qualities make it difficultto compare studies and isolate any consistentresults.

In this study, we examine whether weight re-lates to proximity to light rail stations, grocerystores, and the central business district (CBD).Light rail stations are chosen because tran-sit use (Wener and Evans 2007) and residen-tial proximity to transit stations (McCormack,Giles-Corti, and Bulsara 2008) are associatedwith more walking and lower BMI (Rundleet al. 2007; Brown and Werner 2009). Anotheranalysis (Brown et al. 2009) also found thata resident’s BMI is inversely associated withproximity to light rail stations but not to busstops or parks.

We examine distance to the CBD becauseproximity to light rail stations in Salt LakeCounty relates to proximity to the CBD (seeFigure 1). In addition, CBD residents gen-erally walk more (Ewing and Cervero 2001;Chen and McKnight 2007), because CBDs of-fer more conducive walking environments, res-ident preferences, and difficulties with trafficcongestion and parking. Adding distance to theCBD to the analysis allows us to determinewhether light rail and CBD proximity have dis-tinct effects.

Grocery stores can support walking andhealthy eating. One study found that parksand grocery stores are the most frequentwalking destinations among fifteen surveyedoptions (Tilt, Unfried, and Roca 2007). Prox-imity to grocery stores is associated withmore walking (Moudon et al. 2007) andhealthier BMI (Inagami et al. 2006), al-

though other studies found no relation-ship (Forsyth et al. 2008; McCormack,Giles-Corti, and Bulsara 2008). These in-consistencies could be due to grocery storevariations in price, food selection, and qual-ity. We focus on large grocery stores thattypically offer lower prices (Kaufman et al.1997; Kaufman 1999) and healthier foods thansmaller ones (Sallis, Nader, and Atkins 1986;Jetter and Cassady 2006).

Proxy ScoresWe consider two census proxy measures ofmixed land use: housing age (i.e., median yearwhen housing structures were built) and pro-portion of residents who walk to work (U.S.Census Bureau 2000). Neighborhoods witholder housing often have more mixed usesas well as a variety of other walkability fea-tures, including well-connected streets, side-walks, pedestrian-oriented buildings (Handy1996a, 1996b), trees, and narrower streets(Southworth and Owens 1993). Similarly, theproportion of residents who walk to work is hy-pothesized to indicate, at a minimum, the coex-istence of residential and employment land useswithin walking distance. Few residents walk towork (about 2–3 percent on average) in SaltLake County or nationally, so it is unlikely thatan analysis of countywide BMIs would be sub-stantially affected by those who walk to work(and presumably have lower BMIs as a result).Neighborhoods with more residents who walkto work will generally have a wider range andnumber of walkable destinations and more ac-cessible pedestrian pathways (Craig et al. 2002).Both proxies consistently relate to lower BMIsin prior studies (Smith et al. 2008; Brown et al.2009; Zick et al. 2009), although these studiesuse only one geographic scale for each proxy,unlike this study.

Although comparisons across types of mixeduse measures are rare, Brown and colleagues(2009) compared entropy scores, land ar-eas, destination-oriented measures, and cen-sus proxy variables for relationships to weightoutcomes. They found that the entropy scorewith six land use categories adopted from Franket al. (2006) improves prediction of BMI, butthe six categories entered separately improveprediction even more. Proximity to light railstations and the proxy variables also relate to

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Figure 1 Study area—Salt Lake County, Utah. (Color figure available online.)

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


lower BMI. This study builds on this work byexamining predictors across the three levels ofgeographic scale and exploring the utility ofcombining multiple geographic scales in onemodel. In addition, mixed use measures are ex-panded to include Shannon’s and Simpson’sstatistical indexes, as well as distances to gro-cery stores and the CBD.

Data and Methods

Sampling and BMI DataIndividual-level BMI information was derivedfrom a driver license database that contains all453,927 license holders in Salt Lake County,Utah, in 2005. This study used a randomsubset of 4,960 individuals. Twenty individu-als were randomly sampled from 248 censusblock groups that were also randomly sampledfrom 549 census block groups in the county(Figure 1); note that eighteen relatively un-populated (with <150 driver license records)or sparsely populated fringe block groups havebeen excluded. Young adults (<25 years old)who might not have established their own resi-dence and elderly adults (≥65 years old) whoseBMI has more complex associations with healthwere excluded. Given our interests in problemsof higher BMIs compared to healthy BMIs, wealso excluded the small fraction of individu-als who are underweight (BMI <18.5) becausethey might have underlying health conditionsthat affect their BMI and its relationship tothe built environment differently in compari-son with obese individuals. Preliminary anal-yses confirmed that the sample did not differsignificantly from the countywide databasewith respect to gender-specific ages andBMIs (analyses available from the authorson request). BMIs were calculated from self-reported weight and height taken from Utahdriver licenses. The driver license data wereobtained from the Utah Population Database(UPDB), a health-related research databasethat includes records from the Driver LicenseDivision of the Utah Department of PublicSafety. To protect confidentiality of driver li-cense holders, all personal information fromthe Driver License Division was removed be-fore the data were provided to the investiga-tors on this research project. This project hasbeen approved by the University of Utah Insti-

tutional Review Board and the Utah Resourcefor Genetic and Epidemiologic Research. Aspart of this process, the UPDB staff retainedidentifying address information, linked driverlicense data (height, weight, gender, and age) tocensus-block groups via Universal TransverseMercator coordinates and then provided theresearchers with a data set without individualaddresses.

Measures of WalkabilityThe DIGIT Lab at the University of Utahprovided the street centerline data and parcel-level land use data obtained from the SaltLake County Assessor’s Office. The UtahTransportation Authority provided data on thecounty’s light rail transit system. Dun and Brad-street business data were used to identify largegrocery stores (annual sales volume ≥ $1 mil-lion). Finally, the sociodemographic measureswere obtained from the 2000 U.S. Census (U.S.Census Bureau 2000).

The census block group, tract, and 1-kmbuffer were used to compose all measures ofmixed land use, except for the destination-oriented distances. The street-network bufferswere created around individuals’ residences us-ing the street centerline data. For census-basedmeasures (i.e., population density, median yearstructure built, and percentage of residents whowalk to work), a value for each buffer wascomputed as a weighted average of the valuesfor block groups that overlap with the buffer,where the weight for a block group is pro-portional to its overlapping area. For mixeduse measures based on land use categories, theparcel-level land use data were intersected withboundaries of the three geographic units toidentify the areas of the six land uses withineach unit. The street intersection density mea-sure was based on the street centerline dataand excludes intersections involving interstatehighways and intersections of fewer than threestreets.

Figure 1 shows three destination-orientedmeasures of mixed use: distance to the closestlight rail station (abbreviated as LR), distanceto the CBD, and distance to the closestlarge grocery store. All distances measure theshortest path along street centerlines. Thedistance to the CBD is measured as the shortestdistance from a residence to any intersection

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


within and on the boundary of the CBD asdefined by Wood (2005).

Statistical MethodsWe use generalized estimating equations(GEE) to examine the association between in-dividuals’ BMIs and walkability features in theirneighborhoods. GEE is an extension of gener-alized linear models with the quasi-likelihoodapproach and is designed to analyze longitudi-nal and other correlated data (Liang and Zeger1986; Zeger and Liang 1986; Hanley et al.2003). BMI values for individuals selected fromthe same block group are correlated with oneanother according to estimates from a prelimi-nary analysis (p < .001). Consequently, our dataviolate the assumption of uncorrelated errorterms when using the ordinary least squares re-gression, necessitating the use of GEE models.

The literature indicates that neighborhoodsociodemographic status relates to residents’health, including obesity, independent of theirown sociodemographic status (Krieger andGordon 1999; Yen and Syme 1999; Robertand Reither 2004; Janssen et al. 2006; Smithet al. 2008; Zick et al. 2009). Some stud-ies also suggested that census-based aggre-gate measures can be used as valid proxies forindividual-level sociodemographic status, es-pecially at smaller geographic scales (Krieger1991, 1992; Soobader et al. 2001). We thus con-trolled for individual age and six neighborhoodsociodemographic variables (neighborhood in-come, median age of neighborhood residents,and proportions of African American, Hawai-ian/Pacific Islander, Hispanic, and Asian). Be-cause the potential impact of neighborhood so-ciodemographic status is not necessarily limitedto the walkable proximity of individuals’ home,the block group scale was chosen to measurethe six neighborhood variables.

We used models that include only the sevencontrol variables as a baseline (called Level 0model) and evaluated the goodness of fit ofother models relative to the baseline. We firstcreate models for the three geographic scalesseparately and examine the utility of each. Thenwe combined measures of mixed use at eachscale to test whether the overall model fit isimproved by selecting appropriate neighbor-hood scales differently for each of the mixed usemeasures.

Results

Table 2 presents descriptive statistics for all ofthe variables. The street intersection densityfor the street-network buffer scale is over 15percent higher than for the block group andcensus tract scales because the network buffersare more compact and concentrated along thestreets. In terms of mixed land use measures,the block group and buffer scales tend to bemore similar than the census tract scale. Thislikely reflects the size difference of the threegeographic scales; the average size of censustracts, block groups, and 1-km buffers are 3.34km2 (SD = 2.88), 1.21 km2 (SD = 1.73), and1.28 km2 (SD = 0.39), respectively. Such largedifferences across geographic scales underscorethe possibility that choices of scale could be im-portant for BMI prediction. Table 3 shows par-tial correlations between walkability measuresand BMI, controlling for the seven sociodemo-graphic variables. Most correlations are mod-est, but the distance to LR, the distance tothe CBD, and the two census proxies all havesignificant associations with BMI in the ex-pected directions, consistently across genderand geographic scales. Evidence for variabil-ity is present as well, with variables significantfor twelve of thirty-nine gender comparisons invarying degrees; additionally, seven of thirteenvariables are significant at one geographic scalebut not at other scales.

Table 4 summarizes goodness-of-fit statis-tics for models with various combinationsof built environment variables at the threegeographic scales. Model fit is measured bycorrected quasi-likelihood under the inde-pendence model criterion (QICC), whichpenalizes goodness-of-fit measures for modelcomplexity. Lower values of QICC indicatebetter fitting models.

Population Density and Pedestrian-FriendlyDesignLevel 1 models add population density andstreet intersection density to the baseline Level0 models, resulting in large improvements inQICC at every geographic scale. Increasedpopulation density is significantly associatedwith lower BMI, except among females at theblock group scale. Higher intersection den-sity is unexpectedly associated with higher

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Table 2 Descriptive statistics by gender

Malea Femaleb

M SD M SD

Weight statusBMI, individual 26.51 4.52 24.72 4.94

Sociodemographic measuresAge, individual 40.85 10.99 41.74 11.20Median age (both sexes), BG 30.08 5.29 30.02 5.24Proportion African American (%), BG 0.01 0.02 0.01 0.02Proportion Hawaiian and Pacific Islander (%), BG 0.01 0.03 0.01 0.02Proportion Hispanic (%), BG 0.12 0.13 0.10 0.11Proportion Asian (%), BG 0.02 0.03 0.02 0.03Median family income (in $1,000), BG 55.07 19.63 57.05 18.99

Density measurePopulation density (per sq. kilometer), BG 2,167.26 1,189.19 2,165.40 1,165.17Population density (per sq. kilometer), CT 2,064.36 1,015.62 2,080.88 991.62Population density (per sq. kilometer), buffer 2,084.30 801.92 2,093.80 758.75

Design measureIntersection density (per sq. kilometer), BG 42.14 16.91 43.28 16.90Intersection density (per sq. kilometer), CT 40.40 13.68 41.32 13.35Intersection density (per sq. kilometer), buffer 49.00 12.02 49.87 11.78

Area-based measures of mixed useBlock group scale

Area of single family residential (in sq. kilometers), BG 0.398 0.390 0.412 0.382Area of multifamily residential (in sq. kilometers), BG 0.065 0.088 0.062 0.094Area of retail (in sq. kilometers), BG 0.071 0.133 0.062 0.120Area of office (in sq. kilometers), BG 0.065 0.170 0.060 0.182Area of education (in sq. kilometers), BG 0.019 0.120 0.024 0.153Area of entertainment (in sq. kilometers), BG 0.009 0.059 0.007 0.051Total area (in sq. kilometers), BG 0.628 0.584 0.628 0.603Frank’s entropy score, BG 0.508 0.236 0.493 0.232Shannon’s index, BG 0.760 0.377 0.732 0.363Simpson’s index, BG 0.598 0.211 0.614 0.205

Census tract scaleArea of single family residential (in sq. kilometers), CT 1.170 0.757 1.230 0.770Area of multifamily residential (in sq. kilometers), CT 0.216 0.224 0.206 0.224Area of retail (in sq. kilometers), CT 0.198 0.236 0.179 0.217Area of office (in sq. kilometers), CT 0.176 0.322 0.160 0.312Area of education (in sq. kilometers), CT 0.047 0.142 0.053 0.180Area of entertainment (in sq. kilometers), CT 0.024 0.101 0.020 0.090Total area (in sq. kilometers), CT 1.831 1.095 1.847 1.099Frank’s entropy score, CT 0.551 0.201 0.532 0.193Shannon’s index, CT 0.898 0.345 0.865 0.330Simpson’s index, CT 0.543 0.192 0.562 0.185

1 km street-network buffer scaleArea of single family residential (in sq. kilometers), buffer 0.567 0.274 0.593 0.268Area of multifamily residential (in sq. kilometers), buffer 0.101 0.099 0.099 0.103Area of retail (in sq. kilometers), buffer 0.067 0.077 0.060 0.066Area of office (in sq. kilometers), buffer 0.051 0.067 0.047 0.065Area of education (in sq. kilometers), buffer 0.019 0.027 0.019 0.031Area of entertainment (in sq. kilometers), buffer 0.004 0.017 0.004 0.019Total area (in sq. kilometers), buffer 0.808 0.262 0.822 0.262Frank’s entropy score, buffer 0.499 0.224 0.479 0.219Shannon’s index, buffer 0.786 0.371 0.749 0.361Simpson’s index, buffer 0.600 0.208 0.620 0.203

Destination-oriented measures of mixed useDistance to the closest light rail station (in kilometers) 5.54 3.65 5.80 3.72Distance to CBD (in kilometers) 14.02 8.19 14.78 8.16Distance to the closest large grocery store (in kilometers) 1.65 0.81 1.66 0.82

Census proxy measuresBlock group scale

Proportion walk to work (%), BG 2.55 4.30 2.24 3.99Median year structure built, BG 1,969.9 15.5 1,970.2 15.8

Census tract scaleProportion walk to work (%), CT 2.62 3.89 2.39 3.58Median year structure built, CT 1,970.5 15.3 1,970.5 15.7

(Continued on next page)

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Table 2 Descriptive statistics by gender (Continued)

Malea Femaleb

M SD M SD

1-km street-network buffer scaleProportion walk to work (%), buffer 2.51 3.58 2.28 3.33Median year structure built, buffer 1,970.0 14.2 1,970.3 14.7

Note: BMI = body mass index; BG = block group; CT = census tract; CBD = central business district.an = 2,617.bn = 2,343.

male BMI. Subsequent models also showpositive associations (for both genders but moreconsistently for males), suggesting that inter-section density is a poor walkability measurein this sample when using BMI as the healthoutcome. All subsequent models add one ormore measures of mixed land use to the Level1 model.

Land Use DiversityAmong the three summary indexes of mixeduse that are based on areas of the six land usecategories listed in Table 1, Simpson’s index ispreferable because it yields superior QICCs inthree of six gender–scale combinations and itnever has the worst QICC (see Levels 2a–2c inTable 4). Increased diversity measured by any

Table 3 Partial correlation between individuals’ body mass index and built environment measures

MaleBMI

FemaleBMI

Distance to the closest light rail station (inkilometers)

0.054∗∗

0.112∗∗

Distance to CBD (in kilometers) 0.082∗∗ 0.113∗∗Distance to the closest large grocery store (in

kilometers)–0.008 0.025

BG CT Buffer

MaleBMI

FemaleBMI

MaleBMI

FemaleBMI

MaleBMI

FemaleBMI

Population density (per sq. kilometer) –0.024 –0.012 –0.040∗ –0.037 –0.037 –0.040Intersection density (per sq. kilometer) 0.014 0.015 0.000 –0.008 0.039∗ –0.003Frank’s entropy score (higher scores, more

diversity)–0.036 –0.032 –0.043∗ –0.010 –0.043∗ –0.032

Shannon’s index (higher scores, more diversity) –0.044∗ –0.028 –0.040∗ –0.006 –0.047∗ –0.044∗Simpson’s index (lower scores, more diversity) 0.042∗ 0.030 0.043∗ 0.016 0.046∗ 0.046∗Area of single-family residential (in sq.

kilometers)0.048∗ 0.012 0.047∗ 0.050∗ 0.022 0.008

Area of multifamily residential (in sq. kilometers) –0.014 0.030 –0.008 0.050∗ –0.069∗∗ –0.021Area of retail (in sq. kilometers) –0.007 –0.007 0.015 0.021 –0.020 –0.049∗Area of office (in sq. kilometers) –0.007 –0.016 0.015 0.020 –0.049∗ –0.071∗∗Area of education (in sq. kilometers) –0.008 –0.029 –0.009 –0.024 –0.021 0.028Area of entertainment (in sq. kilometers) –0.027 –0.040 0.008 –0.011 0.006 –0.028Proportion walk to work (%) –0.046∗ –0.076∗∗ –0.050∗ –0.103∗∗ –0.065∗∗ –0.103∗∗Median year structure built (larger values, newer

housing)0.081∗∗ 0.111∗∗ 0.076∗∗ 0.110∗∗ 0.079∗∗ 0.119∗∗

Note: Correlation coefficients controlled for individual age, neighborhood income, median age of neighborhood resi-dents, and race/ethnic composition of neighborhoods measured at the block group scale. BMI = body mass index; BG= block group; CT = census tract; CBD = central business district.∗Significant at 5%.∗∗Significant at 1%.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14

Tab

le4

Goo

dnes

s-of

-fit

stat

istic

sfo

rm

odel

sas

soci

atin

gin

divi

dual

body

mas

sin

dex

with

vario

usco

mbi

natio

nsof

built

envi

ronm

ent

mea

sure

sby

corr

ecte

dqu

asi-l

ikel

ihoo

dun

der

the

inde

pend

ence

crite

rion

BG

scale

CT

scale

Bu

ffer

scale

Le

ve

lM

ale

Fe

male

Male

Fe

male

Male

Fe

male

0O

nly

7co

ntro

ls,w

ithou

tan

y3D

varia

bles

QIC

C51

,456

.552

,088

.71

Leve

l0+

popu

latio

nde

nsity

and

inte

rsec

tion

dens

ityQ

ICC

51,3

5452

,010

.251

,221

.551

,964

.751

,101

.951

,983

.2Im

prov

emen

tfr

omLe

vel0

–102

.5–7

8.5

–235

.0–1

24.0

–354

.6–1

05.5

2Le

vel1

+ar

ea-b

ased

dive

rsity

mea

sure

s2a

Fran

k’s

inde

xQ

ICC

51,2

93.1

52,0

0251

,149

.751

,962

.651

,062

.351

,930

Impr

ovem

ent

from

Leve

l0–1

63.4

–86.7

–306

.8–1

26.1

–394

.2–1

58.7

2bS

hann

on’s

inde

xQ

ICC

51,2

54.8

52,0

13.1

51,1

56.1

51,9

6551

,048

.951

,895

.8Im

prov

emen

tfr

omLe

vel0

–201

.7–7

5.6

–300

.4–1

23.7

–407

.6–1

92.9

2cS

imps

on’s

inde

xQ

ICC

51,2

6652

,006

.951

,140

.751

,952

.751

,051

.551

,879

.3Im

prov

emen

tfr

omLe

vel0

–190

.5–8

1.8

–315

.8–1

36.0

–405

.0–2

09.4

2dFr

ank’

s6

cate

gorie

sQ

ICC

51,2

13.2

51,8

94.4

51,1

29.4

51,6

97.6

50,9

34.8

51,6

08.1

6Im

prov

emen

tfr

omLe

vel0

–243

.3–1

94.3

–327

.1–3

91.1

–521

.7–4

80.5

3Le

vel1

+de

stin

atio

n-ba

sed

dive

rsity

mea

sure

s3a

Dis

tanc

eto

LRst

atio

nQ

ICC

51,2

30.8

51,4

12.4

51,1

01.1

51,3

38.6

51,0

06.2

51,3

77.6

Impr

ovem

ent

from

Leve

l0–2

25.7

–676

.3–3

55.4

–750

.1–4

50.3

–711

.13b

Dis

tanc

esto

LR&

CB

DQ

ICC

51,0

62.2

51,1

84.4

50,9

82.3

51,1

71.9

50,8

82.2

51,1

91.1

Impr

ovem

ent

from

Leve

l0–3

94.3

–904

.3–4

74.2

–916

.8–5

74.3

–897

.63c

Dis

tanc

eto

larg

egr

ocer

QIC

C51

,351

.852

,008

.251

,210

.351

,952

.951

,084

.751

,977

.9Im

prov

emen

tfr

omLe

vel0

–104

.7–8

0.5

–246

.2–1

35.8

–371

.8–1

10.8

3dD

ista

nces

toLR

,CB

D,&

larg

egr

ocer

QIC

C51

,048

.351

,172

.250

,959

.451

,167

.950

,856

.951

,186

.6Im

prov

emen

tfr

omLe

vel0

–408

.2–9

16.5

–497

.1–9

20.8

–599

.6–9

02.1

Leve

l4Le

vel1

+pr

oxy-

base

ddi

vers

itym

easu

res

4aPe

rcen

tage

wal

kto

wor

kQ

ICC

51,2

96.4

51,7

87.6

51,1

67.6

51,5

24.4

50,9

83.1

51,5

03.8

Impr

ovem

ent

from

Leve

l0–1

60.1

–301

.1–2

88.9

–564

.3–4

73.4

–584

.94b

Med

ian

year

stru

ctur

ebu

iltQ

ICC

51,0

20.0

51,3

82.4

50,9

47.9

51,3

81.1

50,8

44.3

51,3

27.9

Impr

ovem

ent

from

Leve

l0–4

36.5

–706

.3–5

08.6

–707

.6–6

12.2

–760

.8Le

vel5

Com

bina

tion

ofLe

vels

3ban

d4b

Leve

l1+

dist

ance

sto

LR&

CB

D+

med

ian

year

stru

ctur

ebu

iltQ

ICC

50,9

60.5

51,0

07.2

50,9

08.4

51,0

30.1

50,8

1951

,028

.1Im

prov

emen

tfr

omLe

vel0

–496

.0–1

,081

.5–5

48.1

–1,0

58.6

–637

.5–1

,060

.6Im

prov

emen

tfr

omLe

vel3

b–1

01.7

–177

.2–7

3.9

–141

.8–6

3.2

–163

Impr

ovem

ent

from

Leve

l4b

–59.5

–375

.2–3

9.5

–351

.0–2

5.3

–299

.8(C

ontin

ued

onne

xtpa

ge)

167

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14

Tab

le4

Goo

dnes

s-of

-fit

stat

istic

sfo

rm

odel

sas

soci

atin

gin

divi

dual

body

mas

sin

dex

with

vario

usco

mbi

natio

nsof

built

envi

ronm

ent

mea

sure

sby

corr

ecte

dqu

asi-l

ikel

ihoo

dun

der

the

inde

pend

ence

crite

rion

(Con

tinue

d)

BG

scale

CT

scale

Bu

ffer

scale

Le

ve

lM

ale

Fe

male

Male

Fe

male

Male

Fe

male

Leve

l6Le

vel5

+si

xla

ndus

es,%

wal

kto

wor

k,or

both

Leve

l1+

dist

ance

sto

LR&

CB

D+

med

ian

year

stru

ctur

ebu

ilt+

mor

e6a

Leve

l5+

six

land

uses

QIC

C50

,859

.250

,802

.650

,877

.550

,744

.750

,737

.550

,743

.6Im

prov

emen

tfr

omLe

vel0

–597

.3–1

,286

.1–5

79.0

–1,3

44.0

–719

.0–1

,345

.1Im

prov

emen

tfr

omLe

vel5

–101

.3–2

04.6

–30.

9–2

85.4

–81.

5–2

84.5

6bLe

vel5

+%w

alk

tow

ork

QIC

C50

,958

.450

,992

.950

,910

.450

,980

.250

,806

.850

,960

.3Im

prov

emen

tfr

omLe

vel0

–498

.1–1

,095

.8–5

46.1

–1,1

08.5

–649

.7–1

,128

.4Im

prov

emen

tfr

omLe

vel5

–2.1

–14.

32.

0–4

9.9

–12.

2–6

7.8

6cLe

vel5

+si

xla

ndus

es+%

wal

kto

wor

kQ

ICC

50,8

55.9

50,7

95.4

50,8

79.3

50,7

12.4

50,7

38.6

50,6

79.2

Impr

ovem

ent

from

Leve

l0–6

00.6

–1,2

93.3

–577

.2–1

,376

.3–7

17.9

–1,4

09.5

Impr

ovem

ent

from

Leve

l5–1

04.6

–211

.8–2

9.1

–317

.7–8

0.4

–348

.9

Not

e:Va

lues

inita

lics

show

impr

ovem

ent

offit

inco

mpa

rison

with

othe

rm

odel

s,w

here

larg

erne

gativ

eva

lues

are

asso

ciat

edw

ithla

rger

impr

ovem

ent.

Sha

ded

valu

esin

dica

teth

ebe

stfit

mod

elfo

rea

chle

velo

fm

odel

sby

gend

er.Q

ICC

=co

rrec

ted

quas

i-lik

elih

ood

unde

rth

ein

depe

nden

cem

odel

crite

rion;

LR=

light

rail;

CB

D=

cent

ralb

usin

ess

dist

rict.

168

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


summary index is always associated with lowerBMI, although the association is not necessarilysignificant, but in all six gender–scale combina-tions summary indexes do not perform as wellas the six land use categories added separately(Level 2d, QICC improvements of 11 to 271points). This finding, which is consistent withthose reported by Brown et al. (2009), mightbe an indication that what matters to neigh-borhood walkability is the presence of walkableland uses rather than their level of mixture.

Destination-oriented measures of mixed useyield varied results (Levels 3a–3d). The dis-tance to LR considerably improves model fitfor all gender–scale combinations, especiallyamong females (Level 3a). The LR variable isalso significant, with increased distance associ-ated with higher BMI. Adding the distance tothe CBD to Level 3a models further improvesmodel fit, and the variable is significant for allgender–scale combinations. The significance ofthe distance to LR disappears for male models,however, perhaps reflecting moderate but sig-nificant correlations between the two variables(0.486 for males and 0.475 for females). Theresult implies that the proximity to the CBDoffers some benefit to both males and femalesin terms of their weight status, but the prox-imity to LR is beneficial only for females. Thedistance to the closest large grocery store, onthe other hand, leads to much smaller modelimprovements (Table 4, Levels 3c and 3d).

Both census proxy variables (Levels 4a and4b) yield large QICC improvements, with ahigher percentage of residents who walk towork and older neighborhood housing (i.e.,earlier median year built) both significantlyassociated with lower BMI. Older housing isespecially powerful and consistent across allgender–scale combinations.

Comparisons of models of Levels 2 through4 reveal several interesting patterns. First,the six land use categories included sepa-rately (Level 2d) perform best among the fourarea-based measures of mixed use, but theirQICC improvements are much smaller thanthose achieved by the combination of distancesto LR and the CBD (Level 3b) and the medianyear built variable (Level 4b). The walk to workvariable also outperforms the six categories forsome gender–scale combinations. Second, thecombination of distances to LR and the CBDconsistently performs best for females, whereas

the median year built variable consistently per-forms best for males. Third, models at the 1 kmbuffer scale generally achieve the best QICC,except among females, where distances to LRand the CBD perform best when other variables(i.e., Level 0 population and intersection den-sities) are aggregated to the census tract scale.

Combination of Multiple Mixed UseMeasuresUsing multiple area-based measures in a singlemodel makes little sense because of their sim-ilarity in the conceptualization of mixed use.Other measures, however, can be combined toachieve a better understanding of neighbor-hood characteristics that relate to individualBMI. Perhaps measures such as housing ageand proximities to LR and the CBD indicatenot only mixed use but also other walkable fea-tures of neighborhoods, such as green spaceand sidewalks. Therefore, the following analy-ses examine whether models can be improvedby combining multiple mixed use measures.

Level 5 models combine the best singlemixed use measures from above (Level 3b dis-tances to LR and the CBD and Level 4b hous-ing age) and yield QICC improvements from25 to 375 points compared to the originaltwo models. As in the original models, themedian year built variable is significant in allgender–scale combinations, and the distance toLR is significant only for females. The signif-icance of the distance to the CBD disappears,which likely reflects its relatively high correla-tion with the median year built variable (partialrs are about 0.6–0.7). Interestingly, the blockgroup scale provides the best model fit for fe-males, which rarely happened in earlier models.

Level 6 models add the six land use cate-gories, the walk to work variable, or both toLevel 5 models. Noticeable improvements inQICC are attained by the six land use cate-gories (Level 6a) in most gender–scale combi-nations, whereas improvement by the walk towork variable is marginal except for females atthe census tract and buffer scales (Level 6b).Adding both variables simultaneously (Level6c) barely improves the model fit in compar-ison with Level 6a models, with an exceptionof the female model at the buffer scale. Thewalk to work variable is insignificant in mostcases, although several of the six land uses are

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Table 5 Best estimated models relating individual body mass index and built environment measures

Independent variables Beta SE p value

MaleIntercept –30.190 16.510 0.067Age, individual 0.072 0.008 0.000∗∗Median age (both sexes), BG 0.022 0.019 0.247Proportion African American (%), BG 7.229 5.711 0.206Proportion Hawaiian and Pacific Islander (%), BG 3.546 3.168 0.263Proportion Hispanic (%), BG 0.963 0.927 0.299Proportion Asian (%), BG –6.407 3.177 0.044∗Median family income (in $1,000), BG –0.020 0.006 0.001∗∗Population density (per sq. kilometer), CT 0.000 0.000 0.113Intersection density (per sq. kilometer), buffer 0.025 0.009 0.006∗∗Distance to the closest light rail station (in kilometers) 0.003 0.033 0.920Distance to CBD (in kilometers) 0.003 0.020 0.875Median year structure built, BG 0.027 0.008 0.001∗∗Area of single-family residential (in sq. kilometers), buffer 0.253 0.491 0.606Area of multifamily residential (in sq. kilometers), buffer –2.319 1.051 0.027∗Area of retail (in sq. kilometers), buffer 0.839 1.259 0.505Area of office (in sq. kilometers), buffer –1.218 1.522 0.424Area of education (in sq. kilometers), buffer –1.697 3.170 0.592Area of entertainment (in sq. kilometers), buffer 3.455 7.933 0.663

FemaleIntercept –33.285 31.768 0.295Age, individual 0.099 0.009 0.000∗∗Median age (both sexes), BG 0.037 0.020 0.065Proportion African American (%), BG 8.686 7.606 0.253Proportion Hawaiian and Pacific Islander (%), BG 11.168 3.451 0.001∗∗Proportion Hispanic (%), BG 3.515 1.583 0.026∗Proportion Asian (%), BG –9.398 3.770 0.013∗Median family income (in $1,000), BG –0.051 0.009 0.000∗∗Population density (per sq. kilometer), buffer 0.000 0.000 0.434Intersection density (per sq. kilometer), BG 0.005 0.007 0.456Distance to the closest light rail station (in kilometers) 0.103 0.039 0.009∗∗Distance to CBD (in kilometers) 0.014 0.034 0.681Median year structure built, buffer 0.028 0.016 0.085Proportion walk to work (%), buffer –0.086 0.059 0.142Area of single-family residential (in sq. kilometers), buffer –0.266 0.549 0.627Area of multifamily residential (in sq. kilometers), buffer 2.361 1.631 0.148Area of retail (in sq. kilometers), buffer –1.384 2.175 0.525Area of office (in sq. kilometers), buffer –1.245 2.186 0.569Area of education (in sq. kilometers), buffer 9.256 5.691 0.104Area of entertainment (in sq. kilometers), buffer –7.191 3.114 0.021∗

Note: BG = block group; CT = census tract; CBD = central business district.∗Significant at 5%.∗∗Significant at 1%.

significant in both Level 6a and 6c models forsome gender–scale combinations. Overall, thebest model fit is achieved by Level 6a modelfor males (i.e., distances to LR and the CBD+ housing age + six land use areas) and Level6c model for females (i.e., male model above+ proportion walk to work), both at the bufferscale.

Finally, to examine the potential benefit ofchoosing geographic scales for individual builtenvironment measures rather than choosingone for the entire model, we modify the Level6a model for males and the Level 6c model

for females to include measures at geographicscales that give the strongest partial correla-tions in Table 3. For the six land use categories,the buffer scale was chosen because it led tothe largest QICC improvements in Level 2d,6a, and 6c models. The resulting QICC val-ues improve on the previously best models inTable 4 for both genders. Table 5 presents de-tails of these best fit models. For males, lowerBMIs are associated with lower intersectiondensity at the census tract level (p = 0.006),older housing at the block group level (p =.001), and more multifamily residential use at

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


the buffer level (p = 0.027). Lower BMIs offemales are associated with closer light rail sta-tions (p = 0.009) and more entertainment useat the buffer level (p = 0.021). These significantassociations are in the hypothesized directions,except for the intersection density for males.

Summary and Discussions

Despite the growing literature on walkabilityand health, few guidelines exist for researchersto decide on proper geographic scales and walk-ability measures, especially mixed land use mea-sures. In this article, we examine three geo-graphic scales and four types of mixed landuse measures in relationship to BMI. Analy-ses indicate that the use of the street-networkbuffer generally results in relatively better fit-ting models.

The advantage of buffer measures is not nec-essarily consistent across genders or measuresof the built environment, however. Further, thebest fitting models demonstrate that the perfor-mance of models could be improved by choos-ing an appropriate geographic scale for eachmeasure of the built environment instead ofchoosing one scale for all variables. We recog-nize that our results might not generalize to allgeographic circumstances. We chose the bestscale simply by relying on partial correlationsand we limited testing to three levels of geo-graphic scale. This analysis nonetheless illus-trates a potential pitfall when any fixed neigh-borhood definition is used to measure all builtenvironmental features.

Almost all of the alternative measures ofmixed use examined in this study are useful inpredicting individual BMI. The census proxyof the median housing age and the distanceto the closest light rail station are particularlypromising, followed by the six land use cate-gories included separately. None of the threestatistical summary indexes depicted in Table1 outperform the six land use categories in-cluded separately. In addition, some combina-tions of the alternative measures demonstratefurther improvement in model fit. Together,these findings suggest that no single measureof mixed use will likely fully capture how landuse relates to neighborhood BMIs; mixed useis complex and merits complex and multiplemeasures.

The variable measuring proximity to largegrocery stores is notable for its failure toimprove model fit. This might happen becausecommon shopping patterns of Americansrequire motorized transportation to carry mul-tiple, heavy bags. Motorized transportationalso allows them to choose stores not in theproximity of their residence. Additionally,grocery stores provide not only healthy foodbut also less nutritious food, and they might belocated near unhealthy food outlets such as fastfood restaurants. Weight status is determinedby physical activity and eating behavior; thus,food environments that relate to both compo-nents might have more complex associationswith BMI than what could be measured bythe proximity to grocery stores alone. Anotherunexpected, notable finding in this study isthat mixed use measures at the census tractscale generally resulted in better model fit thanthose at the block group scale. This is counter-intuitive, as the smaller block group scale moreclosely approximates walkable distances. Theseunexpected findings might result from the factthat walkability in this study is solely restrictedto residential neighborhood. Although this is awidely adopted approach in studies concernedwith neighborhood effects on human health,individuals travel to many areas for workand shopping, including those outside theirresidential neighborhoods. Consequently,walkability measures that take into con-sideration human spatial behaviors such asspace–time accessibility developed in timegeography (Kwan 1998; Miller 1999) might bejustified to better capture the built environ-mental features that are relevant to individuals’health.

Results also demonstrate gender differencesin the association between individual BMI andbuilt environment measures as well as in thepreferred geographic scales of analysis for sev-eral measures. Consistent with prior research(Rundle et al. 2007; Brown and Werner 2009;Brown et al. 2009), proximity to LR stationsrelates to lower female BMI, regardless of theproximity to the CBD. In contrast, proximity toLR relates to lower male BMI only when prox-imity to the CBD is not controlled. This mightreflect gender differences in transit use (Crane2007; Pucher and Renne 2003) or private ve-hicle ownership. In addition, different subsetsof the six land use categories show significant

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


associations with male and female BMIs. Theseresults suggest the importance of examininghow walkability features are used and perceiveddifferently by men and women.

To the best of our knowledge, self-reporteddriver license weight and height informationhas rarely been used in obesity research. Anadvantage of driver license data is its extensivecoverage of the adult population. The originalUPDB driver license data includes almost 90percent of Salt Lake County residents betweenthe ages of twenty-five and sixty-four, whichenhances the external validity of the study’sfinding.

A possible disadvantage of using driver li-cense data is that they might exclude themost economically disadvantaged individualswho might also have a higher obesity risk.The use of driver license data also imposespotential limitations of self-reported weightand a time lag between the physical environ-ment and weight measures. Past studies havefound a tendency for individuals to underre-port weight and overreport height (Nawaz et al.2001; Gorber et al. 2007). Nevertheless, self-reported weights, such as those in the Behav-ioral Risk Factor Surveillance System and theNational Health and Nutrition ExaminationSurvey, have proved valuable for monitoringobesity trends in the United States (Ezzati et al.2006; Centers for Disease Control and Preven-tion 2007).

Individuals’ records in the data correspondto their most recent renewal; in Utah, re-newals are required every five years or after ad-dress changes, name changes, or loss of license.The data therefore represent the most recentheight and weight data from 1995 through2005, whereas census and other built envi-ronment measures are from different years. Inaddition, for individuals who changed their res-idences during this time period, reported BMImeasures might not correspond to the built en-vironment of their current residences; we arecurrently investigating this issue of residentialchange. Given these potential limitations andthe fact that adults between the ages of twenty-five and sixty-four typically gain weight overtime (U.S. Department of Agriculture 2005),BMI measures in this study are likely to un-derestimate true BMIs and thus our findingsshould be viewed as conservative.

There are additional research questions thatare beyond the scope of this study. First, thebuffer distance of 1 km was chosen primarilyfor compatibility with prior studies; we did nottest for optimal buffer distances. Studies thatexamine multiple buffer distances have founddifferences across distances in terms of the mag-nitude of associations (Berke et al. 2007) andthe associated built environment features (Mc-Cormack, Giles-Corti, and Bulsara 2008). Thisimplies that optimal distances likely vary for dif-ferent measures of built environment. Conse-quently, it is essential to develop theoretical aswell as empirical bases to decide on appropriatebuffer distances. Second, the six land use cate-gories used in this study are from Frank et al.(2006). They could be classified differently inother municipalities or data sets, which mighthamper comparative studies. Third, data werenot available that would allow us to controlfor additional individual-level covariates (e.g.,income) and behaviors (e.g., walking). Somestudies suggest that women are more likely touse public transit than men (Pucher and Renne2003; Crane 2007), which might explain thegender differences we observed in proximity toLR. Future research is needed to fully under-stand the sources of these differences. Fourth,these results are cross-sectional and do not con-sider residential self-selection. Most past stud-ies that relate walkability to physical activity orobesity risk have not explicitly investigate thisissue. If more physically active individuals self-select into more walkable neighborhoods, thiscould introduce a bias into the estimated re-lationships. Thus, care must be taken not tointerpret our estimated relationships as causal.

Three variables used in this study merit spe-cial attention in future research. The censusproxy measures—median year structure builtand the percentage of residents walking towork—are useful measures, given their ease ofcollection and consistent relations with BMI(Smith et al. 2008; Brown et al. 2009; Zicket al. 2009). However, the question of ex-actly how these variables affect walkability andBMI deserves further attention. If our findingsare replicated elsewhere, census proxy variablesmight enable researchers and policymakers toidentify potential hot spots of obesity risk with-out requiring detailed land use information.Similarly, the observed health benefit of the

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


proximity to transit stations merits wider study,including investigations of effects that aredirect (walking to rail stations) and indirect(walking to other destinations clustered aroundrail stations). LR offers societal benefits of lesspollution and oil dependence; if LR consis-tently relates to lower BMI, then the healthbenefits of LR deserve broader considerationfrom policymakers and researchers. �

Literature Cited

Berke, E. M., T. D. Koepsell, A. V. Moudon, R. E.Hoskins, and E. B. Larson. 2007. Association ofthe built environment with physical activity andobesity in older persons. American Journal of PublicHealth 97 (3): 486–92.

Blanck, H. M., W. H. Dietz, D. A. Galuska, C. Gille-spie, R. Hamre, L. K. Khan, M. K. Serdula, E.S. Ford, W. S. Garvin, A. H. Mokdad, and D.Densmore. 2006. State-specific prevalence of obe-sity among adults—United States, 2005. Morbidityand Mortality Weekly Report 55 (36): 985–88.

Booth, M. L., A. Bauman, N. Owen, and C. J.Gore. 1997. Physical activity preferences, pre-ferred sources of assistance, and perceived barri-ers to increased activity among physically inactiveAustralians. Preventive Medicine 26 (1): 131–37.

Brown, B. B., and C. M. Werner. 2009. Before andafter a new light rail stop: Resident attitudes, travelbehavior, and obesity. Journal of the American Plan-ning Association 75 (1): 5–12.

Brown, B. B., I. Yamada, K. R. Smith, C. D. Zick,L. Kowaleski-Jones, and J. X. Fan. 2009. Mixedland use and walkability: Variations in land usemeasures and relationships with BMI, overweight,and obesity. Health & Place 15:1130–41.

Brownson, R. C., C. M. Hoehner, K. Day, A. Forsyth,and J. F. Sallis. 2009. Measuring the built en-vironment for physical activity: State of the sci-ence. American Journal of Preventive Medicine 36 (4,Suppl. 1): S99–S123.e12.

Canepa, B. 2007. Bursting the bubble: Determiningthe transit-oriented development’s walkable limits.Transportation Research Record: Journal of the Trans-portation Research Board 1992:28–34.

Centers for Disease Control and Prevention. 2007.Health risks in the United States: Behavioral Risk Fac-tor Surveillance System at a glance 2007. Atlanta, GA:Centers for Disease Control and Prevention.

Cervero, R., and K. Kockelman. 1997. Travel de-mand and the 3Ds: Density, diversity, and design.Transportation Research Part D-Transport and Envi-ronment 2 (3): 199–219.

Chen, C., and C. E. McKnight. 2007. Does the builtenvironment make a difference? Additional evi-dence from the daily activity and travel behav-

ior of homemakers living in New York City andsuburbs. Journal of Transport Geography 15 (5): 380–95.

Cohen, D. A., S. Ashwood, M. Scott, A. Overton,K. R. Evenson, C. C. Voorhees, A. Bedimo-Rung,and T. L. McKenzie. 2006. Proximity to schooland physical activity among middle school girls:The trial of activity for adolescent girls. Journal ofPhysical Activity and Health 3 (S1): S129–S138.

Cohen, D. A., A. Sehgal, S. Williamson, D. Golinelli,N. Lurie, and T. L. McKenzie. 2007. Contributionof public parks to physical activity. American Jour-nal of Public Health 97 (3): 509–14.

Craig, C. L., R. C. Brownson, S. E. Cragg, and A.L. Dunn. 2002. Exploring the effect of the en-vironment on physical activity—A study examin-ing walking to work. American Journal of PreventiveMedicine 23 (2): 36–43.

Crane, R. 2007. Is there a quiet revolution in women’stravel? Revisiting the gender gap in commuting.Journal of the American Planning Association 73 (3):298–316.

Ewing, R., and R. Cervero. 2001. Travel and the builtenvironment—A synthesis. Transportation ResearchRecord 1780:87–114.

Ewing, R., T. Schmid, R. Killingsworth, A. Zlot, andS. Raudenbush. 2003. Relationship between urbansprawl and physical activity, obesity, and morbid-ity. American Journal of Health Promotion 18 (1):47–57.

Ezzati, M., H. Martin, S. Skjold, S. V. Hoorn, and C.J. L. Murray. 2006. Trends in national and state-level obesity in the USA after correction for self-report bias: Analysis of health surveys. Journal ofthe Royal Society of Medicine 99 (5): 250–57.

Fisher, K. J., F. Z. Li, Y. Michael, and M. Cleveland.2004. Neighborhood-level influences on physicalactivity among older adults: A multilevel analysis.Journal of Aging and Physical Activity 12 (1): 45–63.

Flegal, K. M., M. D. Carroll, C. L. Ogden, and L.R. Curtin. 2010. Prevalence and trends in obesityamong US adults, 1999–2008. Journal of the Amer-ican Medical Association 303 (3): 235–41.

Forsyth, A. E., ed. 2005. Environment and physicalactivity: GIS protocols. Minneapolis: MetropolitanDesign Center, University of Minnesota.

Forsyth, A., M. Hearst, J. M. Oakes, and K. H.Schmitz. 2008. Design and destinations: Factorsinfluencing walking and total physical activity. Ur-ban Studies 45 (9): 1973–96.

Forsyth, A., K. H. Schmitz, M. Oakes, J. Zimmer-man, and J. Koepp. 2006. Standards for environ-mental measurement using GIS: Toward a pro-tocol for protocols. Journal of Physical Activity andHealth 3 (Suppl. 1): S241–S257.

Frank, L. D., M. A. Andresen, and T. L. Schmid.2004. Obesity relationships with community

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


design, physical activity, and time spent in cars.American Journal of Preventive Medicine 27 (2):87–96.

Frank, L. D., J. Kerr, J. F. Sallis, R. Miles, and J.Chapman. 2008. A hierarchy of sociodemographicand environmental correlates of walking and obe-sity. Preventive Medicine 47:172–78.

Frank, L. D., B. E. Saelens, K. E. Powell, andJ. E. Chapman. 2007. Stepping towards causa-tion: Do built environments or neighborhood andtravel preferences explain physical activity, driv-ing, and obesity? Social Science & Medicine 65 (9):1898–1914.

Frank, L. D., J. F. Sallis, T. L. Conway, J. E. Chap-man, B. E. Saelens, and W. Bachman. 2006. Manypathways from land use to health: Associations be-tween neighborhood walkability and active trans-portation, body mass index, and air quality. Journalof the American Planning Association 72 (1): 75–87.

Frank, L., T. Schmid, J. Sallis, J. Chapman, and B.Saelens. 2005. Linking objectively measured phys-ical activity with objectively measured urban form:findings from SMARTRAQ. American Journal ofPreventive Medicine 28 (Suppl. 2): 117–25.

Galster, G. C. 2001. On the nature of neighborhood.Urban Studies 38 (12): 2111–24.

Gauvin, L., E. Robitaille, M. Riva, L. McLaren,C. Dassa, and L. Potvin. 2007. Conceptualizingand operationalizing neighbourhoods—The co-nundrum of identifying territorial units. CanadianJournal of Public Health-Revue Canadienne De SantePublique 98:S18–S26.

Giles-Corti, B., M. H. Broomhall, M. Knuiman, C.Collins, K. Douglas, K. Ng, A. Lange, and R. J.Donovan. 2005. Increasing walking: How impor-tant is distance to, attractiveness, and size of publicopen space? American Journal of Preventive Medicine28 (2, Suppl. 2): 169–76.

Giles-Corti, B., and R. J. Donovan. 2002. The rela-tive influence of individual, social and physical en-vironment determinants of physical activity. SocialScience & Medicine 54 (12): 1793–1812.

Gorber, S. C., M. Tremblay, D. Moher, and B. Gor-ber. 2007. A comparison of direct vs. self-reportmeasures for assessing height, weight and bodymass index: A systematic review. Obesity Reviews8 (4): 307–26.

Guo, J. Y., and C. R. Bhat. 2007. Operationalizingthe concept of neighborhood: Application to resi-dential location choice analysis. Journal of TransportGeography 15 (1): 31–45.

Handy, S. L. 1996a. Understanding the link betweenurban form and nonwork travel behavior. Jour-nal of Planning Education and Research 15 (3): 183–98.

———. 1996b. Urban form and pedestrian choices: Astudy of four Austin neighborhoods. TransportationResearch Record 1552:135–44.

Hanley, J. A., A. Negassa, M. D. d. Edwardes, andJ. E. Forrester. 2003. Statistical analysis of corre-lated data using generalized estimating equations:An orientation. American Journal of Epidemiology157 (4): 364–75.

Hanson, S. 2006. Active living research in light ofthe TRB/IMO report. Journal of Physical Activityand Health 3 (Suppl. 1): S258–S266.

Hedley, A. A., C. L. Ogden, C. L. Johnson, M.D. Carroll, L. R. Curtin, and K. M. Flegal.2004. Prevalence of overweight and obesity amongU.S. children, adolescents, and adults, 1999–2002.Journal of American Medical Association 291 (23):2847–50.

Hill, J. O., and J. C. Peters. 1998. Environmentalcontributions to the obesity epidemic. Science 280(5368): 1371–74.

Inagami, S., D. A. Cohen, B. K. Finch, and S. M.Asch. 2006. You are where you shop—Grocerystore locations, weight, and neighborhoods. Amer-ican Journal of Preventive Medicine 31 (1): 10–17.

Janssen, I., W. F. Boyce, K. Simpson, and W. Pick-ett. 2006. Influence of individual- and area-levelmeasures of socioeconomic status on obesity, un-healthy eating, and physical inactivity in Canadianadolescents. American Journal of Clinical Nutrition83 (1): 139–45.

Jetter, K. M., and D. L. Cassady. 2006. The availabil-ity and cost of healthier food alternatives. AmericanJournal of Preventive Medicine 30 (1): 38–44.

Kaufman, P. R. 1999. Rural poor have less access tosupermarkets, large grocery stores. Rural Develop-ment Perspectives 13 (3): 19–26.

Kaufman, P. R., J. M. MacDonald, S. M. Lutz, andD. M. Smallwood. 1997. Do the poor pay morefor food? Item selection and price differences af-fect low-income household food costs. AgriculturalEconomic Report No. 759. Washington, DC: U.S.Department of Agriculture, Food and Rural Eco-nomics Division, Economic Research Service, U.S.Department of Agriculture.

King, W. C., S. H. Belle, J. S. Brach, L. R. Simkin-Silverman, T. Soska, and A. M. Kriska. 2005.Objective measures of neighborhood environmentand physical activity in older women. AmericanJournal of Preventive Medicine 28 (5): 461–69.

Kockelman, K. 1997. Travel behavior as function ofaccessibility, land use mixing, and land use balance:Evidence from San Francisco Bay area. Transporta-tion Research Record: Journal of the TransportationResearch Board 1607 (1): 116–25.

Krebs, C. J. 1989. Ecological methodology. New York:Harper & Row.

Krieger, N. 1991. Women and social class: A method-ological study comparing individual, household,and census measures as predictors of black/whitedifferences in reproductive history. Journal of Epi-demiology and Community Health 45 (1): 35–42.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


———. 1992. Overcoming the absence of socioeco-nomic data in medical records: Validation and ap-plication of a census-based methodology. AmericanJournal of Public Health 82 (5): 703–10.

Krieger, N., J. T. Chen, P. D. Waterman, D.H. Rehkopf, and S. V. Subramanian. 2003.Race/ethnicity, gender, and monitoring socioe-conomic gradients in health: A comparison ofarea-based socioeconomic measures—The pub-lic health disparities geocoding project. AmericanJournal of Public Health 93 (10): 1655–71.

Krieger, N., and D. Gordon. 1999. Re: Use of census-based aggregate variables to proxy for socioeco-nomic group: Evidence from national samples.American Journal of Epidemiology 150 (8): 892–94.

Kwan, M.-P. 1998. Space-time and integral measuresof individual accessibility: A comparative analysisusing a point-based framework. Geographical Anal-ysis 30 (3): 191–216.

Laraia, B., L. Messer, K. Evenson, and J. S. Kauf-man. 2007. Neighborhood factors associated withphysical activity and adequacy of weight gain dur-ing pregnancy. Journal of Urban Health—Bulletin ofthe New York Academy of Medicine 84 (6): 793–806.

Lee, C., and A. V. Moudon. 2004. Physical activityand environment research in the health field: Im-plications for urban and transportation planningpractice and research. Journal of Planning Litera-ture 19 (2): 147–81.

Li, F. Z., P. A. Harmer, B. J. Cardinal, M. Bosworth,A. Acock, D. Johnson-Shelton, and J. M. Moore.2008. Built environment, adiposity, and physicalactivity in adults aged 50–75. American Journal ofPreventive Medicine 35 (1): 38–46.

Liang, K. Y., and S. L. Zeger. 1986. Longitudi-nal data analysis using generalized linear models.Biometrika 73:13–22.

Lopez, R. P. 2007. Neighborhood risk factors forobesity. Obesity 15 (8): 2111–19.

McCormack, G. R., B. Giles-Corti, and M. Bulsara.2008. The relationship between destination prox-imity, destination mix and physical activity behav-iors. Preventive Medicine 46 (1): 33–40.

Messer, L. C. 2007. Invited commentary: Beyondthe metrics for measuring neighborhood effects.American Journal of Epidemiology 165 (8): 868–71.

Miller, H. J. 1999. Measuring space-time accessibil-ity benefits within transportation networks: Basictheory and computational procedures. GeographicalAnalysis 31:187–212.

Mobley, L. R., E. D. Root, E. A. Finkelstein, O.Khavjou, R. P. Farris, and J. C. Will. 2006. Envi-ronment, obesity, and cardiovascular disease risk inlow-income women. American Journal of PreventiveMedicine 30 (4): 327–32.

Mokdad, A., M. Serdula, W. Dietz, B. Bowman,J. Marks, and J. Koplan. 1999. The spreadof the obesity epidemic in the United States,

1991–1998. Journal of the American Medical Asso-ciation 282:1519–22.

Morency, C., A. Paez, M. J. Roorda, R. Mercado,and S. Farber. 2009. Distance traveled in threeCanadian cities: Spatial analysis from the perspec-tive of vulnerable population segments. Journal ofTransport Geography 19 (1): 39–50.

Moudon, A. V., C. Lee, A. D. Cheadle, C. Garvin,D. B. Johnson, T. L. Schmid, and R. D. Weath-ers. 2007. Attributes of environments supportingwalking. American Journal of Health Promotion 21(5): 448–59.

Moudon, A. V., C. Lee, A. D. Cheadle, C. Garvin,D. B. Johnson, T. Schmid, R. D. Weathers, andL. Lin. 2006. Operational definitions of walka-ble neighborhood: Theoretical and empirical In-sights. Journal of Physical Activity and Health 3 (S1):S99–S117.

Nagendra, H. 2002. Opposite trends in response forthe Shannon and Simpson indices of landscape di-versity. Applied Geography 22 (2): 175–86.

Nawaz, H., W. Chan, M. Abdulrahman, D. Larson,and D. L. Katz. 2001. Self-reported weight andheight: Implications for obesity research. Amer-ican Journal of Preventive Medicine 20 (4): 294–98.

Norman, G. J., S. K. Nutter, S. Ryan, J. F. Sallis, K.J. Calfas, and K. Patrick. 2006. Community designand access to recreational facilities as correlates ofadolescent physical activity and body-mass index.Journal of Physical Activity and Health 3 (Suppl. 1):S118–S128.

O’Campo, P. 2003. Invited commentary: Advancingtheory and methods for multilevel models of resi-dential neighborhoods and health. American Jour-nal of Epidemiology 157 (1): 9–13.

Ogden, C. L., M. D. Carroll, L. R. Curtin, M. A.McDowell, C. J. Tabak, and K. M. Flegal. 2006.Prevalence of overweight and obesity in the UnitedStates, 1999–2004. Journal of the American MedicalAssociation 295 (13): 1549–55.

Oliver, L. N., N. Schuurman, and A. W. Hall. 2007.Comparing circular and network buffers to exam-ine the influence of land use on walking for leisureand errands. International Journal of Health Geo-graphics 6:41.

Openshaw, S. 1984. The modifiable areal unit problem.Norwich, UK: GeoBooks.

Pucher, J., and J. L. Renne. 2003. Socioeconomicsof urban travel: Evidence from the 2001 NHTS.Transportation Quarterly 57 (3): 49–77.

Ravera, O. 2001. A comparison between diversity,similarity and biotic indices applied to the macroin-vertebrate community of a small stream: TheRavella river (Como Province, Northern Italy).Aquatic Ecology 35 (2): 97–107.

Robert, S. A., and E. A. Reither. 2004. A multi-level analysis of race, community disadvantage, and

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


body mass index among adults in the US. Social Sci-ence and Medicine 59:2421–34.

Rundle, A., A. V. D. Roux, L. M. Freeman, D. Miller,K. M. Neckerman, and C. C. Weiss. 2007. The ur-ban built environment and obesity in New YorkCity: A multilevel analysis. American Journal ofHealth Promotion 21 (4): 326–34.

Saelens, B. E., and S. L. Handy. 2008. Builtenvironment correlates of walking: A review.Medicine and Science in Sports and Exercise 40:S550–S566.

Sallis, J. F., R. Nader, and J. Atkins. 1986. San Diegosurveyed for heart healthy foods and exercise facil-ities. Public Health Reports 101:216–18.

Shannon, C. E., and W. Weaver. 1949. The mathe-matical theory of communication. Urbana: Universityof Illinois Press.

Simpson, E. H. 1949. Measurement of diversity. Na-ture 163:688.

Smith, K. R., B. B. Brown, I. Yamada, L. Kowaleski-Jones, C. D. Zick, and J. X. Fan. 2008. Walkabilityand body mass index: Density, design, and newdiversity measures. American Journal of PreventiveMedicine 35 (3): 237–44.

Song, Y., and D. A. Rodriguez. 2005. The measure-ment of the level of mixed land uses: A syntheticapproach. White Paper Series. Chapel Hill, NC:Carolina Transportation Program.

Soobader, M., F. B. LeClere, W. Hadden, and B.Maury. 2001. Using aggregate geographic data toproxy individual socioeconomic status: Does sizematter? American Journal of Public Health 91 (4):632–36.

Southworth, M., and P. M. Owens. 1993. The evolv-ing metropolis: Studies of community, neighbor-hood, and street form at the urban edge. Journal ofthe American Planning Association 59 (3): 271–87.

Tilt, J. H., T. M. Unfried, and B. Roca. 2007. Usingobjective and subjective measures of neighborhoodgreenness and accessible destinations for under-standing walking trips and BMI in Seattle, Wash-ington. American Journal of Health Promotion 21 (4):371–79.

U.S. Census Bureau. 2000. Census 2000. Wash-ington, DC: U.S. Census Bureau. http://factfinder.census.gov (last accessed 30 March2010).

U.S. Department of Agriculture. 2005. Dietary guide-lines for americans. Washington, DC: U.S. Depart-ment of Health and Human Services.

Weiss, L., D. Ompad, S. Galea, and D. Vlahov.2007. Defining neighborhood boundaries for ur-ban health research. American Journal of PreventiveMedicine 32 (6S): S154–S159.

Wener, R. E., and G. W. Evans. 2007. A morningstroll: Levels of physical activity in car and masstransit commuting. Environment and Behavior 39(1): 62–74.

Wood, J. A. 2005. Economic benchmarks for Salt LakeCity’s central business district. Salt Lake City, UT:The Downtown Alliance.

Yen, I. H., and S. L. Syme. 1999. The social envi-ronment and health: A discussion of the epidemi-ologic literature. Annual Review of Public Health20:287–308.

Zeger, S. L., and K. Y. Liang. 1986. The analysis ofdiscrete and continuous longitudinal data. Biomet-rics 42:121–30.

Zick, C. D., K. R. Smith, J. X. Fan, B. B. Brown, I.Yamada, and L. Kowaleski-Jones. 2009. Runningto the store? The relationship between neighbor-hood environments and the risk of obesity. SocialScience & Medicine 69 (10): 1493–1500.

IKUHO YAMADA is an Associate Professor in theInterfaculty Initiative in Information Studies and theCenter for Spatial Information Science at the Univer-sity of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033 Japan. E-mail: [email protected] primary research interests reside in spatial statis-tics, GIScience, and spatial-temporal approaches topublic health and urban transportation issues.

BARBARA B. BROWN is a Professor in the Depart-ment of Family and Consumer Studies and Investi-gator at IPIA at the University of Utah, 228 S 1400E Room 228, Salt Lake City, UT 84112. E-mail:[email protected]. She is an environmen-tal psychologist who studies environmental, psycho-logical, and social aspects of neighborhoods as theyrelate to walkability, new urbanism, neighborhoodrevitalization, and transit-oriented designs.

KEN R. SMITH is a Professor in the Departmentof Family and Consumer Studies, an Investigatorwith the IPIA, and Director of the Pedigree andPopulation Resource at the University of Utah, 228S 1400 E Room 228, Salt Lake City, UT 84112.E-mail: [email protected]. He is a biodemog-rapher interested in familial aspects of health, aging,and longevity.

CATHLEEN D. ZICK is a Professor in the De-partment of Family and Consumer Studies and In-vestigator at IPIA at the University of Utah, 228 S1400 E Room 228, Salt Lake City, UT 84112. E-mail: [email protected]. She is a family and consumereconomist interested in household time allocation,household structure and economic well-being, andfamily/consumer policy.

LORI KOWALESKI-JONES is an Associate Pro-fessor in the Department of Family and Consumer

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14


Studies and Investigator at IPIA as well as at theHuntsman Cancer Institute at the University of Utah,228 S 1400 E Room 228, Salt Lake City, UT 84112.E-mail: [email protected]. She is in-terested in the ways in which youth are affected byfamily structure, community resources, and publicpolicy.

JESSIE X. FAN is a Professor in the Departmentof Family and Consumer Studies and Investigatorat IPIA at the University of Utah, 228 S 1400 ERoom 228, Salt Lake City, UT 84112. E-mail:[email protected]. She is a consumer and familyeconomist interested in consumer expenditure,saving, and credit use decisions and relevant policyissues.

Dow

nloa

ded

by [

Uni

vers

ity o

f C

hica

go L

ibra

ry]

at 0

6:15

18

Nov

embe

r 20

14

mixed land use and obesity: an empirical comparison of alternative land use measures and geographic...

Documents