urban studies - mitweb.mit.edu/ebj/www/55_files/zegras lee ben-joseph... · 2012-02-28 · (alfonzo...
TRANSCRIPT
http://usj.sagepub.com/Urban Studies
http://usj.sagepub.com/content/early/2012/01/05/0042098011429485The online version of this article can be found at:
DOI: 10.1177/0042098011429485
published online 5 January 2012Urban StudChristopher Zegras, Jae Seung Lee and Eran Ben-Joseph
Baby Boomers' Local Travel Behaviour in Suburban Boston, USBy Community or Design? Age-restricted Neighbourhoods, Physical Design and
Published by:
http://www.sagepublications.com
On behalf of:
Urban Studies Journal Limited
can be found at:Urban StudiesAdditional services and information for
http://usj.sagepub.com/cgi/alertsEmail Alerts:
http://usj.sagepub.com/subscriptionsSubscriptions:
http://www.sagepub.com/journalsReprints.navReprints:
http://www.sagepub.com/journalsPermissions.navPermissions:
What is This?
- Jan 5, 2012OnlineFirst Version of Record >>
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
By Community or Design? Age-restrictedNeighbourhoods, Physical Design and BabyBoomers’ Local Travel Behaviour inSuburban Boston, US
Christopher Zegras, Jae Seung Lee and Eran Ben-Joseph
[Paper first received, October 2009; in final form, August 2011]
Abstract
This article analyses the travel behaviour, residential choices and related preferencesof 55+ baby boomers in suburban Boston, USA, looking specifically at age-restrictedneighbourhoods. For this highly auto-dependent group, do neighbourhood-relatedcharacteristics influence local-level recreational walk/bike and social activity trip-making? The analysis aims to discern community (for example, social network)versus physical (for example, street network) influences. Structural equation models,incorporating attitudes and residential choice, are used to control for self-selectionand to account for direct and indirect effects among exogenous and endogenousvariables. The analysis reveals modest neighbourhood effects. Living in age-restricted, as opposed to unrestricted, suburban neighbourhoods modestly increasesthe likelihood of residents being active (i.e. making at least one local recreationalwalk/bike trip) and the number of local social trips. Overall, the age-restricted com-munity status has greater influence on recreational and social activity trip-makingthan the neighbourhood physical characteristics, although some community–neighbourhood interaction exists.
1. Introduction
Globally, the growing numbers of olderadults, combined with changes in metropol-itan settlement patterns, lifestyles and atti-tudes, have important implications forurban futures (for example, Champion,
2001). In many industrialised countries,‘baby boomers’, the generation born duringthe period of sustained high birth rates fol-lowing World War II, are now associatedwith distinctive approaches to consumption,
Christopher Zegras, Jae Seung Lee and Eran Ben-Joseph are in the Department of Urban Studiesand Planning, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. E-mail:[email protected], [email protected] and [email protected].
1–30, 2012
0042-0980 Print/1360-063X Online� 2012 Urban Studies Journal Limited
DOI: 10.1177/0042098011429485 at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
politics, personal finance, work and retire-ment, health and leisure (for example,Phillipson et al., 2008). Many of the indus-trialised world’s baby boomers came of ageduring a period of mass motorisation and sub-urbanisation and their travel behaviour hasemerged as an important issue. Particularly inthe US, the majority of older adults live in thesuburbs and the great majority of their trips(87 per cent in 2009) are by car (US DOT,2009). These lifestyle preferences, and ageingitself, pose challenges related to promotingactive lifestyles and healthy ageing and reduc-ing transport environmental and safety con-cerns (Rosenbloom, 2003). Examining babyboomers’ residential preferences and travelbehaviours can help, at least, to inform neigh-bourhood design, transport policy and mobi-lity service provision for the older adultcohort.
The range of innovations in products,markets and services increasingly gearedtowards an older adult society (Coughlin,2007) includes residential living alternatives,such as age-restricted neighbourhoods. TheUS’ first age-restricted neighbourhood,Youngtown, Arizona, was built in 1954 on a320-acre cattle ranch outside Phoenix andwas designed as a socially active, affordableand child-free setting (Blechman, 2008). Inthe US as of 2005, 43 per cent (29.6 million)of owner-occupied housing had at least onemember aged 55+ ; of these, approximately3.3 per cent (1 million) were in age-restricted neighbourhoods; age-restrictedhousing accounted for 11 per cent of newhomes bought by persons 55+ in 2003/04(Emrath and Liu, 2007). To a lesser extent,similar developments are appearing in otherWestern nations (Grant and Mittelsteadt,2004; Kennedy and Coates, 2008).
Our study explores the relationshipbetween age-restricted neighbourhoods andbaby boomers’ local travel habits. Ostensiblydesigned for older adult lifestyle prefe-rences, age-restricted neighbourhoods
might influence physical and/or social activ-ity among residents, leading to healthier life-styles. We examine this possibility, focusingon recreational walk/bike and local socialtrip-making among ‘leading-edge’ babyboomers (age 55–64 during data collection in2008): comparing age-restricted neighbour-hoods in suburban Boston with nearby non-age-restricted neighbourhoods; and assessingthe effects of neighbourhoods’ physical char-acteristics. That is, we test two sources ofbehavioural effects: those arising from social(and other unobserved) characteristics ofage-restricted neighbourhoods and thoseresulting from particular physical attributes.Although the setting is a metropolitan area inthe US, insights may apply elsewhere.
2. Background and ResearchQuestions
2.1 Key Concepts and Questions
In the US, baby boomers are generallyrecognised as those born between 1946 and1964—78.2 million persons (25 per cent ofthe nation) in 2005. We focus on ‘leading-edge’ baby boomers, now approachingretirement age and currently qualifying forage-restricted (55+ ) housing residency.This cohort is the ‘‘key demographic tar-geted by developers and marketers of activeadult housing’’ in the US (Heudorfer, 2005,p. 22). Here, ‘baby boomers’ refers to thisleading-edge cohort while ‘older adults’refers more generally to those 55+ .
The two basic categories of older adultneighbourhoods are: planned develop-ments, which include continuing careretirement communities offering on-sitenursing/care facilities and leisure-orientedretirement communities, typically orientedaround recreation (for example, golfcourses); and unplanned communities—i.e.‘naturally occurring retirement commu-nities’ (NORCs) that organically evolve
2 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
into neighbourhoods with the majority ofresidents aged 55+ (Hunt and Gunter-Hunt, 1985).
We examine a particular type of plannedolder adult neighbourhood, the age-restricted, active adult neighbourhood,1
and behavioural differences its residentsmay display relative to residents in unrest-ricted neighbourhoods. From this point:‘neighbourhood’ means a ‘‘geographicallybounded unit in which residents share prox-imity and the circumstances that come withit’’ (Chaskin, 1995, p. l); and, ‘community’means the broader network of interpersonalrelationships providing ‘‘sociability, sup-port, information, a sense of belonging, andsocial identity’’ (Wellman, 2005, p. 53). Acommunity might coincide with a neigh-bourhood; age-restricted neighbourhoodsaim, partly, to create community. Weexclude assisted living and congregated carefacilities to control for the potentially differ-ent travel capabilities of individuals withassisted-living needs and, subsequently, thepossible influences of neighbourhooddesigns specific to such residents. Thus,‘age-restricted’ refers to age-restricted,active adult neighbourhoods and ‘unrest-ricted’ refers to neighbourhoods withoutexplicit age restrictions. We use the age-restricted status as a proxy for community.
Finally, we study two types of individuals’local activities
—local recreational walking and bicycleuse—hereafter, ‘recreational non-motorised transport’ (NMT), becauseincreasing physical activity helpshealthy ageing and local NMT can sat-isfy recommendations for older adults’regular moderate physical activity (forexample, DiPietro, 2001; Eyler et al.,2003); and
—local social engagement—hereafter,‘social trips’—since being socially ‘dis-engaged’ may lessen physical andmental health and residential
neighbourhoods can maintain andincrease social networks via proximityand shared physical settings, enhan-cing residents’ well-being (for exam-ple, Kweon et al., 1998; Yang andStark, 2010).
Separating these trip types adds impor-tant nuance to the analysis, as neighbour-hoods and communities might vary in theirimpacts on different travel behaviours andindividuals may choose particular settings tosatisfy certain behavioural preferences; thesettings, in turn, may then influence otherbehaviours. In this paper, being ‘social’ andmaking social trips refer to individual char-acteristics and activities; ‘community’ refersto the broader network of interpersonal rela-tionships, as already defined and as we dis-tinguish based on the restricted/unrestrictedneighbourhood of residence.
2.2 Research Precedents
Scholars and others have long been inter-ested in older adults’ travel behaviour (forexample, Wachs, 1979). Relevant recentresearch includes transport’s contributionto older adults’ well-being in Vancouver(Cvitkovich and Wister, 2001), trip genera-tion rates and travel distances in London(Schmocker et al., 2005) and satisfactionwith travel opportunities in Sweden(Wretstrand et al., 2009). Studies have onlymore recently focused specifically on therelationship between the built environmentand older adults’ travel behaviour.
Due to our research focus, we limit theliterature review to studies of: ‘objective’measures of the built environment andrelationship with NMT use; and, effects ofneighbourhood and community character-istics on older adults’ walking behaviourand social engagement.
The built environment and walking.Research consistently reveals associations
BY COMMUNITY OR DESIGN? 3
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
between utilitarian walking and factors likeproximity to destinations and public tran-sit, street connectivity, mixed land use andhigher residential and job density (forexample, Baran et al., 2008; Giles-Corti andDonovan, 2002; Giles-Corti et al., 2005;Huston et al., 2003; Lee and Moudon,2006; Moudon et al., 2005; Saelens andSallis, 2003). On the other hand, researchfocused on walking for recreation and exer-cise provides inconsistent results (Owenet al., 2004). Some studies show that side-walks (Giles-Corti and Donovan, 2002),accessible destinations (Giles-Corti et al.,2005), hilliness (Lee and Moudon, 2006)and perception of attractiveness and safety(Alfonzo et al., 2008; Giles-Corti andDonovan, 2002) are associated with a higherlevel of recreational walking; other studiesfail to reveal such correlations (Rodriguezet al., 2006; Saelens and Sallis, 2003).
Older adults’ NMT use. King et al.(2003), examining older women (averageage 74) in suburban and urbanPennsylvania, find a positive correlationbetween physical activities (pedometer-measured) and convenient destinations andperceived walkability. Berke et al. (2007a)find neighbourhood walkability in KingCounty, Washington—measured via a spa-tial buffer of households and accountingfor characteristics like dwelling unit densityand proximity of grocery stores—to beinversely associated with depressive symp-toms in older (65+ ) men (but notwomen). Berke et al. (2007b) also find astatistically significant relationship betweenthe same walkability measure and fre-quency of older persons’ (65+ ) walking forphysical activity. Examining older people’s(65+ ) travel behaviour in northernCalifornia and controlling for attitudes,Cao et al. (2010) find that several neigh-bourhood characteristics (for example,
safety, distances) influence walk trip fre-quencies. Joseph and Zimring (2007)examine older adults’ (age 77–83) pathchoice in three continuing care retirementneighbourhoods in Atlanta, finding anassociation between: well-connected,destination-oriented paths and utilitarianwalking; and longer, well-connected pathswithout steps and recreational walking.Finally, using multilevel regression, Nagelet al. (2008) find that high-volume streetsand proximity to destinations positivelyinfluence total walking time among olderadults (average age 74) in Portland(Oregon), while low-volume streets have anegative influence on total walking time.They find no association between the builtenvironment and the odds of not walking,suggesting no neighbourhood influence onsedentary older adults’ walking behaviour.
Neighbourhood, community and olderadults’ social activities and/or well-being. Early US studies took a building-level perspective, often focusing ongovernment-supported housing (Lawtonet al., 1975). In Portland (Oregon),Chapman and Beaudet (1983) found olderadults’ (average age 78) interactions withneighbours to be highest in ‘good quality’neighbourhoods, more distant from thecity centre, and with low shares of olderpeople. Kweon et al. (1998) found a posi-tive association between time spent incommon outdoor green spaces and mea-sures of social integration and ‘sense oflocal community’ among poor 64+ adults(average age 68) in Chicago’s age-integrated public housing. Finally, Yangand Stark (2010), using qualitative meth-ods, find apparent behavioural influencesof social features related to expectations ofencounters and homogeneity of residentsin assisted living facilities (stand-alonebuildings).
4 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
2.3 Age-restricted Neighbourhoods andTravel Behaviour: Hypotheses
Little research has focused specifically onlocal travel behaviour in age-restrictedneighbourhoods. As mentioned, Joseph andZimring (2007) examined walk path choicein continuing care retirement neighbour-hoods. Flynn and Boenau (2007) estimatedvehicular traffic counts for a suburbanVirginia age-restricted neighbourhood, find-ing trip rates comparable to those recom-mended for detached senior adult housingby the Institute of Transportation Engineers.
Avoiding the broader debate about age-restricted neighbourhoods (see Blechman,2008), we identify features that may influ-ence local NMT use and social engagement.Specifically, relative to unrestricted neigh-bourhoods, age-restricted neighbourhoodsmay differ by (Hebbert, 2008)
—demographics: people of similar agesand interests, combined with physicaldisconnection from surroundingneighbourhoods, may decrease thelikelihood of encountering strangersin day-to-day activities;
—community: programmes, events, club-houses may increase residents’ activitylevels;
—suitability: targeting the 55+ demo-graphic and offering lifestyle choiceamenities (like golf courses, pools)may support more active living; and
—walkability: trails and sidewalks, andlittle, if any, through-traffic mayincrease walking.
Age-restricted and unrestricted neigh-bourhoods also share many similarities. Thegreat majority (71 per cent) of age-restrictedneighbourhoods in the US are suburban,even more suburban than overall locationsof older adult households (Emrath and Liu,2007).2 This implies limited connectivity toother neighbourhoods, limited local retail,
dispersed employment and other services,and limited public transport. In this subur-ban context, we focus locally, where physicaland community differences and, thus,potential behavioural effects may arise.
Do age-restricted neighbourhoods influ-ence local travel behaviour? The socialecological model offers a theoretical framefor local travel behaviour in age-restrictedneighbourhoods, emphasising the recipro-cal interactions between behavioural andenvironmental factors. Presuming thatchanges in community alter individualbehaviours, the model focuses on relation-ships between environmental interventionsand interpersonal, organisational and othercommunity factors (Sallis and Owen,1996). Age-restricted neighbourhoods maysupport older adults through peer groups,social programmes and higher perceivedsafety, among other things (for example,Ahrentzen, 2010). We hypothesise that,after controlling for physical characteristics,age-restricted neighbourhoods have morerecreational NMT and social trips due tocommunity effects.
Do neighbourhood characteristics influ-ence local travel behaviour? As Maatet al. (2005) propose, a neighbourhood’sphysical characteristics may influence travelbehaviour via effects on net utility—theutility of travel (for example, number,quality, distribution of destinations) less itsdisutility (actual and perceived travelcosts). Consider, for example, prototypicalstreet configurations: linear, loop and grid(see Figure 1). The latter two reduce non-duplicative routes (reducing travel’s disuti-lity) and, by clustering dwellings, increaseopportunities to meet neighbours (increas-ing travel’s utility). Thus, we hypothesisethat grid- and loop-type neighbourhoodspromote more recreational NMT and
BY COMMUNITY OR DESIGN? 5
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
social trips, as do higher intersection den-sity, neighbourhood facilities (for example,parks, golf course) and proximity to otherdestinations, including public transportstops.
2.4 Analytical Challenges and SpecificModelling Precedents
Aiming to show whether neighbourhoods’community and physical characteristicsproduce different activity patterns, poses theclassic causality challenge, associated with‘self-selection’ (Mokhtarian and Cao, 2008).At least two related forms of bias may bepresent: simultaneity bias (for example,individuals who prefer walking choosing tolive in walkable neighbourhoods); andomitted variable bias (unobserved variables,like preferences for walking, produce thetravel outcome (walking), but also correlatewith neighbourhood characteristics).3 Inother words, the presumed exogenouscausal variable, the neighbourhood, is actu-ally endogenous, which can produce incon-sistent and biased estimators. Mokhtarianand Cao (2008) review the issues and possi-ble analytical and research design solutions.Cao et al. (2009) review 38 empiricalstudies using nine different approaches tocontrol for ‘self-selection’—direct ques-tioning, statistical control, instrumentalvariables, sample selection models, propen-sity score matching, other joint models ofresidential and travel choices (for example,structural equation models) and longitudi-nal studies.
Only statistical control and structuralequation models are reviewed here.Statistical control directly incorporates atti-tudes and preferences into the behaviouralmodel, thereby isolating these effects fromneighbourhood-level effects. Studies typi-cally use specialised survey data, includingattitudes and preferences (for example,measured on a Likert scale), in a two-step
approach: factor analysis on the indicators(since multiple preferences/attitudes aremeasured); and, behavioural modelling,including fitted values from the first step(for example, Cao et al., 2006, 2010).Problematically, the estimation of thesecond step is inconsistent because thefitted latent variables (from the first step)include measurement error by droppingerror terms (Ben-Akiva et al., 2002).
The latter problem can be addressed withstructural equation modelling (SEM), ananalytical tool introduced in the travel beha-viour field in the 1980s (Golob, 2003) andmore recently applied to the self-selectionissue (Cao et al., 2009). A full SEM usessimultaneously estimated measurementmodels, for endogenous and exogenousvariables, and a structural model, and cancapture influences of exogenous on endo-genous variables and among endogenousvariables (Golob, 2003). SEM measurementmodels are similar to exploratory factor ana-lytical approaches, except in restricting theparameters defining factors and specifyingcovariances among unexplained portions ofboth unobserved and latent variables(Golob, 2003). The estimated parametersmake the predicted variance–covariancematrix as similar as possible to the observedvariance–covariance matrix, subject tomodel constraints. SEM can distinguishbetween direct and total effects and, withsimultaneous measurement equations oflatent variables, allows consistent incorpora-tion of attitudes and preferences in beha-vioural models and captures potentialbi-directional influences between attitudesand travel behaviour (Mokhtarian and Cao,2008).
Few studies have used SEM to introducelatent attitudinal variables in the built envi-ronment/travel behaviour context. Abreuet al. (2006) used SEM in analysing adultworkers’ travel in Lisbon, treating short-and longer-term travel behaviours and
6 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Figure 1. Three categories of neighbourhood street patterns, descriptive diagrams and proto-typical examples of the categorisation.Source: World Imagery, provided by ESRI (http://www.arcis.com/home/item.html?id=10df2279f9684e4a9f6a7f08febac2a9).
BY COMMUNITY OR DESIGN? 7
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
residence and workplace land use character-istics (latent variables identified throughexploratory factor analysis) as endogenousvariables and individual socioeconomicvariables as exogenous. The approach par-tially accounts for self-selection while notexplicitly including attitudinal effects; thestructural and measurement models are notestimated simultaneously. Bagley andMokhtarian (2002) included attitudes in aSEM, including endogenous variables (tworesidential type variables, one job locationvariable, three travel demand variables andthree attitude variables) and exogenous vari-ables (socio-demographics, lifestyle factors,attitude measures). They found that atti-tudes and lifestyles exerted the greatestinfluence on travel behaviour, while residen-tial location type had little impact. Thestudy represented neighbourhood charac-teristics via factor scores on two dimensions(traditional versus suburban) and includedlatent variables as fitted values of factoranalysis on indicators, rather than simulta-neously estimating structural and measure-ment equations. Similar to Abreu et al.(2006), their model is path analysis ratherthan complete SEM.
In summary, for the highly automobile-dependent, yet relatively understudied, babyboomer generation in the suburban US, weask the question: do neighbourhood-relatedcharacteristics influence local-level recrea-tional walk/bike and social activity trip-making? Drawing from social ecologicaltheory and utility-based travel behaviourtheory, our analysis aims to discern com-munity (for example, social network) versusphysical (for example, street network) influ-ences. Unlike most previous research in thisfield, we use full structural equation models,incorporating attitudes and residentialchoice, to control for self-selection and toaccount for direct and indirect effectsamong exogenous and endogenousvariables.
3. Research Context and Design
Greater Boston includes 164 cities andtowns, with 4.45 million persons (in 2000),across 2832 square miles (6107 square km).Just over 20 per cent of residents are olderadult (US Census Bureau, 2002), a cohortexpected to increase by 50 per cent between2000 and 2020 (Heudorfer, 2005).Approximately 8.5 per cent of GreaterBoston residents in 2000 were ‘leading-edge’ boomers (US Census Bureau, 2002),a group slightly more suburban than theoverall population.4
These demographic trends, and local landuse policies and fiscal considerations, havefuelled age-restricted development. State-wide, Heudorfer (2005) found 150 age-restricted neighbourhood developmentscompleted or under construction in 93 citiesand towns, implying a supply of more than10 000 housing units, with another 170 age-restricted developments in pre-constructionor seeking permissions in 109 towns. Mostdevelopments have fewer than 100 dwellingunits and include walking paths, meetingrooms and clubhouses, with fewer providingon-site shops, bike trails and golf facilities(Heudorfer, 2005).
3.1 Survey Design and Data
We use a quasi-experimental, cross-sectionalresearch design comparing suburban age-restricted and unrestricted neighbourhoodsin Greater Boston. The age-restricted neigh-bourhoods were first identified—via realestate listings, information from developersand other resources5—based on the follow-ing criteria: built out and occupied; entirelyor mainly age-restricted; and ‘active adult’(for example, not a continuing care facility).Thirty-five age-restricted neighbourhoodsmet the initial criteria. From this list, 20neighbourhoods were selected (see Table 1),by filtering out recent developments (to
8 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
ensure potential residency of at least threeyears) and small developments (less than 30units on a single street). The final sampledage-restricted neighbourhoods range in sizefrom 40 to 1100 dwelling units with a meanof 160 and median of 66 units. Our modelscontrol for the possible influence of neigh-bourhood size by including total streetlength in each age-restricted and unrestrictedneighbourhood. Overall, the selected neigh-bourhoods are biased towards more recentdevelopments and/or ones with recent realestate activity.
Each age-restricted neighbourhood wasmatched with unrestricted surroundingsusing postal codes to approximate similarregional accessibility and demographics.Mailing addresses were requested fromUSAData, a commercial data vendor, forresidents aged 55–65, generating 34 108names. We identified 1237 households inage-restricted neighbourhoods by matchingstreet names against the purchased list. Wethen randomly sampled 5763 households
from unrestricted areas, producing a totalsample size of 7000 households. We purpo-sely oversampled unrestricted areas, expect-ing to receive a lower response rate fromthe cohort of interest there. Our samplingapproach is endogenously stratified.
Mailed survey packages included a $5non-contingent cash incentive, a travelsurvey for retrospective trip counts over thepast week; attitudinal questions, such as pre-ferences for walking and cycling (five-pointLikert scale); and household/individualquestions (for example, income, employ-ment status). We received 1650 householdresponses, 1422 after excluding problematicresponses (effective response rate of 20 percent): 349 from age-restricted neighbour-hoods (28 per cent response rate) and 1073from unrestricted neighbourhoods (19 percent response rate). Households included1859 individuals (470 age-restricted; 1389unrestricted).6 Among the 20 age-restrictedneighbourhoods, responses came from 15(Table 1).
Table 1. Age-restricted neighbourhoods examined (15 from which we received responses: 28per cent response rate)
ID Community Households Persons Map
1 Adams Farm 14 218 Deerfield Estate 7 109 Delapond Village 2 211 Eagle Ridge 11 1517 Leisurewoods 25 3120 Oak Point 95 12821 Pinehills 87 11623 Red Mill 6 825 Southport 35 4527 Spyglass Landing 5 630 The Village at Meadwood 16 2231 The Village at Orchard Meadow 17 2332 Village at Quail Run 11 1533 Vickery Hills 14 2335 Wellington Crossing 4 5
Total 349 470
Notes: The shading on the map indicates towns with one or more ARAACs, as tabulated byHeudorfer (2005). Numbered dots indicate locations of ARAACs identified for this study.
BY COMMUNITY OR DESIGN? 9
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Neighbourhood characteristics weremeasured using a geographical informationsystem (GIS) and public and private datasources, based on household location (iden-tifiable to the centroid of a 250-metre gridcell; see the example in Figure 1).7
3.2 Measures and Descriptive Statistics
Table 2 includes descriptive statistics of keyvariables, including outcomes of interest,reported weekly: NMT trips, representingrecreational walking/biking trips; and socialtrips, measuring visits to neighbours andrepresenting local social engagement.Respondents in age-restricted neighbour-hoods have only slightly higher averageweekly trip rates for both trip purposes. Alarge share of individuals in both neigh-bourhood types report making zero NMTand social trips during the week (hereafter,these individuals are ‘non-active’ and ‘non-social’). Unrestricted neighbourhoods havea 10 per cent higher share of non-activeand a 13 per cent higher share of non-social individuals. Baby boomers residingin age-restricted neighbourhoods tend tobe less employed, slightly healthier andslightly older, with fewer owning a bike ormore than three cars.
Age-restricted neighbourhoods havemore local facilities, such as public spaces,and, primarily, loop street patterns. Nonehas grid streets. Nearly 50 per cent of unrest-ricted neighbourhoods have linear streetpatterns. Other physical characteristics—such as intersection density, destinationsand proximity to public transport—do notsignificantly differ between sampledrestricted/unrestricted neighbourhoods.
Exploratory factor analysis on theresponses to the questions regarding resi-dential preferences led us to hypothesisetwo latent variables: Pro Walkability, denot-ing preference for walkable neighbour-hoods, and Pro Segregation, representing
preference for neighbourhoods segregatedby age and social class. Confirmatory factoranalysis confirms this latent structure:fixing the indicators most highly correlatedwith the two latent variables at 1 for identi-fication, all other indicators significantlycontribute to the latent variables. Thislatent construct serves as a measurementmodel in the following SEM.8
4. Behavioural Modelling
The large share of zero-reported NMT andsocial trips (Table 2) indicates censoring—ordinary count models may be inappropri-ate. We employ a zero-inflated model,allowing zeros to remain in the count modelby estimating an individual’s likelihood ofbeing in the ‘zero’ group. Taking recrea-tional NMT trips as an example, a binarylogit model estimates the probability ofbeing non-active and active. These probabil-ities weight the zeros in the count modelsuch that the probability of observing zerofor an individual equals the probability ofbeing non-active plus the probability ofbeing active, multiplied by the probability ofobserving zero in the count model (Jones,2005) (Figure 2, Equations (1)–(3)). Thisproduces two sets of coefficients. The logitmodel results indicate the variables’ influ-ence on the likelihood of being non-active;negative coefficients imply a higher prob-ability of being active. The count model esti-mates trip counts for the active group;positive coefficients mean a higher fre-quency of recreational NMT trips.
We apply zero-inflated negative binomial(ZINB) models9 with SEM that simultane-ously incorporates attitudes possibly affect-ing residential choice/travel behaviour anda residential choice model. Three types ofrelationship are examined—residentialchoice, residential preference and travelbehaviour (Figure 2)—and three models
10 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
le2.
Des
crip
tive
stat
isti
csby
nei
ghbourh
ood
type
and
test
sofdif
fere
nce
s
Var
iabl
esT
otal
Gro
up
mea
n(S
.D.)
NM
ean
(S.D
.)A
ge-r
estr
icte
dU
nre
stri
cted
Mea
nd
iffe
ren
ce
Dep
end
ent
vari
able
sN
MT
trip
Las
tw
eek,
ho
wm
any
tim
esd
idyo
uw
alk
or
cycl
efo
rex
erci
sein
you
rn
eigh
bo
urh
oo
d?
1761
2.23
5(2
.417
)2.
629
(2.4
51)
2.10
1(2
.391
)0.
528**
Ind
ivid
ual
sre
po
rtin
gze
roN
MT
trip
so
ver
pas
tw
eek
(i.e
.n
on
-act
ive)
704
0.40
00.
324
0.42
60.
102**
Soci
altr
ipL
ast
wee
k,h
ow
man
yti
mes
did
you
visi
tyo
ur
nei
ghb
ou
rs?
1755
0.80
1(1
.322
)1.
084
(1.4
72)
0.70
6(1
.253
)0.
378**
Ind
ivid
ual
sre
po
rtin
gze
roso
cial
trip
so
ver
pas
tw
eek
(i.e
.n
on
-so
cial
)10
750.
613
0.51
40.
646
0.13
2**
Qu
esti
onpr
edic
tor
(tre
atm
ent)
AR
AA
CA
ge-r
estr
icti
on
stat
us
(0.n
ot
rest
rict
ed;1
.age
-res
tric
ted
)18
590.
253
——
—
Soci
o-ec
onom
icch
arac
teri
stic
sE
mpl
oyE
mp
loym
ent
stat
us
(0.
un
emp
loye
d;
1.em
plo
yed
)18
460.
637
0.51
00.
680
0.17
0**
Hea
lth
yH
ealt
hst
atu
s(0
.u
nh
ealt
hy;
1.h
ealt
hy)
1859
0.85
10.
892
0.83
70.
054*
Mal
eG
end
er(0
.fe
mal
e;1.
mal
e)18
490.
472
0.44
40.
482
0.03
8A
geR
esid
ents
’ag
e17
6061
.195
(3.8
75)
62.6
51(3
.750
)60
.687
(3.7
90)
1.96
3**
Hig
h-i
nco
me
Hig
han
nu
alh
ou
seh
old
inco
me
($10
0kan
do
ver)
(0.
oth
erw
ise;
1.h
igh
inco
me)
1761
0.29
80.
330
0.28
70.
043
Mid
-in
com
eM
ediu
man
nu
alh
ou
seh
old
inco
me
($50
k–99
.9k)
(0.
oth
erw
ise,
1.m
ediu
min
com
e)17
610.
496
0.50
10.
495
0.00
6
Low
-in
com
e(b
ase)
Lo
wan
nu
alh
ou
seh
old
inco
me
(les
sth
an49
.9k)
(0.
oth
erw
ise,
1.lo
win
com
e)17
610.
205
0.16
90.
218
0.04
9*
Th
ree
veh
icle
sT
hre
ean
dm
ore
veh
icle
sin
ah
ou
seh
old
(0.
less
than
3ve
hic
les;
1.3+
veh
icle
s)16
680.
259
0.13
70.
301
0.16
4**
Bik
eB
ikes
ina
ho
use
ho
ld(0
.N
ob
icyc
les;
1.m
ore
than
on
eb
icyc
le)
1645
0.57
30.
519
0.59
20.
072**
(con
tin
ued
)
BY COMMUNITY OR DESIGN? 11
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
le2.
(Conti
nued
)
Var
iabl
esT
otal
Gro
up
mea
n(S
.D.)
NM
ean
(S.D
.)A
ge-r
estr
icte
dU
nre
stri
cted
Mea
nd
iffe
ren
ce
Nei
ghbo
urh
ood
char
acte
rist
ics
Gri
dN
eigh
bo
urh
oo
dst
reet
pat
tern
sgr
idty
pe
(0.o
ther
wis
e;1.
grid
)45
80.
234
—0.
242
0.24
2*
Loo
pN
eigh
bo
urh
oo
dst
reet
pat
tern
slo
op
typ
e(0
.o
ther
wis
e;1.
loo
p)
458
0.29
50.
800
0.27
80.
522**
Lin
ear
Nei
ghb
ou
rho
od
stre
etp
atte
rns
lin
ear
typ
e(0
.o
ther
wis
e;1.
lin
ear)
458
0.47
20.
200
0.48
10.
281*
Inte
rsec
td
ensi
tyT
rue
inte
rsec
tio
nd
ensi
ty(t
rue
inte
rsec
tio
ns/
100m
etre
so
fst
reet
s)45
80.
322
0.39
20.
320
0.07
2
Fac
ilit
ies
Pre
sen
ceo
fp
ub
lic
spac
eso
rsp
ort
sfa
cili
ties
inn
eigh
bo
urh
oo
ds
(0.
no
,1.
yes)
458
0.34
90.
733
0.33
60.
397**
Des
tin
atio
n40
0P
rese
nce
of
‘pla
ces
of
inte
rest
’in
Go
ogl
eE
arth
wit
hin
400m
etre
s(0
.n
o;
1.ye
s)45
80.
448
0.60
00.
442
0.15
8
MB
TA
bus
stop
Pre
sen
ceo
fM
BT
Ab
us
sto
ps
wit
hin
1km
(0.
no
;1.
yes)
458
0.06
60.
133
0.06
30.
070
Com
mu
ter
rail
Pre
sen
ceo
fco
mm
ute
rra
ilw
ith
in1
km(0
.n
o;
1.ye
s)45
80.
222
0.20
00.
223
0.02
3St
reet
len
gth
To
tal
stre
etle
ngt
ho
fa
nei
ghb
ou
rho
od
(km
)45
82.
977
5.39
62.
895
2.50
1
Not
es:*
p\
0.05
,**
p\
0.01
,in
dic
atin
gsi
gnif
ican
cele
vels
of
dif
fere
nce
of
mea
ns/
pro
po
rtio
ns;
—in
dic
ates
no
tap
pli
cab
le.
Som
eto
tals
dif
fer
du
eto
mis
sin
git
ems
inth
esa
mp
le.
12 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
are estimated (the Appendix provides fullresults). Model 1 has no control for self-selection. Model 2 attempts to control for
individuals’ self-selection for neighbour-hood physical characteristics by simultane-ously estimating the ZINB model and the
Figure 2. Path diagrams and equations of three models that hypothesise relationships amongthe built environment, residential preference and travel behaviour.
BY COMMUNITY OR DESIGN? 13
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
latent variable (attitudinal) model’s struc-tural and measurement equations. Model 2estimates residential preferences conditionalupon socioeconomic characteristics; thus,travel behaviour and residential preferencesare endogenous while socioeconomic statusand neighbourhood physical characteristicsand age-restricted status are exogenous.Model 3 includes a binary choice model forage-restricted status, assuming that peopleselect age-restricted neighbourhoods to sat-isfy community preferences, and that neigh-bourhood and individual characteristicsinfluence age-restricted choice.10 In thiscase, age-restricted neighbourhood choice,residential preferences and individual andneighbourhood characteristics jointly affectlocal travel behaviour. We estimate themodels in Mplus 5.0, using a normal theorymaximum likelihood (ML) estimator andaccounting for non-independence amongobservations from the same household11
(Muthen, 1998–2004; Muthen and Muthen,1998–2007).
4.1 Recreational NMT
The recreational NMT trip results directlysupport our first hypothesis regarding theeffect of age-restricted setting, with somesupport for neighbourhood physical charac-teristics, but only ‘bundled’ with age-restric-tion. Figure 3 orients the discussion.
Examine, first, the age-restricted neigh-bourhood choice: after controlling forneighbourhood characteristics, age-restrictedneighbourhoods attract older, higher-income people who prefer segregated neigh-bourhoods. Males are less likely to chooseage-restriction. Age-restricted neighbour-hoods with loop-type streets, higher inter-section density and on-site facilities aremore attractive.
Looking at the likelihood of being non-active, neighbourhood physical characteristics—loop street type, intersection density,
presence of local facilities, total streetlength—do have an influence, but onlyindirectly, via the age-restricted choice.Community and design primarily influencethe non-active likelihood, with only nearbydestinations exerting a significant effect onnumber of trips among the active. Nearbycommuter rail, interestingly, negatively cor-relates with number of recreational NMTtrips.
For individuals, being employed increasesthe likelihood of being non-active anddecreases the number of NMT trips amongthe active. Being healthy decreases the likeli-hood of being non-active. Finally, ‘pro-Walkables’ are less likely to be non-active,while ‘pro-Segregated’ are more likely to be.This latter effect is partly offset by the pro-Segregated choosing age-restricted neigh-bourhoods that, in turn, increase the likeli-hood of being active.
There is little evidence of self-selectionfor local NMT trips. While both latent atti-tudinal constructs significantly affect thechoice to be active, they do not change thesign, significance or magnitude of the age-restricted effect. Age-restricted communitysettings increase the chance that residentswill make local recreational NMT trips;perhaps, in part, due to neighbourhoodphysical characteristics. Other than nearbydestinations’ effect on NMT trip counts,neighbourhood physical characteristics donot directly affect baby boomers’ beingactive or the number of recreational NMTtrips.
4.2 Social Trips
The social trip results also only partially sup-port our hypotheses, with distinct, some-what counter-intuitive, differences relativeto recreational NMT. Age-restricted settingsand, indirectly, their bundled physical char-acteristics exert an uncertain influence onthe number of social trips. Physical
14 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Figu
re3.
Path
dia
gram
and
resu
lts
ofre
crea
tion
NM
T(l
eft)
and
soci
altr
ips
(rig
ht)
model
s.N
ote
s:R
esults
are
from
model
3in
the
Appen
dix
(Tab
les
A1,A
2);
yp
\0.1
0;*
p\
0.0
5;**
p\0.0
1.
BY COMMUNITY OR DESIGN? 15
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
characteristics themselves only modestly(and uncertainly) influence the likelihood ofbeing ‘social’. Again, Figure 3 guides thediscussion.
The same age-restricted choice model asfor recreational NMT holds. However, con-trary to the NMT case, age-restricted neigh-bourhoods do not affect being social; amongthe social, age-restriction increases socialtrip-making. This result should be viewedwith some uncertainty (p-value = 0.075)and suggests residential self-selection vis-a-vis social trip-making (compare the signif-icance of the age-restricted neighbourhoodcoefficient from model 1 with models 2 and3; Appendix, Table A2). Those inclined tomake more social trips may select age-restricted settings (and, possibly, theirphysical characteristics) to satisfy socialtrip-making tendencies. Regarding directphysical effects, street typologies are insig-nificant. Nearby commuter rail is associatedwith being social and making more localsocial trips (p\0.10).
For individuals, being employedincreases the likelihood of being non-social.Unsurprisingly, being employed reducesweekly local social trip-making. Olderboomers have greater likelihood of makingmore social trips. Finally, those preferringsegregated neighbourhoods have a higherlikelihood of making more social trips.
Social trips offer stronger evidence of self-selection in this study. While age-restrictedneighbourhoods appear to be associatedwith more weekly social trips among thesocially inclined, statistical support for thiseffect declines once accounting for attitudesand residential preferences.
5. Implications and Shortcomings
Our findings must be viewed in light of thedemographic geography of older adults inthe metropolitan US: the majority live in
auto-dependent suburbia. Among oursampled individuals, for example, 93 percent of daily reported trips were by auto-mobile (Hebbert, 2008), even higher thanthe automobile mode share for GreaterBoston’s baby boomers.12 This study shedslittle light on the larger challenges implied.Nonetheless, with respect to two types oflocal travel activities that may be influencedby suburban neighbourhood and communitycharacteristics and play an important role inhealthy ageing, some influences emerge.
We find modest effects of neighbourhoodage-restricted status and physical characteris-tics on weekly recreational NMT and socialtrip-making. Distinguishing between thosewho do and do not make a recreationalNMT or social trip provides useful informa-tion. Eyler et al. (2003), studying adults inthe US, identified three types of walker: regu-lar, occasional and never. Occasional andnever walkers lacked time for walking andnever walkers reported feeling unhealthier,while regular walkers reported more self-confidence and social support for walking.Our recreational NMT results support thesefindings and suggest a design and commu-nity (social network) role: those with a ‘pro-Walkable’ mindset are more likely to beactive; the community and, indirectly, designaspects of age-restricted neighbourhoodsincrease residents’ likelihood of being active,after controlling for self-selection. This pro-vides some support for the social ecologicalmodel of health promotion—the social-physical setting of the age-restricted neigh-bourhoods apparently provides a mediumfor active living (for example, Wister, 2005).Among the active, however, the neighbour-hood has no effect on increased recreationalNMT trip-making, although nearby destina-tions do play a role. The age-restricted effectmay come from social settings (i.e. commu-nity) or other unobserved (or non-compara-ble) physical characteristics distinguishingage-restricted from unrestricted suburbs. For
16 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
example, the age-restricted neighbourhoodsstudied have more local facilities (for exam-ple, clubhouses) than typical suburbs (Table2); while insignificant in the NMT models,these variables’ effects may be masked by theage-restricted label.
As in the recreational NMT case, someage-restricted physical characteristics (inter-section density, neighbourhood facilities anddestinations) indirectly influence social trip-making among the social. In this case, how-ever, residents may be purposefully choosingage-restricted settings and their related designattributes: age-restricted settings will not‘make’ people social, but may attract thosewith higher social trip-making tendencies.
Our findings indicate the importance ofdistinguishing between trip types, includingwhen attempting to control for self-selection. The results confirm intuition: anindividual may choose a neighbourhood tosatisfy desired local social activity; this resi-dential choice to satisfy one activity prefer-ence might then induce changes in otheractivities.
5.1 Limitations and Future Research
Our results are only directly applicable to aspecific demographic, geography and time ofyear (i.e. April 2008) and may not be generali-sable. Even for the specific groups and areasstudied, it is likely that the sampling proce-dure suffers from biases that further limit theresults’ validity and generalisability.
The age-restricted effects may be con-founded by our not knowing whether someof the unrestricted neighbourhoods alsohave a high share of older adults (i.e. beingNORCs), implying similar communitystructures. This relates to spatialdependence—participation in a particularactivity may be influenced by surroundingneighbourhoods, including how well ‘inte-grated’ the neighbourhoods are with their
surroundings, only crudely proxied here.The age-restricted neighbourhoods’ relativenewness may also confound; newer resi-dents13 may still be ‘exploring’ surround-ings, effects indistinguishable from theage-restricted status. Over time, such effectsmay diminish or intensify—an area for fur-ther study.
Analytically, complete SEM—simultaneously estimating measurementmodels of latent attitudinal variables andbehavioural (structural) models—represents an important advance. It con-trols for self-selection based on attitudesand residential choice and allows testingmore complex relationships, includingdirect and indirect effects. The increasedmodelling sophistication also comes at acost—our particular SEM cannot easilyreveal relative or marginal effects, only sig-nificance and directionality. Furthermore,the design remains cross-sectional, asopposed to temporal (i.e. measuringchange). For example, people living in asociable community and/or a social-oriented neighbourhood may increase, overtime, their socialising, which may thenchange the community (for example, walk-ing groups); revealing these dynamicswould require longitudinal analysis.
Questions can be raised about the out-comes measured: self-reported recreationalNMT trips in the neighbourhood and socialtrips to ‘neighbors’. Respondents may inter-pret the extent of ‘neighbourhood’ and/or‘neighbours’ differently. Further, the mea-sures may be weak proxies for outcomesmore closely related to healthy ageing, suchas: minutes of activity per day, health condi-tions, levels of social engagement, strengthof social networks and/or mental healthconditions. Analogously, the validity andreliability of the attitudes/preferences ques-tions are uncertain and treating the ordinalLikert-value attitude scores as continuous
BY COMMUNITY OR DESIGN? 17
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
variables (in the factor analysis), althoughcommon practice, may be problematic.
Regarding neighbourhood built environ-ment, we attempted ‘objective’ measure-ment, which may not account for designqualities like sense of safety and human scale(Ewing and Handy, 2009) and may ignoreindividual perceptions of relevant factors.Again, these perceptions may change overtime and be influenced by neighbourhoodand/or community changes. Enhancedbehavioural insights might come from com-bining qualitative measures of the built envi-ronment with ‘objective’ measures.
Further research could examine addi-tional travel behaviours among babyboomers and/or compare suburban andurban baby boomers or age-restrictedneighbourhoods with non-age-restrictedmaster planned neighbourhoods. Suchcomparisons may reveal whether themodest behavioural effects of age-restrictedneighbourhoods derive from the communitystructure, physical features or their recipro-cal interactions. Additional topics worthexamining include: the potential to retrofitexisting neighbourhoods to serve the needsof older adults more effectively; whetherspatial concentrations of older adults in sub-urbia increase possibilities for new transportand/or other older adult services; and therelationship between commuter rail proxim-ity and local trip-making.
Our results indicate the need to reach abetter understanding of how physical andsocial structures interact to influence olderadults’ activities. Overall, however, therelative locations of older-adult-orientedneighbourhoods need attention. For exam-ple, just 13 per cent of the age-restrictedneighbourhoods studied are within 1 kmof a bus stop and 20 per cent within 1 kmof commuter rail. As ageing meansreduced driving capabilities, this relativeautomobile-dependency may pose aproblem.
6. Conclusion
We studied a neighbourhood type cateringfor older adults—age-restricted, activeadult neighbourhoods—and attempted todiscern community (for example, socialnetwork) versus physical (for example,street network) influences on suburbanbaby boomers’ travel behaviour. Usingstructural equation models, the analysisattempts to control for self-selection basedon attitudes and residential choice, allow-ing for direct and indirect effects amongexogenous and endogenous variables.
The age-restricted neighbourhoodsattract older, higher-income baby boomerswho prefer age-segregation. These commu-nities increase the likelihood of boomersbeing active—i.e. making at least one localrecreational NMT trip—but not thenumber of NMT trips among the active.Physical characteristics have only an indi-rect effect, by influencing the decision tolive in age-restricted settings. In contrast,age-restriction has no effect on being social(i.e. the likelihood of ever visiting neigh-bours); among the social, however, age-restriction increases social trip-making,although perhaps due to self-selection. Inother words, age-restricted neighbourhoodsare associated with higher levels of localsocial activity, but because they attractmore socially inclined residents. The age-restricted effect may stem from a sense ofcommunity fostered in age-restrictedneighbourhoods and/or unobserved orintermingling physical characteristics.
Our analysis indicates the importanceof distinguishing between trip types whencontrolling for self-selection in the builtenvironment/travel behaviour research. Italso suffers from a range of limitations,including generalisability, unknown rela-tive magnitude of effects and inability toassess impacts over time. While thisresearch offers some insight into the
18 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
influence of age-restricted neighbourhoodson baby boomers’ local travel behaviours,it says nothing about the regional travelpatterns of this highly suburbanised,automobile-dependent generation.
Notes
1. In the US, the Department of Housing andUrban Development (HUD) uses seniorhousing, or 55 and older community;residential developer Del Webb refers to‘active adult communities’ (HarrisInteractive, 2005); the National Association ofHomebuilders suggests that ‘age-qualified’ ispreferred (Emrath and Liu, 2007).
2. Overall older adult (and age-restricted)household locations in the US: 23 per cent(14 per cent) central cities; 50 per cent (71per cent) suburbs; and 27 per cent (15 percent) outside metro areas (derived fromEmrath and Liu, 2007, Tables 1 and 2).
3. A sample selection problem may also exist:among the possible sub-samples of babyboomers, factors influencing residentiallocation choice for our age-restrictedsub-sample could also influence behaviour. Inour case, this problem effectively appears as aform of omitted variable bias (Hebbert, 2008).
4. Based on share of census population in2000, accumulated over the correspondingcensus block centroid’s distance fromBoston’s central business district.
5. Heudorfer (2005) inventoried (apparentlybased on a survey of town officials)age-restricted housing in the state, but didnot identify individual developments.
6. Problematic responses included: addressesnon-geo-locatable or outside the studyarea (due to mail forwarding); nohousehold survey page; age outside thecohort of interest. Hebbert (2008) providesdetail on survey design, implementationand results.
7. The grid cell approach ensures anonymity,making it impossible to identify thehousehold’s address. The centroid of the250-metre grid cell serves as the household‘location’. Each grid centroid was visually
associated to a neighbourhood based onprimary street characteristics (see Figure 1).Basic data, including roads, parcels,commuter rail, come from MassGIS (http://www.mass.gov/mgis/), although availabledata were limited. For example, no buildingfootprints for sample neighbourhoods couldbe located and road networks in the newerage-restricted neighbourhoods wereoutdated. We updated missing data using ahigh resolution aerial photo from ESRI(http://www.arcgis.com/home/item.html?id=a5fef63517cd4a099b437e55713d3d54) toclassify street patterns and computeintersection density. For other neighbourhoodcharacteristics (for example, public spaces,outdoor sports facilities), we used GoogleEarth’s satellite imagery and ‘Places’ layer,and, in some cases, site plans andage-restricted neighbourhoods’ webpages.
8. Space constraints preclude including theconfirmatory factor analysis (CFA). Referto the measurement models in Tables A1and A2; full results available upon request.
9. For the NMT and social trip models, theVuong test indicates that ZINB is preferableto a regular negative binomial; and alikelihood ratio test indicates that ZINB ispreferable to a zero-inflated Poisson.
10. As mentioned in the sample description,our sample is endogenously stratified. Inthe age-restricted neighbourhood choice(logit) models, this choice-based samplingresults in an inconsistent alternative specificconstant, while other coefficients areconsistent (Manski and Lerman, 1977). Weattempt to correct for this samplingstrategy by using weights in the choicemodel estimation: p/s for households inage-restricted communities and (1-p)/(1-s) for households from unrestrictedcommunities, where p is the probability ofa household living in an age-restrictedcommunity from the population and s isthe proportion of households from oursample living in an age-restrictedcommunity. As a value for p is notavailable, we use 3.2 per cent (Emrath andLiu, 2007). We test the sensitivity of results
BY COMMUNITY OR DESIGN? 19
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
to this value by estimating the models withp = 1 per cent and 6 per cent (reasonableupper and lower bounds); the results donot vary substantively. Full sensitivityanalyses are available upon request.
11. We sampled households but modelindividuals, thus need to account forpotential correlation of behaviour amongsame-household individuals (i.e. intraclasscorrelation). One option MPlus providesfor dealing with this issue is correctingstandard errors (SEs); using this approach,the SEs in Tables 3 and 4 are ‘corrected’.Our approach may also have intraclasscorrelation among households from thesame neighbourhood, indicating the needfor a multilevel SEM ZINB model. We leavethis approach for future research.
12. The differences may partially result fromundercounting in our survey. The55–65-year-old cohort at the time of themost recent Boston metropolitan areatravel survey (in 1991) had an automobilemode share of 89 per cent (CTPS, 1993);the most recent national travel survey,although with only 194 baby boomers fromthe Boston MA, indicates a 78 per centautomobile mode share for all trips by thiscohort (US DOT, 2009).
13. Residents of age-restricted neighbourhoodsreport having lived there on average for 5years, compared with 19 years forunrestricted neighbourhoods (Hebbert,2008).
Funding Statement
This work was supported by the New EnglandUniversity Transportation Center, under grantDTRS99-G-0001.
Acknowledgements
The authors wish to thank: Frank Hebbert, whoplayed a major role in survey design, implemen-tation and analysis; Joe Coughlin and MosheBen-Akiva, for numerous useful conversations;Kristin Simonson, for assisting with the spatialanalysis; Lee Carpenter, for editorial assistance;and several anonymous referees for detailed andinsightful comments and suggestions.
References
Abreu, J., Golob, T. and Goulias, K. (2006)Effects of land use characteristics on resi-dence and employment location and travelbehavior of urban adult workers, Transporta-tion Research Record, 1977, pp. 121–131.
Ahrentzen, S. (2010) On their own turf: com-munity design and active aging in a naturallyoccurring retirement community, Journal ofHousing for the Elderly, 24(3/4), pp. 267–290.
Alfonzo, M., Boarnet, M., Day, K., Mcmillan, T.et al. (2008) The relationship of neighbour-hood built environment features and adultparents’ walking, Journal of Urban Design,13(1), pp. 29–51.
Bagley, M. N. and Mokhtarian, P. L. (2002) Theimpact of residential neighborhood type ontravel behavior: a structural equations model-ing approach, The Annals of Regional Science,36, pp. 279–297.
Baran, P., Rodrıguez, D. and Khattak, A. (2008)Space syntax and walking in a new urbanistand suburban neighborhoods, Journal ofUrban Design, 13(1), pp. 5–28.
Ben-Akiva, M., Walker, J., Bernardino, A. T., Gopi-nath, D. A. et al. (2002) Integration of choiceand latent variable models, in: H. Mahmassani(Ed.) In Perpetual Motion: Travel BehaviourResearch Opportunities and Application Chal-lenges, ch. 21. Oxford: Elsevier Pergamon.
Berke, E. M., Gottlieb, L. M., Moudon, A. V.and Larson, E. B. (2007a) Protective asso-ciation between neighborhood walkabil-ity and depression in older men, Journalof the American Geriatrics Society, 55, pp.526–533.
Berke, E. M., Koepsell, T. D., Moudon, A. V.,Hoskins, R. E. et al. (2007b) Association ofthe built environment with physical activityand obesity in older persons, American Jour-nal of Public Health, 97(3), pp. 486–492.
Blechman, A. (2008) Leisureville: Adventures inAmerica’s Retirement Utopias. New York:Atlantic Monthly Press.
Cao, X., Mokhtarian, P. and Handy, S. (2009)Examining the impacts of residential self-selection on travel behaviour: a focus onempirical findings, Transport Reviews, 29(3),pp. 359–395.
Cao, X., Handy, S. and Mokhtarian, P. (2006)The influences of the built environment and
20 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
residential self-selection on pedestrian beha-vior: evidence from Austin, TX, Transporta-tion, 33(1), pp. 1–20.
Cao, X., Mokhtarian, P. and Handy, S. (2010)Neighborhood design and the accessibility ofthe elderly: an empirical analysis in northCalifornia, International Journal of Sustain-able Transportation, 4, pp. 347–371.
Champion, A. G. (2001) A changing demo-graphic regime and evolving polycentricurban regions: consequences for the size,composition and distribution of city popula-tions, Urban Studies, 38(4), pp. 657–677.
Chapman, N. J. and Beaudet, M. (1983) Envi-ronmental predictors of well-being for at-risk older adults in a mid-sized city, Journalof Gerontology, 38(2), pp. 237–244.
Chaskin, R. J. (1995) Defining neighborhood: his-tory, theory, and practice. The Chapin HallCenter for Children, University of Chicago.
Coughlin, J. (2007) Disruptive demographics,design, and the future of everyday environ-ments, Design Management Review, 18(2), pp.53–59.
CTPS (Central Transportation Planning Staff)(1993) 1991 household travel survey. BostonMetropolitan Planning Organization, Boston,MA.
Cvitkovich, Y. and Wister, A. (2001) The impor-tance of transportation and prioritization ofenvironmental needs to sustain well-beingamong older adults, Environment and Beha-vior, 33(6), pp. 809–829.
DiPietro, L. (2001) Physical activity in aging:changes in patterns and their relationship tohealth and function, Journal of GerontologyA, 56(2), pp. 13–22.
Emrath, P. and Liu, H. F. (2007) The age-qualifiedactive adult housing market. National Associa-tion of Home Builders, Washington, DC.
Ewing, R. and Handy, S. (2009) Measuring theunmeasurable: urban design qualities relatedto walkability, Journal of Urban Design, 14(1),pp. 65–84.
Eyler, A., Brownson, R., Bacak, S. and House-mann, R. (2003) Epidemiology of walking forphysical activity in the United States, Medi-cine & Science in Sports & Exercise, 35(9), pp.1529–1536.
Flynn, T. E. and Boenau, A. E. (2007) Trip gener-ation characteristics of age-restricted housing,ITE Journal, 77(2), pp. 30–37.
Giles-Corti, B. and Donovan, R. (2002) The relativeinfluence of individual, social and physical envi-ronment determinants of physical activity, SocialScience & Medicine, 54(12), pp. 1793–1812.
Giles-Corti, B., Broomhall, M., Knuiman, M., Col-lins, C. et al. (2005) Increasing walking: howimportant is distance to, attractiveness, and sizeof public open space?, American Journal of Pre-ventive Medicine, 28(2S2), pp. 169–176.
Golob, T. (2003) Structural equation modelingfor travel behavior research, TransportationResearch Part B, 37(1), pp. 1–25.
Grant, J. and Mittelsteadt, L. (2004) Types ofgated communities, Environment and Plan-ning B, 31, pp. 913–930.
Harris Interactive (2005) Pulte Homes BabyBoomer Study. Harris Interactive (http://www.icaa.cc/1-Organizationalmember/mem-ber_research/viewsonaging/Delwebb2005survey.pdf).
Hebbert, F. (2008) Local travel habits of babyboomers in suburban age-restricted commu-nities. MA thesis, Massachusetts Institute ofTechnology, Cambridge, MA.
Heudorfer, B. (2005) Age restricted active adulthousing in Massachusetts, Boston MA. Citi-zens’ Housing and Planning Association, June(http://www.chapa.org/sites/default/files/f_122952789640BUpdateDec2008_4.pdf).
Hunt, M. E. and Gunter-Hunt, G. (1985) Natu-rally occurring retirement communities, Jour-nal of Housing for the Elderly, 3(3/4), pp. 3–22.
Huston, S., Evenson, K., Bors, P. and Gizlice, Z.(2003) Neighborhood environment, access toplaces for activity, and leisure-time physicalactivity in a diverse North Carolina popula-tion, American Journal of Health Promotion,18, pp. 158–169.
Jones, A. J. (2005) Applied econometrics for healtheconomists: a practical guide, 2nd edn. Pre-pared for the Office of Health Economics,London (http://www.york.ac.uk/res/herc/teach/OHE-Booklet.pdf).
Joseph, A. and Zimring, C. (2007) Where activeolder adults walk: understanding the factorsrelated to path choice for walking amongactive retirement community residents, Envi-ronment and Behavior, 39(1), pp. 75–105.
Kennedy, D. J. and Coates, D. (2008) Retire-ment village resident satisfaction in Austra-lia: a qualitative enquiry, Journal of Housingfor the Elderly, 22(4), pp. 311–334.
BY COMMUNITY OR DESIGN? 21
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
King, W., Brach, J., Belle, S., Killingsworth, R.et al. (2003) The relationship between conve-nience of destinations and walking levels inolder women, American Journal of Health Pro-motion, 18(1), pp. 74–82.
Kweon, B.-S., Sullivan, W. and Wiley, A. (1998)Green common spaces and the social integra-tion of inner-city older adults, Environmentand Behavior, 30(6), pp. 832–858.
Lawton, M. P., Nahemow, L. and Teaff, J.(1975) Housing characteristics and the well-being of elderly tenants in federally assistedhousing, Journal of Gerontology, 30(5), pp.601–607.
Lee, C. and Moudon, A. V. (2006) Correlates ofwalking for transportation or recreation pur-poses, Journal of Physical Activity and Health,3(S1), pp. S77–S98.
Maat, K., Wee, B. van and Stead, D. (2005) Landuse and travel behaviour: expected effectsfrom the perspective of utility theory andactivity-based theories, Environment andPlanning B, 32, pp. 33–46.
Manski, C. and Lerman, S. (1977) The estimationof choice probabilities from choice-basedsamples, Econometrica, 45(8), pp. 1977–1988.
Mokhtarian, P. and Cao, X. (2008) Examining theimpacts of residential self-selection on travelbehavior: a focus on methodologies, Transpor-tation Research B, 42(3), pp. 204–228.
Moudon, A. V., Lee, C., Cheadle, A., Garvin, C.et al. (2005) Operational definition of walk-able neighborhood, Journal of Physical Activ-ity and Health, 3(S1), pp. S99–S117.
Muthen, B. O. (1998–2004) Mplus Technical Appen-dices. Los Angeles, CA: Muthen & Muthen.
Muthen, L. K. and Muthen, B. O. (1998–2007)Mplus User’s Guide, 5th edn. Los Angeles:Muthen & Muthen.
Nagel, C., Carlson, N., Bosworth, M. and Michael,Y. (2008) The relation between neighborhoodbuilt environment and walking activity amongolder adults, American Journal of Epidemiology,168(4), pp. 461–468.
Owen, N., Humpel, N., Leslie, E., Bauman, A.et al. (2004) Understanding environmentalinfluences on walking: review and researchagenda, American Journal of Preventive Medi-cine, 27(1), pp. 67–76.
Phillipson, C., Leach, R., Money, A. and Biggs, S.(2008) Social and cultural constructions ofageing: the case of the baby boomers, Sociological
Research Online, 13(3)5 (http://www.socresonline.org.uk/13/3/5.html).
Rodriguez, D., Khattak, A. and Kelly, E. (2006)Can new urbanism encourage physical activity?Comparing a new urbanist neighborhood withconventional suburbs, Journal of the AmericanPlanning Association, 72(1), pp. 43–54.
Rosenbloom, S. (2003) The mobility needs ofolder Americans: implications for transportationreauthorization. Transportation Reform Series,The Brookings Institution, Washington, DC.
Saelens, B. and Sallis, J. (2003) Neighborhood-based differences in physical activity: an envi-ronment scale evaluation, American Journalof Public Health, 93(9), pp. 1552–1558.
Sallis, J. and Owen, N. (1996) Ecological models,in: K. Glanz, B. K. Rimer and F. M. Lewis(Eds) Health Behavior and Health Education:Theory, Research, and Practice, 2nd edn, pp.403–424. San Francisco, CA: Jossey-Bass.
Schmocker, J.-D., Quddus, M. A., Noland, R. B.and Bell, M. G. H. (2005) Estimating trip gen-eration of elderly and disabled people: analy-sis of London data, Transportation ResearchRecord, no. 1924, pp. 9–18.
US Census Bureau (2002) Census 2000 summaryfile 2 - geographic area: Boston–Worcester–Lawrence, MA–NH–ME–CT CMSA (http://library.case.edu/media/kelvinsmithlibrary/census/sf2techdoc.pdf).
US DOT (US Department of Transportation,Federal Highway Administration) (2009)2009 national household travel survey (http://nhts.ornl.gov).
Wachs, M. (1979) Transportation for the Elderly:Changing Lifestyles, Changing Needs. Berkeley,CA: University of California Press.
Wellman, B. (2005) Community: from neigh-borhood to network, Communications of theACM, 48(10), pp. 53–55.
Wister, A. (2005) Baby Boomer Health Dynamics:How Are We Aging? Toronto, ON: Universityof Toronto Press.
Wretstrand, A., Svensson, H., Fristedt, S. andFalkmer, T. (2009) Older people and localpublic transit: mobility effects of accessibilityimprovements in Sweden, Journal of Trans-port and Land Use, 2(2), pp. 49–65.
Yang, H.-Y. and Stark, S. L. (2010) The role of envi-ronmental features in social engagement amongresidents living in assisted living facilities, Journalof Housing for the Elderly, 24(1), pp. 28–43.
22 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
1.
Rec
reat
ional
NM
Ttr
ips
zero
-inflat
edneg
ativ
ebin
om
ial(Z
INB
)m
odel
resu
lts
(N=
1456)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Log
itm
odel
(zer
o-in
flat
ion
)es
tim
atin
g‘l
ikel
ihoo
dof
bein
gin
non
-act
ive
grou
p’A
ge-r
estr
icte
d2
0.51
9*(0
.244
)2
0.60
2*(0
.249
)2
0.61
0*(0
.250
)P
roW
alka
bili
ty2
0.21
2*(0
.098
)2
0.21
3*(0
.097
)P
roSe
greg
atio
n0.
210*
(0.0
95)
0.20
8*(0
.094
)G
rid
20.
172
(0.1
82)
20.
188
(0.1
81)
20.
188
(0.1
81)
Loo
p0.
059
(0.1
79)
0.05
5(0
.181
)0.
056
(0.1
81)
Inte
rsec
tion
den
sity
20.
296
(0.4
22)
20.
211
(0.4
13)
20.
210
(0.4
13)
Fac
ilit
ies
20.
027
(0.1
65)
20.
010
(0.1
66)
20.
009
(0.1
66)
Des
tin
atio
n40
00.
017
(0.1
60)
0.06
0(0
.161
)0.
060
(0.1
61)
MB
TA
bus
stop
20.
166
(0.2
83)
20.
092
(0.2
84)
20.
093
(0.2
84)
Com
mu
ter
rail
20.
053
(0.1
80)
20.
041
(0.1
84)
20.
041
(0.1
84)
Stre
etle
ngt
h0.
004
(0.0
09)
0.00
2(0
.009
)0.
002
(0.0
09)
Em
ploy
0.57
8**
(0.1
66)
0.60
6**
(0.1
67)
0.60
6**
(0.1
67)
Hea
lth
y2
0.74
3**
(0.1
96)
20.
788**
(0.1
96)
20.
788**
(0.1
96)
Mal
e0.
027
(0.1
34)
20.
048
(0.1
37)
20.
048
(0.1
37)
Age
0.01
1(0
.020
)0.
012
(0.0
20)
0.01
2(0
.020
)H
igh
inco
me
20.
129
(0.2
05)
20.
102
(0.2
07)
20.
102
(0.2
07)
Mid
inco
me
0.05
3(0
.190
)0.
067
(0.1
90)
0.06
6(0
.190
)T
hre
eve
hic
les
20.
096
(0.1
64)
20.
165
(0.1
67)
20.
165
(0.1
67)
Bik
e2
0.44
9**
(0.1
46)
20.
430**
(0.1
48)
20.
430**
(0.1
48)
Con
stan
t2
0.46
6(1
.295
)2
0.50
9(1
.311
)2
0.51
1(1
.311
)
Neg
ativ
ebi
nom
ial
mod
eles
tim
atin
g‘n
um
ber
ofN
MT
trip
sam
ong
acti
vegr
oup’
Age
-res
tric
ted
0.05
3(0
.087
)0.
078
(0.0
87)
0.08
0(0
.088
)P
roW
alka
bili
ty2
0.03
2(0
.038
)2
0.03
1(0
.038
)P
roSe
greg
atio
n2
0.03
7(0
.036
)2
0.03
7(0
.036
)G
rid
0.00
1(0
.069
)0.
007
(0.0
69)
0.00
7(0
.069
)
(con
tin
ued
)
Appen
dix
BY COMMUNITY OR DESIGN? 23
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
1.(C
onti
nued
)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Loo
p2
0.03
4(0
.068
)2
0.02
9(0
.068
)2
0.02
9(0
.068
)In
ters
ecti
ond
ensi
ty2
0.05
0(0
.115
)2
0.05
2(0
.113
)2
0.05
2(0
.113
)F
acil
itie
s0.
000
(0.0
66)
0.01
0(0
.066
)0.
010
(0.0
66)
Des
tin
atio
n40
00.
128
(0.0
60)
0.13
9*(0
.060
)0.
139*
(0.0
60)
MB
TA
Bu
sst
op0.
101
(0.1
08)
0.09
9(0
.105
)0.
099
(0.1
05)
Com
mu
ter
rail
20.
163*
(0.0
68)
20.
177*
(0.0
68)
20.
177*
(0.0
68)
Stre
etle
ngt
h0.
001
(0.0
03)
0.00
0(0
.003
)0.
000
(0.0
03)
Em
ploy
20.
227**
(0.0
55)
20.
234**
(0.0
56)
20.
234**
(0.0
56)
Hea
lth
y0.
065
(0.0
78)
0.06
3(0
.078
)0.
063
(0.0
78)
Mal
e2
0.02
0(0
.049
)2
0.02
2(0
.049
)2
0.02
2(0
.049
)A
ge2
0.00
4(0
.007
)2
0.00
3(0
.007
)2
0.00
3(0
.007
)H
igh
inco
me
20.
015
(0.0
74)
20.
014
(0.0
74)
20.
014
(0.0
74)
Mid
inco
me
20.
035
(0.0
71)
20.
041
(0.0
71)
20.
041
(0.0
71)
Th
ree
veh
icle
s2
0.03
8(0
.063
)2
0.04
6(0
.062
)2
0.04
6(0
.062
)B
ike
0.03
4(0
.055
)0.
030
(0.0
55)
0.03
0(0
.055
)C
onst
ant
1.56
9**
(0.4
22)
1.55
2**
(0.4
22)
1.55
2**
(0.4
22)
Alp
ha
0.08
9**
(0.0
24)
0.08
8**
(0.0
24)
0.08
8**
(0.0
24)
Mea
sure
men
tm
odel
esti
mat
ing
Pro
Wal
kabi
lity
I 1:
Ip
refe
rto
hav
esh
op
san
dse
rvic
esw
ith
inw
alki
ng
dis
tan
ce1.
000
(0.0
00)
1.00
0(0
.000
)I 2
:I
do
no
tva
lue
spac
ear
ou
nd
my
ho
use
mo
reth
ansh
op
sn
earb
y0.
889**
(0.0
94)
0.88
8**
(0.0
94)
I 3:
Ili
kea
nei
ghb
ou
rho
od
con
tain
ing
ho
usi
ng,
sho
ps
and
serv
ices
0.92
7**
(0.0
46)
0.92
6**
(0.0
46)
I 4:
Id
on
ot
pre
fer
alo
to
fsp
ace
bet
wee
nm
yh
om
ean
dth
est
reet
0.51
5**
(0.0
80)
0.51
5**
(0.0
80)
I 5:
Ip
refe
ra
ho
use
clo
seto
the
sid
ewal
kso
that
Ica
nse
ep
asse
rsb
y0.
621**
(0.0
50)
0.62
0**
(0.0
50)
(con
tin
ued
)
24 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
1.(C
onti
nued
)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Pro
Segr
egat
ion
I 6:
Ip
refe
rn
eigh
bo
urs
atth
esa
me
stag
eo
fli
feas
me
1.00
0(0
.000
)1.
000
(0.0
00)
I 7:
Ip
refe
rli
vin
gar
ou
nd
peo
ple
wh
oar
esi
mil
arto
me
0.58
9**
(0.0
82)
0.58
6**
(0.0
81)
I 8:
Iam
con
cern
edab
ou
tst
ran
gers
wal
kin
gth
rou
ghm
yn
eigh
bo
urh
oo
d0.
462**
(0.0
87)
0.45
7**
(0.0
85)
I 9:
Ili
keto
live
ina
nei
ghb
ou
rho
od
wit
ho
ut
chil
dre
nin
it0.
354**
(0.0
61)
0.35
3**
(0.0
60)
Stru
ctu
ral
(MIM
IC)
mod
eles
tim
atin
gP
roW
alka
bili
tyE
mpl
oy2
0.01
0(0
.067
)2
0.01
1(0
.068
)H
ealt
hy
20.
024
(0.0
83)
20.
024
(0.0
84)
Mal
e2
0.22
6**
(0.0
53)
20.
227**
(0.0
53)
Age
0.00
4(0
.008
)0.
003
(0.0
08)
Hig
hin
com
e0.
044
(0.0
91)
0.04
4(0
.091
)M
idin
com
e2
0.09
4(0
.080
)2
0.09
4(0
.080
)T
hre
eve
hic
les
20.
301**
(0.0
76)
20.
301**
(0.0
76)
Bik
e2
0.04
6(0
.068
)2
0.04
6(0
.068
)
Pro
Segr
egat
ion
Em
ploy
20.
167*
(0.0
65)
20.
168*
(0.0
65)
Hea
lth
y0.
124
(0.0
81)
0.12
5(0
.080
)M
ale
0.12
6*(0
.049
)0.
126*
(0.0
49)
Age
20.
001
(0.0
08)
20.
001
(0.0
08)
Hig
hin
com
e2
0.03
0(0
.091
)2
0.03
1(0
.091
)M
idin
com
e2
0.07
7(0
.079
)2
0.07
7(0
.079
)T
hre
eve
hic
les
0.02
9(0
.070
)0.
029
(0.0
70)
Bik
e2
0.12
8*(0
.065
)2
0.12
8*(0
.065
)
Log
itm
odel
esti
mat
ing
‘lik
elih
ood
ofch
oosi
ng
age-
rest
rict
edn
eigh
bou
rhoo
d’
Pro
Wal
kabi
lity
0.14
0(0
.179
)P
roSe
greg
atio
n0.
810**
(0.1
47)
(con
tin
ued
)
BY COMMUNITY OR DESIGN? 25
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
1.(C
onti
nued
)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Loo
p5.
979**
(0.7
08)
Inte
rsec
tion
Den
sity
2.20
6**
(0.3
84)
Fac
ilit
ies
3.39
3**
(0.3
41)
Des
tin
atio
n40
02
0.11
6(0
.299
)M
BT
Abu
sst
op2
2.64
7**
(0.9
40)
Com
mu
ter
rail
0.19
9(0
.433
)St
reet
len
gth
0.10
6**
(0.0
14)
Em
ploy
20.
259
(0.2
70)
Hea
lth
y0.
530
(0.3
40)
Mal
e2
0.44
2*(0
.192
)A
ge0.
205**
(0.0
37)
Hig
hin
com
e1.
302**
(0.3
97)
Mid
inco
me
0.47
0(0
.329
)T
hre
eve
hic
les
20.
268
(0.4
37)
Bik
e2
0.32
3(0
.267
)
Nu
mb
ero
fp
aram
eter
s37
8410
2N
um
ber
of
ob
serv
edva
riab
les
1827
27Id
enti
fica
tio
nO
veri
den
tifi
ed84
\0.
5*27*
(27+
1)O
veri
den
tifi
ed10
2\
0.5*
27*
(27+
1)
Not
es:
yp
\0.
10;*
p\
0.05
;**
p\
0.01
.
26 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
2.
Soci
altr
ips
zero
-infl
ated
neg
ativ
ebin
om
ial(Z
INB
)m
odel
resu
lts
(N=
1410)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Log
itm
odel
(zer
o-in
flat
ion
)es
tim
atin
g‘l
ikel
ihoo
dof
bein
gin
non
-act
ive
grou
p’A
ge-r
estr
icte
d0.
030
(0.5
12)
20.
102
(0.6
11)
20.
111
(0.6
21)
Pro
Wal
kabi
lity
20.
240
(0.2
97)
20.
238
(0.2
97)
Pro
Segr
egat
ion
0.19
2(0
.283
)0.
194
(0.2
81)
Gri
d0.
543
(0.5
87)
0.40
0(0
.528
)0.
399
(0.5
26)
Loo
p0.
147
(0.5
47)
0.08
7(0
.453
)0.
089
(0.4
53)
Inte
rsec
tion
den
sity
21.
379
(1.3
37)
21.
175
(1.0
62)
21.
175
(1.0
62)
Fac
ilit
ies
20.
486
(0.5
18)
20.
360
(0.4
30)
20.
360
(0.4
30)
Des
tin
atio
n40
02
0.39
5(0
.447
)2
0.21
7(0
.452
)2
0.21
7(0
.451
)M
BT
Abu
sst
op2
1.40
4(2
.151
)2
1.06
6(2
.118
)2
1.06
4(2
.109
)C
omm
ute
rra
il1.
200y
(0.6
19)
1.10
8y(0
.600
)1.
110y
(0.6
03)
Stre
etle
ngt
h2
0.00
8(0
.017
)2
0.01
3(0
.016
)2
0.01
3(0
.016
)E
mpl
oy0.
525
(0.3
95)
0.56
0(0
.360
)0.
560
(0.3
60)
Hea
lth
y0.
370
(0.5
73)
0.37
4(0
.595
)0.
374
(0.5
95)
Mal
e0.
819*
(0.3
70)
0.74
8y(0
.421
)0.
749y
(0.4
22)
Age
0.05
1(0
.048
)0.
040
(0.0
45)
0.04
0(0
.045
)H
igh
inco
me
0.40
0(0
.756
)0.
382
(0.6
82)
0.38
4(0
.683
)M
idin
com
e0.
452
(0.6
00)
0.45
8(0
.602
)0.
460
(0.6
05)
Th
ree
veh
icle
s2
0.36
0(0
.649
)2
0.40
2(0
.675
)2
0.40
4(0
.677
)B
ike
0.07
4(0
.321
)0.
048
(0.3
11)
0.04
9(0
.312
)C
onst
ant
24.
486
(3.5
44)
23.
778
(3.5
13)
23.
776
(3.5
19)
Neg
ativ
ebi
nom
ial
mod
eles
tim
atin
g‘n
um
ber
ofso
cial
trip
sam
ong
soci
algr
oup’
Age
-res
tric
ted
0.57
6**
(0.2
12)
0.47
9*(0
.239
)0.
471y
(0.2
42)
Pro
Wal
kabi
lity
20.
018
(0.1
28)
20.
017
(0.1
28)
Pro
Segr
egat
ion
0.15
6(0
.111
)0.
156
(0.1
10)
Gri
d0.
115
(0.1
96)
0.04
6(0
.202
)0.
046
(0.2
01)
(con
tin
ued
)
BY COMMUNITY OR DESIGN? 27
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
2.(C
onti
nued
)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Loo
p2
0.15
7(0
.200
)2
0.18
2(0
.184
)2
0.18
0(0
.184
)In
ters
ecti
ond
ensi
ty2
0.26
4(0
.303
)2
0.24
4(0
.273
)2
0.24
2(0
.273
)F
acil
itie
s2
0.18
8(0
.186
)2
0.16
6(0
.173
)2
0.16
5(0
.173
)D
esti
nat
ion
400
0.14
2(0
.173
)0.
196
(0.1
95)
0.19
6(0
.195
)M
BT
Abu
sst
op2
0.65
9(0
.481
)2
0.60
3(0
.595
)2
0.60
3(0
.593
)C
omm
ute
rra
il0.
342*
(0.1
67)
0.35
2y(0
.182
)0.
353y
(0.1
82)
Stre
etle
ngt
h2
0.00
8(0
.007
)2
0.00
9(0
.006
)2
0.00
9(0
.006
)E
mpl
oy2
0.36
8*(0
.168
)2
0.32
7*(0
.151
)2
0.32
7*(0
.151
)H
ealt
hy
20.
006
(0.1
87)
20.
006
(0.1
89)
20.
006
(0.1
89)
Mal
e0.
144
(0.1
54)
0.14
4(0
.153
)0.
145
(0.1
52)
Age
0.04
8*(0
.020
)0.
043*
(0.0
20)
0.04
3*(0
.020
)H
igh
inco
me
20.
262
(0.2
68)
20.
259
(0.2
53)
20.
258
(0.2
54)
Mid
inco
me
20.
155
(0.1
98)
20.
126
(0.2
07)
20.
125
(0.2
07)
Th
ree
veh
icle
s2
0.18
5(0
.244
)2
0.19
4(0
.261
)2
0.19
5(0
.261
)B
ike
0.26
2*(0
.133
)0.
261*
(0.1
31)
0.26
2*(0
.131
)C
onst
ant
22.
518y
(1.2
85)
22.
243y
(1.3
19)
22.
241y
(1.3
20)
Alp
ha
0.36
0y(0
.215
)0.
301
(0.2
00)
0.30
1(0
.201
)M
easu
rem
ent
mod
eles
tim
atin
gP
roW
alka
bili
tyI 1
:I
pre
fer
toh
ave
sho
ps
and
serv
ices
wit
hin
wal
kin
gd
ista
nce
1.00
0(0
.000
)1.
000
(0.0
00)
I 2:
Id
on
ot
valu
esp
ace
aro
un
dm
yh
ou
sem
ore
than
sho
ps
nea
rby
0.89
3**
(0.0
93)
0.89
3**
(0.0
93)
I 3:
Ili
kea
nei
ghb
ou
rho
od
con
tain
ing
ho
usi
ng,
sho
ps
and
serv
ices
0.92
9**
(0.0
45)
0.92
9**
(0.0
45)
I 4:
Id
on
ot
pre
fer
alo
to
fsp
ace
bet
wee
nm
yh
om
ean
dth
est
reet
0.52
0**
(0.0
80)
0.52
0**
(0.0
80)
(con
tin
ued
)
28 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
2.(C
onti
nued
)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
I 5:
Ip
refe
ra
ho
use
clo
seto
the
sid
ewal
kso
that
Ica
nse
ep
asse
rsb
y0.
625**
(0.0
51)
0.62
5**
(0.0
51)
Pro
Segr
egat
ion
I 6:
Ip
refe
rn
eigh
bo
urs
atth
esa
me
stag
eo
fli
feas
me
1.00
0(0
.000
)1.
000
(0.0
00)
I 7:
Ip
refe
rli
vin
gar
ou
nd
peo
ple
wh
oar
esi
mil
arto
me
0.59
3**
(0.0
85)
0.58
8**
(0.0
84)
I 8:
Iam
con
cern
edab
ou
tst
ran
gers
wal
kin
gth
rou
ghm
yn
eigh
bo
urh
oo
d0.
465**
(0.0
90)
0.45
9**
(0.0
88)
I 9:
Ili
keto
live
ina
nei
ghb
ou
rho
od
wit
ho
ut
chil
dre
nin
it0.
354**
(0.0
60)
0.35
2**
(0.0
60)
Stru
ctu
ral
(MIM
IC)
mod
eles
tim
atin
gP
roW
alka
bili
tyE
mpl
oy2
0.01
0(0
.067
)2
0.01
0(0
.067
)H
ealt
hy
20.
024
(0.0
83)
20.
024
(0.0
83)
Mal
e2
0.22
6**
(0.0
53)
20.
225**
(0.0
53)
Age
0.00
4(0
.008
)0.
003
(0.0
08)
Hig
hin
com
e0.
044
(0.0
91)
0.04
4(0
.091
)M
idin
com
e2
0.09
3(0
.079
)2
0.09
3(0
.079
)T
hre
eve
hic
les
20.
300**
(0.0
76)
20.
300**
(0.0
76)
Bik
e2
0.04
6(0
.067
)2
0.04
6(0
.067
)
Pro
Segr
egat
ion
Em
ploy
20.
167*
(0.0
65)
20.
168*
(0.0
65)
Hea
lth
y0.
124
(0.0
81)
0.12
4(0
.081
)M
ale
0.12
6*(0
.049
)0.
126*
(0.0
49)
Age
20.
001
(0.0
08)
20.
001
(0.0
08)
Hig
hin
com
e2
0.03
1(0
.091
)2
0.03
1(0
.092
)M
idin
com
e2
0.07
8(0
.079
)2
0.07
8(0
.079
)T
hre
eve
hic
les
0.02
9(0
.070
)0.
029
(0.0
70)
Bik
e2
0.12
8*(0
.064
)2
0.12
8*(0
.065
)
(con
tin
ued
)
BY COMMUNITY OR DESIGN? 29
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from
Tab
leA
2.(C
onti
nued
)
Mod
el1:
ZIN
Bw
ith
out
late
nt
vari
able
sM
odel
2:SE
Mw
ith
late
nt
vari
able
sM
odel
3:SE
Mw
ith
age-
rest
rict
edch
oice
mod
el
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Coe
ffic
ien
t(S
.E.)
Log
itM
odel
Est
imat
ing
‘Lik
elih
ood
ofch
oosi
ng
age-
rest
rict
edn
eigh
bou
rhoo
d’
Pro
Wal
kabi
lity
0.13
8(0
.180
)P
roSe
greg
atio
n0.
815**
(0.1
48)
Loo
p5.
982**
(0.7
08)
Inte
rsec
tion
den
sity
2.20
5**
(0.3
84)
Fac
ilit
ies
3.39
7**
(0.3
41)
Des
tin
atio
n40
0-0
.112
(0.3
00)
MB
TA
bus
stop
22.
641**
(0.9
40)
Com
mu
ter
rail
0.19
6(0
.432
)St
reet
len
gth
0.10
6**
(0.0
14)
Em
ploy
-0.2
61(0
.270
)H
ealt
hy
0.52
9(0
.340
)M
ale
20.
445*
(0.1
92)
Age
0.20
5**
(0.0
37)
Hig
hin
com
e1.
303**
(0.3
97)
Mid
inco
me
0.46
7(0
.329
)T
hre
eve
hic
les
20.
266
(0.4
36)
Bik
e2
0.32
1(0
.267
)
Nu
mb
ero
fp
aram
eter
s37
8410
2N
um
ber
of
ob
serv
edva
riab
les
1827
27Id
enti
fica
tio
nO
veri
den
tifi
ed84
\0.
5*27*
(27+
1)O
veri
den
tifi
ed10
2\
0.5*
27*
(27+
1)
Not
es:
yp\
0.10
;*
p\0.
05;**
p\0.
01.
30 CHRISTOPHER ZEGRAS ET AL.
at MASSACHUSETTS INST OF TECH on February 28, 2012usj.sagepub.comDownloaded from