Examining The Impacts of Neighborhood Design and Residential Self Selection On Active Travel A Methodological Assessment

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Urban Geography

ISSN: 0272-3638 (Print) 1938-2847 (Online) Journal homepage: https://www.tandfonline.com/loi/rurb20

Examining the impacts of neighborhood design


and residential self-selection on active travel: a
methodological assessment

Xinyu (Jason) Cao

To cite this article: Xinyu (Jason) Cao (2015) Examining the impacts of neighborhood design and
residential self-selection on active travel: a methodological assessment, Urban Geography, 36:2,
236-255, DOI: 10.1080/02723638.2014.956420

To link to this article: https://doi.org/10.1080/02723638.2014.956420

Published online: 24 Oct 2014.

Submit your article to this journal

Article views: 618

View related articles

View Crossmark data

Citing articles: 27 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=rurb20
Urban Geography, 2015
Vol. 36, No. 2, 236–255, http://dx.doi.org/10.1080/02723638.2014.956420

Examining the impacts of neighborhood design and residential


self-selection on active travel: a methodological assessment
Xinyu (Jason) Cao*

Humphrey School of Public Affairs, University of Minnesota, 301 19th Ave S Minneapolis, MN
55455, USA
(Received 4 April 2013; accepted 2 May 2014)

A limited number of studies have addressed residential self-selection when considering


the relationship between the built environment (BE) and physical activity. Ignoring
self-selection may bias the estimate of the strength of the relationship and misinform
public policy. Using 2011 data from the Twin Cities, this study employs cross-sectional
and quasi-longitudinal analyses to disentangle the impacts between self-selection
(through residential preferences and travel attitudes) and neighborhood characteristics
on active travel (AT). The two approaches produce somewhat different outcomes:
attitudinal factors are extensively present in the cross-sectional analysis and the
associations between neighborhood characteristics and AT are greatly attenuated after
controlling for attitudinal factors. In contrast, neighborhood characteristics play a
dominant role in the quasi-longitudinal analysis, while attitudinal and demographic
characteristics carry additional impacts on AT. Overall, both preferences and attitudes
help explain the difference in AT. I find that while physical activity infrastructure
impacts biking, physical activity infrastructure together with transit and accessibility
affects walking, even after controlling for demographic and attitudinal characteristics.
The results highlight the importance of shaping the BE in order to provide an
infrastructure of alternative modes of transport and access to different land uses in
policy efforts to increase AT.
Keywords: land use; residential self-selection; Smart Growth; physical activity; travel

The prevalence of obesity, an epidemic in the United States, is associated with the
growing risk of many health issues such as hypertension and type 2 diabetes (Finer,
2006). Obesity results from excessive energy intake and/or physical inactivity. In the
circles of planning and health, conventional land use and transportation policies (such as
zoning regulations, parking standards, and highway expansion) are criticized for their
contribution to physical inactivity (Brownson, Boehmer, & Luke, 2004; Handy, Boarnet,
Ewing, & Killingsworth, 2002; Wells, Evans, & Yang, 2010). The conceptual logic is that
these policies and processes facilitate low-density and auto-oriented development, leading
people to drive more and walk or bike less.
Built environment (BE) interventions have become popular to facilitate active travel
(AT), that is, means of travel that has a focus on physical activity such as walking and
biking. The interventions aim to change transportation systems and urban design through
alternative land use and transportation policies. For instance, Safe Route to School (SR2S)
has been widely implemented to promote children’s AT to school. In 2010, the Minnesota
Governor signed the Complete Streets legislation into law to develop road networks

*Email: cao@umn.edu

© 2014 Taylor & Francis


Urban Geography 237

designed and operated for all modes of transportation: transit riders, bicyclists, and
pedestrians, along with the needs of motorists.
Previous studies have investigated the relationships between the BE and physical
activity (including AT) to justify BE interventions. The interest is so intensive that dozens
of literature reviews have been produced to synthesize empirical findings and guide future
development since 1990 (Ding & Gebel, 2012). Among them, Durand and colleagues
(Durand, Andalib, Dunton, Wolch, & Pentz, 2011) reviewed physical activity outcomes
associated with BE elements along the 10 Principles of Smart Growth. They concluded
that a vast majority of studies found few associations between the elements and physical
activity outcomes (excluding walking), whereas at least half of the studies found sig-
nificant associations between walking and the elements related to the following four
Smart Growth Principles: “create a range of housing opportunities and choices,” “mix
land uses,” “strengthen and direct development towards existing communities,” and “take
advantage of compact building design.” McCormack and Shiell (2011) also concluded
that the BE is more likely to be associated with walking for transport than other forms of
physical activity. Overall, density, diversity, design, accessibility, and safety are common
BE correlates of AT (Saelens & Handy, 2008).
The association between the BE and AT is likely to be confounded by residential self-
selection, which may result from demographic and attitudinal characteristics (Mokhtarian
& Cao, 2008). In this context, self-selection means that people choose where to live based
on their predispositions toward AT. For example, an individual who prefers walking may
consciously live in a pedestrian-friendly neighborhood and hence walk more. In this case,
it may not be the neighborhood but the preference that causes walking. If the self-selection
effect exists but is not controlled for through research design and/or econometric models,
we may misestimate the impact of the BE on AT (Cao, Mokhtarian, & Handy, 2009;
McCormack & Shiell, 2011; Stevens & Brown, 2011), leading to erroneous policy
implications. The consequences of flawed BE interventions are often long-standing
because once implemented, buildings and transportation infrastructure last for decades.
“Sorting out causality by excluding alternative explanations” in future research has
been repeatedly recommended in review studies (Ding & Gebel, 2012, p. 104). However,
the effort has so far been rather limited (Boone-Heinonen, Gordon-Larsen, Guilkey,
Jacobs, & Popkin, 2011; McCormack & Shiell, 2011). Among thousands of studies on
the connections between the BE and physical activity, McCormack and Shiell (2011)
found only 20 cross-sectional and 13 quasi-experimental studies published between 1996
and 2010. They concluded that after controlling for self-selection, previous studies
produce mixed outcomes on the impacts of most BE measures on physical activity, except
for neighborhood type, land use mix, and measures of walkability. This study enriches the
literature by offering additional insights into the connections between neighborhood
characteristics and AT in the Minneapolis—St. Paul metropolitan area, Twin Cities.
Few studies have examined the comparability of different methodologies used in
empirical research. Generally, cross-sectional studies tend to produce relatively consistent
findings: the BE does affect physical activity after controlling for self-selection. However,
longitudinal studies offer mixed outcomes: the BE may or may not influence physical
activity (McCormack & Shiell, 2011). Because a robust longitudinal design is still “in its
infancy” (McCormack & Shiell, 2011, p. 8), McCormack and Shiell call for further
longitudinal research. Moreover, few studies have explored whether econometric treat-
ments of cross-sectional data yield similar results to longitudinal analyses of the same
respondents. This understanding is critical since planners heavily rely on cross-sectional
238 X. (Jason) Cao

studies to inform policies. This study attempts to fill the gap by comparing a cross-
sectional analysis and a quasi-longitudinal analysis.
Another aim of this study was to disentangle the influences of both neighborhood
characteristics and residential self-selection on AT. The cross-sectional statistical control
approach and quasi-longitudinal analysis are applied to 2011 data from the Twin Cities.
The following questions will be answered: (1) Is the difference in AT attributable to a self-
selection effect? (2) Do neighborhood characteristics affect AT after controlling for
demographic and attitudinal characteristics? (3) What neighborhood characteristics play
a more important role in encouraging AT than others? (4) Do the two approaches produce
consistent results? The “Methodology” section describes the complex roles of attitudes in
the relationships between the BE and AT. It presents my analysis approaches, data, and
variables. The next section discusses results. The final section summarizes the key
findings and limitations of this study.

Methodology
To infer causality, scientific research generally requires at least three prerequisites:
statistical association, nonspuriousness, and time precedence (Singleton & Straits,
2005). Since quantitative empirical studies always present statistically significant (or
insignificant) results, nonspuriousness and time precedence are the focus of causal
inference. A spurious relationship between variables refers to an association that can be
explained by a third-party (antecedent) variable. To establish nonspuriousness in a non-
experimental study, an appropriate method is to show that the relationship still holds when
all relevant third-party variables are controlled for (i.e., statistical control). In reality, we
are seldom able to control for all third-party variables (Singleton & Straits, 2005). Time
precedence means that a cause must precede its effect in time, or at least the direction of
influence must be from a cause to an effect (Singleton & Straits, 2005). A panel study is
desirable for establishing time precedence. From a cross-sectional analysis, however, it
can be difficult to tell whether the choice of the BE precedes travel choice or travel choice
precedes residential choice. For example, highly walkable neighborhoods are significantly
associated with a greater prevalence of pedestrian travel (Cervero & Duncan, 2003). A
common inference is that the influence is from the BE to AT through an intervening
variable—such as travel costs. This is a strong causal mechanism from the perspective of
transportation economics (Boarnet & Crane, 2001). Alternatively, it may mean that
individuals who walk a lot intentionally choose to live in highly walkable neighborhoods
to match their predispositions toward AT. This brings about the confounding effects of
attitudes.

The roles of attitudes


Almost all recent studies on the BE and travel behavior/physical activity have controlled
for demographic characteristics. However, few studies have measured attitudinal vari-
ables. Thus, the roles of attitudes in the associations between the BE and AT are still
exploratory because of limited previous research. As shown in Figure 1, attitudes may act
as either antecedent or intervening factors (Cao, Mokhtarian, & Handy, 2011; Mokhtarian
& Cao, 2008). Figure 1(a) illustrates a potentially spurious relationship between pedes-
trian-friendly environments and walking behavior, which can be addressed by controlling
for walking preference. In Figure 1(b), a large amount of walking (which may or may not
have very much to do with pedestrian-friendly environments) may establish or strengthen
Urban Geography 239

(a) Attitudes are antecedent (b) Attitudes are intervening


(in one direction)

Select a pedestrian-friendly Select a pedestrian-friendly


neighborhood neighborhood

Form or reinforce a Form or reinforce a


preference for walking preference for walking

Walk frequently Walk frequently

(c) Attitudes are intervening (d) Attitudes are secondary/ irrelevant


(in the other direction)

Select a pedestrian-friendly Select a pedestrian-friendly


neighborhood neighborhood

Form or reinforce a Form or reinforce a


preference for walking preference for walking

Walk frequently Walk frequently

Causality Association

Figure 1. Some potential relationships among attitudes, built environment, and walking.

an individual’s preference for pedestrian travel, which may in turn encourage her choice
of pedestrian-friendly environments. However, an individual’s current and short-term
travel behavior is not a logical predictor of her previous residential choice. Therefore,
in a cross-sectional analysis, the direction of influence from the (previously chosen) BE to
(presently chosen) AT is generally taken to be the case rather than the inverse inference:
presuming that causality runs from AT to the BE.
Attitudes may again serve as an intervening variable but in the other direction, as
shown in Figure 1(c). In particular, if attitudes are measured at the current time, these
attitudes may be more a function of prior residential choice than the reverse (Chatman,
2009). Alternatively, as shown in Figure 1(d), the BE may have a primary and direct
influence on AT while attitudes may be secondary or irrelevant to this link, as most
previous studies have implicitly or explicitly assumed. For example, one may walk to
many nearby activities, even if reluctantly (counter to preferences), if the BE makes it too
difficult or expensive to drive; conversely, one may drive to many nearby activities, even
if reluctantly, if the BE is not conducive to walking—heavy, fast, noisy, smelly traffic; no
or broken sidewalks; no aesthetic appeal; etc.
240 X. (Jason) Cao

Analysis approaches
Various approaches have been used for causal inference from the BE to travel behavior
(Mokhtarian & Cao, 2008). This study compares the statistical control approach and
quasi-longitudinal design.
Statistical control is the most commonly used approach to address residential self-
selection in cross-sectional data (Boone-Heinonen et al., 2011; Cao et al., 2009;
McCormack & Shiell, 2011). It assumes a unidirectional influence: attitudes affect
residential and travel choices. In the literature, AT is often expressed as a function of
demographics and BE characteristics. Because most studies control for demographics,
they are dropped in the following illustrations for simplicity. If travel attitudes and
residential preferences (AP) are not measured in the data, their impacts will be captured
by the error term of Equation (1).

AT ¼ f ðBEÞ þ eðAPÞ (1)

Because residential choice and hence the BE are influenced by attitudes, BE elements
and the error term are correlated through attitudes, as shown in Equation (2) (Mokhtarian
& Cao, 2008).

AT ¼ f ðBEðAPÞÞ þ eðAPÞ (2)

The correlation violates the basic assumption of regression, and endogeneity bias
becomes a concern. Statistical control explicitly accounts for the influence of attitudes in
analyzing AT, by measuring them and including them in the equation. If all of the attitudes
that confound the BE–AT causal linkage can be captured in measureable variables, then
the BE is presumably uncorrelated to the error term of Equation (3).

AT ¼ f ðBE; APÞ þ ε (3)

If the inclusion of attitudes and preferences makes BE elements insignificant, a natural


conclusion is that BE effect on AT is due to self-selection. Otherwise, the BE has its own
influence (Mokhtarian & Cao, 2008).
Statistical control helps to eliminate spurious associations. However, it assumes that
attitudes are antecedent to both the choices of the BE and AT (Figure 1(a)). In reality, the
relationships between attitudes and the BE and those between attitudes and AT may be
bidirectional (Figure 1(b) and (c)). If attitudes are adapted from the BE, the statistical
control approach may understate the BE effect on AT.
In contrast, a longitudinal design offers a more robust causal inference than does a
cross-sectional design, because it can specify the temporal sequence between change in
BE and change in AT, and controls for time-invariant third-party variables (Boone-
Heinonen et al., 2011; McCormack & Shiell, 2011; Mokhtarian & Cao, 2008). In
particular, AT is a function of the BE and attitudes at each of two time points (demo-
graphics are also dropped for simplicity):

AT1 ¼ f ðBE1 ; AP1 Þ þ e1 (4)

AT2 ¼ f ðBE2 ; AP2 Þ þ e2 (5)


Urban Geography 241

If the attitudes do not change over time ðAP1 ¼ AP2 Þ, the subtraction of the two
equations yields

ðAT2  AT1 Þ ¼ gfðBE2  BE1 Þg þ ðe2  e1 Þ (6)

or

ΔAT ¼ gðΔBEÞ þ Δe (7)

The influence of such attitudes on change in AT is cancelled out.


Scholars implemented longitudinal studies in several ways (McCormack & Shiell,
2011). Some have compared AT before and after BE interventions (Boarnet, Anderson,
Day, McMillan, & Alfonzo, 2005; Brown & Werner, 2007). However, the locations of
interventions are not random, since such interventions are usually focused on locations
that are often more deficient on the dimensions that the interventions aim to enhance.
Furthermore, the assignments of treatment and control groups are not random either
(Mokhtarian & Cao, 2008). Some have compared AT before and after residential reloca-
tion (Giles-Corti et al., 2013; Krizek, 2003; Mumford, Contant, Weissman, Wolf, &
Glanz, 2011). However, residential relocation is not a random treatment, but driven by
residents’ changes in lifestyle, employment, and so on. Nevertheless, both types of studies
are more desirable than cross-sectional studies. In lieu of expensive and time-consuming
longitudinal design, some scholars have adopted a quasi-longitudinal design by asking
respondents to recall their BE and travel behavior before residential relocation
(Aditjandra, Mulley, & Nelson, 2009; Handy, Cao, & Mokhtarian, 2006). This study
adopts the retrospective approach. However, the cost of the approach is that respondents
might recall their previous neighborhood characteristics differently depending on their
previous pattern of AT.

Data and variables


The data come from a self-administered 10-page survey mailed in May 2011 to
households in five corridors in the Twin Cities. We chose three urban corridors and
two suburban corridors in the region (Figure 1). The urban corridors include the areas
within approximately half mile of Hiawatha Avenue, Nicollet Avenue, and
Bloomington Avenue in South Minneapolis (sections from Lake Street to 50th
Street). These three parallel corridors, mainly developed before World War II, have
similar characteristics in terms of regional location (Figure 2), street patterns and
transit access (Figure 3), and demographics (Table 1). Hiawatha Avenue is served
by light rail transit (LRT), whereas the other two are served by express and local
buses. The suburban corridors have similar income and house ownership to the
Hiawatha corridor but were mainly developed in the 1970s. Coon Rapids is about
14 miles directly north of downtown Minneapolis, and Burnsville is 17 miles directly
south of downtown. They have limited access to transit, and their street networks are
mainly curvilinear (Figure 3).
For each corridor, we purchased two databases of residents from AccuData Integrated
Marketing, a commercial data provider (http://www.accudata.com): a database of
“movers” and a database of “nonmovers.” The “movers” included all current residents
242 X. (Jason) Cao

Figure 2. Locations of corridors.

who had moved to the corridor after the opening of the Hiawatha LRT or after 2004
(if they do not live along the Hiawatha corridor). From this database, we drew a random
sample of about 1,000 residents from the Hiawatha corridor and about 500 residents from
each of Nicollet, Bloomington, Coon Rapids, and Burnsville corridors. The database of
“nonmovers” consisted of a random sample of about 1,000 residents from the Hiawatha
corridor and about 500 residents from each of the four corridors, who were not included in
the “movers” list for each corridor. We oversampled the Hiawatha corridor because the
survey was primarily to test the impact of the Hiawatha LRT on travel behavior.
The survey was pretested with students and staff members of our school as well as
several neighbors and social contacts; survey content was accordingly revised based on
pretesters’ feedback. The survey and two reminder postcards (1 and 2 weeks later) were
Urban Geography 243

Figure 3. Map of corridors.

mailed in May 2011. Ten respondents were drawn randomly and given a $50 gift card as
an incentive for completing the survey. The original database consisted of 6,017
addresses, but only 5,884 were valid. The number of responses totaled 1,303, equivalent
to a 22.2% response rate based on valid addresses only. This is considered quite good for
a survey of this length, since the response rate for a survey administered to the general
population is typically 10–40% (Sommer & Sommer, 1997).
Table 1 compares sample characteristics with the 2010 Census. Overall, home owners
are overrepresented in the sample, although the percentages of owners across different
corridors are similar. Respondents tend to have a smaller household size than the popula-
tion and households with children are underrepresented. These results are not surprising
because 54.5% of the respondents are nonmovers who have been living in their neighbor-
hoods for more than 7 years at the time of the survey. On average, nonmovers have much
smaller households and fewer children than movers, and nearly all nonmovers own their
houses.
244

Table 1. Sample characteristics versus 2010 census.

Hiawatha Nicollet Bloomington Coon Rapids Burnsville

Census Sample Census Sample Census Sample Census Sample Census Sample

Number of people 24,166 508 36,808 197 27,055 241 24,866 175 27,975 182
Percent of female (%) 50 52 49 49 50 51 51 49 52 49
Mean household size 2.27 2.15 2.50 2.21 2.65 2.26 2.53 2.42 2.31 2.24
Percent with kids (%) 26 23 43 22 35 24 32 29 35 22
X. (Jason) Cao

Percent of owners (%) 68 83 49 82 62 84 77 86 58 84

Note: ANOVA tests indicate that the means of the last four variables do not differ among different corridors.
Urban Geography 245

The variables used in this study consist of five groups: AT, neighborhood char-
acteristics, residential preferences, travel attitudes, and demographics. In the survey,
AT was measured using Questions 10–13 of Transportation Physical Activity in the
long form International Physical Activity Questionnaire (IPAQ, http://www.ipaq.ki.se/
ipaq.htm, accessed on 1 March 2011). IPAQ has been tested for reliability and validity
in many countries. In particular, in the survey, respondents were asked to indicate the
number of days during the last 7 days they walked (or bicycled) for at least 10 minutes
at a time to go from place to place and the duration they usually spent on one of those
days. AT was then computed as a continuous scale in MET (metabolic equivalent)-
minutes based on IPAQ guidelines for data processing and analysis (IPAQ, p. 7,
accessed on 1 March 2011):

Walking MET-minutes/week for transport = 3.3 * walking minutes * walking days for
transportation
Cycle MET-minutes/week for transport = 6.0 * cycling minutes * cycle days for transportation
Total Transport MET-minutes/week = sum of Walking + Cycling MET-minutes/week scores
for transportation.

Physical Activity Guidelines for Americans (PAGA) recommends 500–1,000 MET-minutes


of activity to obtain substantial health benefits. The World Health Organization (WHO, 2010)
has similar recommendations:

Adults aged 18–64 should do at least 150 minutes of moderate-intensity aerobic physical
activity [3 or more MET] throughout the week or do at least 75 minutes of vigorous-intensity
aerobic physical activity [6 or more MET] throughout the week or an equivalent combination
of moderate- and vigorous-intensity activity (PAGA, p. 26)

and they should double the durations for additional health benefits.
Following the practice in the physical activity literature and using the PAGA guideline,
the intensity of AT was grouped into four categories: no activity (0 to 32 MET-minutes),
inadequate activity (33–499 MET-minutes), meeting PAGA (500–1,000 MET-minutes), and
exceeding PAGA (1,001 MET-minutes or above). According to ANOVA post hoc tests, we
found that respondents from urban corridors were more likely to conduct AT than suburban
respondents, and the former were more likely to meet or exceed PAGA than the latter
(Table 2).
Changes in AT for movers were measured retrospectively. Because it is difficult for
individuals to accurately recall the specifics of their AT before their relocation, respon-
dents were asked to indicate how their travel differs now, from before they moved. In
particular, two questions asked respondents about walking and biking on a 5-point ordinal
scale from “a lot less now” to “a lot more now.”
Adapted from the methods used by Handy et al. (2006), respondents were asked to
indicate how true 30 characteristics are for their current (for both movers and nonmovers)
and previous (for movers only) neighborhoods, on a 4-point scale from “not at all true” (1)
to “entirely true” (4). The 30 statements cover attributes associated with living units, land
use and transportation systems, safety, social environment, and so on. Perceived neighbor-
hood characteristics are individuals’ cognitive response to neighborhood design. Also the
importance of these items to respondents when/if they were looking for a new place to
live was measured on a 4-point scale from “not at all important” (1) to “extremely
important” (4). The comparison of individuals’ perceived neighborhood characteristics
for their current residence and their neighborhood characteristic preferences indicates how
246

Table 2. Physical activity by corridor.

Corridor N Mean* Median No activity (%) Inadequate PA (%) Meeting PAGA (%) Exceeding PAGA (%) Total (%)

Urban Hiawatha 508 666 284 23.4 41.7 14.5 20.4 100.0
Bloomington 241 604 231 27.8 39.0 14.8 18.4 100.0
Nicollet 197 631 275 25.5 40.2 12.5 21.7 100.0
Suburban Burnsville 182 295 0 51.4 33.5 6.4 8.7 100.0
Coon Rapids 175 332 33 49.4 30.5 7.3 12.8 100.0
X. (Jason) Cao

Note: *NOVA test indicates that the mean differs among different corridors at the 0.001 level.
Urban Geography 247

well their current neighborhoods meet their preferences. A factor analysis reduced these
items (after dropping some items) to eight factors: attractiveness, spaciousness, transit,
safety, quietness, socializing, accessibility, and physical activity infrastructure (Table 3).
Changes in neighborhood characteristics for movers were computed as the difference in
the factors between their current and previous neighborhoods.
To measure attitudes regarding travel, the survey asked respondents whether they
agreed or disagreed with a series of 21 statements on a 5-point scale from “strongly
disagree” (1) to “strongly agree” (5). Factor analysis was then used to extract the
fundamental dimensions spanned by the items, since some of the items are highly
correlated. Seven underlying dimensions were identified: pro-drive, pro-walk, pro-bike,
pro-transit, safety of car, status of car, and pro-travel (Table 4).
The survey also contained a list of demographic characteristics that explains respon-
dents’ AT. The demographics consist of gender, age, education, employment status,
mobility constraints, income, auto ownership, and household structure. Movers’ change-
able demographics were measured retrospectively. They were asked to report the current
number of household members and the number not long before their residential relocation.
Change in household size is the difference between the two variables. Also, movers were
asked to indicate whether their annual household income had increased, decreased, or
stayed about the same, using a 9-point ordinal scale ranging from “decreased by $17,500
or more” to “increased by $17,500 or more.”
Following the survey, we constructed a set of objective BE measures around each
respondent’s current home address using ArcGIS (ESRI, Redlands, CA, USA). Some
variables measure the distance from each respondent’s home address to the nearest feature
such as transit stops. We also measure characteristics around each respondent’s home, using
quarter-mile, half-mile, and one-mile network distance buffers. The variables include
measures related to density (such as population density and housing density), diversity
(such as entropy indices of land use mix and business mix), street connectivity (such as
number of four-/five-way intersections and cul-de-sacs), distance to activities (such as gym
and shops), and access to transit (distance to the closest station and stop density).

Results and discussion


The statistical control approach is used first to evaluate the impacts of neighborhood char-
acteristics and attitudinal factors on AT in the context of a cross-sectional design. Two ordered
logit models are incrementally estimated for AT. Model 1 includes demographics and
neighborhood characteristics and Model 2 further incorporates residential preferences and
travel attitudes. The Veall–Zimmermann R2 is chosen as a goodness-of-fit measure for the
models because it is better than the McFadden R2 when the number of ordinal categories
exceeds three (Veall & Zimmermann, 1996). Although all variables described in the previous
section are tested, parsimonious models are presented: Independent variables that are insig-
nificant at the 0.1 level are manually removed because they make trivial contributions to the
model fit. Table 5 shows that the inclusion of attitudinal factors greatly improves the model fit.
Model 1 shows that both age and any physical or anxiety condition that seriously
limits or prevents respondents from walking outside the home are negatively associated
with AT. After controlling for demographics, perceived physical activity infrastructure and
socializing have positive associations with AT. Regarding objective measures, population
density within a 1-mile buffer from respondents’ home is positively associated with AT
and the association is marginally significant. In contrast, the density of cul-de-sacs within
a quarter-mile buffer has a negative correlation with AT. Because high cul-de-sac density
248

Table 3. Pattern matrix for perceived and preferred neighborhood characteristics.

Physical
activity
Attractiveness Spaciousness infrastructure Transit Accessibility Quietness Safety Socializing

High-quality living unit 0.605


Good investment potential 0.391
Attractive appearance of neighborhood 0.673
High level of upkeep in neighborhood 0.609
Large back yards 0.640
Lots of off-street parking (garages or driveways) 0.481
Sidewalks throughout the neighborhood 0.452
Good bicycle routes beyond the neighborhood 0.839
Parks and open spaces nearby 0.377
Good public transit service (bus or rail) 0.964
Easy access to transit stop/station 0.921
Shopping areas within walking distance 0.393
X. (Jason) Cao

Easy access to a regional shopping mall 0.720


Easy access to downtown 0.375
Religious or civic buildings nearby 0.507
Low level of car traffic on neighborhood streets 0.874
Quiet neighborhood 0.735
Low crime rate within neighborhood 0.426
Safe neighborhood for walking 0.815
Safe neighborhood for kids to play outdoors 0.540
Lots of interaction among neighbors 0.784
Lots of people out and about within the neighborhood 0.752
Notes: The method was principal axis factoring with Oblimin with Kaiser Normalization. Loadings smaller than 0.300 were suppressed.
Table 4. Pattern matrix for travel attitudes.

Safety of car Status of car Pro-travel Pro-drive Pro-transit Pro-bike Pro-walk

Traveling by car is safer overall than walking. 0.748


Traveling my car is safer overall than taking transit. 0.582
Traveling by car is safer overall than riding a bicycle. 0.335 −0.308
It does not matter to me which type of car I drive. −0.642
To me, the car is nothing more than a convenient way to get around. −0.601
To me, the car is a status symbol. 0.324 0.341
Travel time is generally wasted time. −0.615
The only good thing about traveling is arriving at your destination. −0.544
Getting there is half the fun. 0.446 0.384
I like to drive just for fun. 0.692
I like driving. 0.665
I feel free and independent if I drive. 0.491
Urban Geography

I like taking transit. 0.754


Public transit can sometimes be easier for me than driving. 0.737
I prefer to take transit rather than drive whenever possible. 0.711
I prefer to bike rather than drive whenever possible. 0.846
Biking can sometimes be easier for me than driving. 0.829
I like riding a bike. 0.783
I prefer to walk rather than drive whenever possible. 0.734
I like walking. 0.650
Walking can sometimes be easier for me than driving. 0.546
Notes: The method was principal axis factoring with Oblimin with Kaiser Normalization. Loadings smaller than 0.300 were suppressed.
249
250 X. (Jason) Cao

Table 5. Ordered logit models for active travel.

Model 1 Model 2

Variables β p-value β p-value

Threshold 1 −1.181 0.000 −1.119 0.000


Threshold 2 0.576 0.014 0.928 0.000
Threshold 3 1.288 0.000 1.753 0.000
Age −0.010 0.007
Limitations on walking −1.359 0.000 −0.977 0.000
Neighborhood characteristics
Perceived physical activity infrastructure 0.210 0.000
Perceived socializing 0.126 0.026
Population density within 1 mile 0.024 0.072
Density of cul-de-sacs within 1/4 mile −6.418 0.000 −5.173 0.002
Preferences and attitudes
Preferred physical activity infrastructure 0.120 0.035
Preferred transit 0.160 0.011
Pro-bike attitude 0.557 0.000
Pro-walk attitude 0.342 0.000
Pro-transit attitude 0.117 0.049
Pro-travel attitude 0.075 0.083
The number of observations 1,194 1,194
Veall–Zimmermann R2 0.137 0.325

indicates poor street connectivity and connectivity is positively associated with AT


(Ewing & Cervero, 2010; Frank, Engelke, & Schmid, 2003; Marshall & Garrick, 2010),
this negative correlation is reasonable.
Two residential preference variables and four travel attitude variables are positively
associated with AT in Model 2. Specifically, residents who prefer physical activity
infrastructure and attributes of public transit tend to participate in AT more than others;
pro-travel, pro-bike, pro-walk, and pro-transit attitudes are positively associated with AT.
However, age and three neighborhood characteristics (perceived physical activity infra-
structure, perceived socializing, and population density) become insignificant and hence
are removed from Model 2. Cul-de-sac density is still significant, but the magnitude of its
coefficient decreases. It is worth noting that we also tried to include corridor dummy
variables or urban/suburban dummy variables in the model, but none are significant.
Table 6 illustrates parsimonious models for change in walking and biking. Because
attitudes may moderate the impacts of changes in the BE on changes in AT, their current
measures are allowed to enter the model. For example, given the same improvement in
walkability, individuals with a strong walking preference may increase their AT much
more than those with a weak walking preference. Table 6 shows that individuals who
favor bike (or walk) are more likely to increase their biking (or walking, respectively).
Those who prefer physical activity infrastructure are more likely to increase their walking.
Thus, individuals with different attitudes seem to respond to changes in neighborhood
characteristics differently. On the other hand, because attitudes are measured currently, we
are uncertain whether attitudes lead to additional changes in AT or attitudes are reinforced
by current travel patterns if travel has changed. A longitudinal analysis with attitudinal
measures before and after residential relocation would be necessary to answer this
question.
Urban Geography 251

Table 6. Ordered logit models for changes in active travel.

Walk Bike

Variables β p-value β p-value

Threshold 1 −1.690 0.000 −1.147 0.000


Threshold 2 −0.677 0.001 −0.616 0.004
Threshold 3 1.131 0.000 1.579 0.000
Threshold 4 2.150 0.000 2.428 0.000
Pro-bike attitude 0.788 0.000
Pro-walk attitude 0.253 0.000
Preferred physical activity infrastructure 0.210 0.001
Demographics
Change in household income 0.108 0.001 0.089 0.009
Change in household size −0.149 0.065
Perceived neighborhood characteristics
Change in physical activity infrastructure 0.154 0.029 0.215 0.001
Change in safety 0.215 0.000 0.155 0.006
Change in socializing 0.385 0.000
Change in accessibility 0.165 0.011
Change in transit 0.147 0.067
The number of observations 568 538
Veall–Zimmermann R2 0.271 0.250

After controlling for attitudinal factors and demographics, changes in perceived


neighborhood characteristics (the difference between the perceptions of current and
previous neighborhoods) have positive associations with changes in AT. Changes in
physical activity infrastructure and safety are associated with changes in both walking
and biking. Changes in transit, socializing, and accessibility are significant correlates of
change in walking. It is worth noting that if demographics and attitudes are not included
in the models, all of the five perceived neighborhood characteristics are still significant in
the walking model while changes in safety become insignificant at the 0.1 level in the
biking model (results not shown). Overall, the contributions of attitudes and demographics
to changes in walking/biking appear to be additional to those of neighborhood
characteristics.
The models offer insights into our central questions. Is the difference in AT attribu-
table to self-selection effect? Table 5 shows that people who favor alternative modes of
transportation are more likely to participate in AT. T-tests indicate that urban residents are
positively associated with AT and pro-walk and pro-bike attitudes. Therefore, those liking
AT tend to live in urban areas and participate in AT more than others. This suggests a
residential self-selection effect. The displacement of perceived physical activity infra-
structure by preferred physical activity infrastructure in Model 2 also provides evidence
for the argument that residential choice is endogenous to a model of AT.
Do neighborhood characteristics affect AT? The answer is also yes. Cul-de-sacs
density is significant after controlling for demographics and attitudes and hence has an
independent impact on AT beyond self-selection. Changes in neighborhood characteristics
are significantly associated with changes in AT after accounting for the influences of
attitudinal factors and demographics. Controlling for confounding factors presumably
precludes spurious relationships between neighborhood characteristics and AT. Because
changes in neighborhood characteristics occur before changes in AT, the quasi-
252 X. (Jason) Cao

Table 7. The elasticities for changes in active travel.

A lot less A little less About the A little more A lot more
Variables now now same now now

Biking
Change in physical activity −0.133 −0.112 −0.046 0.038 0.104
infrastructure
Change in safety −0.025 −0.025 −0.015 −0.001 0.014
Walking
Change in physical activity −0.102 −0.093 −0.059 −0.007 0.061
infrastructure
Change in safety −0.045 −0.053 −0.051 −0.035 0.008
Change in accessibility −0.073 −0.068 −0.044 −0.007 0.042
Change in transit −0.092 −0.081 −0.047 0.001 0.059
Change in socializing −0.107 −0.139 −0.153 −0.123 −0.004

longitudinal analysis helps to establish time precedence. Therefore, this study suggests
evidence for the three prerequisites of a causal inference and hence produces a more
robust causal inference than do most studies.
Among the significant neighborhood characteristics, which plays a more important
role in affecting AT? To answer this question, the elasticities of significant neighborhood
characteristics are computed. Elasticity measures the percentage increase in the probability
of choosing a particular change in AT (say, “a lot more now”) if an independent variable
grows by 1%. It is a measure of effect size: How important is a variable in practice?
Table 7 shows that the elasticities of changes in neighborhood characteristics fall into the
range of 0.05–0.15, consistent with previous research (Ewing & Cervero, 2001, 2010).
Change in physical activity infrastructure is much more important in increasing biking
than change in safety. Depending on the level of change in walking, the elasticities of
changes in neighborhood characteristics vary a great deal. As shown in the last column of
Table 7, physical activity infrastructure, transit, and accessibility improvements appear to
be more effective for increasing walking than socializing and safety improvements.

Approach comparison
The cross-sectional approach (Table 5) and quasi-longitudinal analysis (Table 6) produce
different results. As shown in Table 5, residential preferences and travel attitudes are
extensively present and the associations between neighborhood characteristics and AT are
greatly attenuated after adjusting for these circumstances. In contrast, neighborhood
characteristics dominate the correlates in Table 6. That is, cross-sectional statistical control
highlights self-selection effects, whereas quasi-longitudinal analysis points to the impor-
tance of the neighborhood design effects that persist even after accounting for all relevant
selection bias.
Why do the two approaches behave so differently? First, statistical control may
impose overstrict controls on neighborhood characteristics. Since attitudes are mostly
measured currently, it is likely that attitudes are adapted from neighborhood characteristics
and AT. If this is true, the influence of neighborhood characteristics on AT will be
understated by (1) ignoring their impacts through mediating variables—attitudes—and
(2) exaggerating the effect of attitudes (since the association between attitudes and AT is
artificially inflated). Moreover, longitudinal analysis accounts for the influence of the
Urban Geography 253

attitudes that do not change over time although they may be significantly associated with
AT at more than one time points. Therefore, attitudinal variables are less likely to appear
in longitudinal models than in cross-sectional models.
However, longitudinal studies may obscure processes of residential choice. For exam-
ple, an individual who prefers AT intentionally chooses a pedestrian-friendly environment.
That is, she self-selects residential location to match her travel preference. This relocation
leads to an increase in walkability, which in turns results in an increase in walking. If the
preference does not change before and after relocation, it may not appear in the model for
change in walking. Therefore, self-selection exists but is not observed. If the preference
before relocation is measured, the self-selection process may be identified through a
structural equations model by linking the preference and change in walkability. If it is
not measured, the statistical control approach can be an important complement to under-
stand the other mechanism underlying the impacts of neighborhood design on AT—a
walkable neighborhood facilitates people to find a place to match their walking preference
and walk more (Cao et al., 2009; Chatman, 2009; Næss, 2009).

Conclusions and implications


This study employs cross-sectional and quasi-longitudinal analyses to disentangle the
influences of neighborhood characteristics and residential self-selection on AT in the Twin
Cities. We find that travel attitudes and residential preferences are extensively associated
with AT. Yet even after accounting for all travel and residential selection biases, neighbor-
hood characteristics have significant and independent impacts on AT.
For implementation, planners must consider three questions. First, which dimensions
of neighborhood design are associated with AT? This study points to physical activity
infrastructure, socializing, safety, transit, and accessibility. Second, what characteristics
are the most important in practice? Physical activity infrastructure is critical for biking.
Physical activity infrastructure, transit, and accessibility are the top three factors in
promoting walking. Third, what policies can bring about these kinds of changes? The
Complete Streets program can create nonmotorized transportation infrastructure that is
essential to AT. Enhancing transit quality of service and improving pedestrian connections
between neighborhoods and transit stations/stops are likely to increase the competitive-
ness of transit (CTOD, 2008) and thence walking to/from transit stations. Directing
development toward existing communities (as opposed to greenfield development) and
promoting mixed-use development in residential areas are able to limit restless expansion
and increase accessibility. Overall, a comprehensive package of strategies needs to be
implemented.
This study has several limitations. First, we investigate individuals’ transportation-
related physical activity and do not capture their activities in other domains. We are
uncertain whether a high level of AT is associated with high level of total physical activity
(Rodríguez, Khattak, & Evenson, 2006). On the other hand, AT is an important form of
physical activity in and of itself in an urban–suburban society where there are clear public
health benefits of additional physical activity of any kind. Second, we explore the
influence of residential environment on AT. Our IPAQ measure may include AT outside
of residential neighborhoods. Future studies should attempt greater precision in matching
the domain of environmental causal factors with the domain where AT choices and
behaviors take place. Third, the changes in AT were measured retrospectively. We
intentionally focus on recall accuracy, rather than measurement precision, as retrospective
measures of perceived characteristics of previous neighborhoods are also subject to recall
254 X. (Jason) Cao

bias. Last but not the least, perceived neighborhood characteristics may be vulnerable to
reporting bias; for example, individuals who participate in AT may be more aware of the
existence of amenities and nonmotorized transportation infrastructure nearby. This bias
may inflate the impact of neighborhood characteristics on AT. Thus, a true longitudinal
study is required to obtain precise estimates of BE influence on AT, as Giles-Corti et al.
(2013) did. Furthermore, researchers could develop a parallel study: a true longitudinal
design is compared to a quasi-longitudinal design when the true longitudinal project is to
be implemented. This will help to quantify the impact of reporting and recall biases.

Acknowledgements
Data collection was supported by the Transitway Impact Research Program in the Twin Cities.
Comments from three anonymous referees greatly improved the paper.

References
Aditjandra, P., Mulley, C., & Nelson, J. (2009). Neighbourhood design impact on travel behaviour:
A comparison of US and UK experience. Projections, 9, 28–52.
Boarnet, Marlon, Anderson, Craig L., Day, Kristen, McMillan, Tracy, & Alfonzo, Mariela (2005).
Evaluation of the California safe routes to school legislation - urban form changes and children’s
active transportation to school. American Journal of Preventive Medicine, 28(2), 134–140.
Boarnet, Marlon, & Crane, Randall (2001). Travel by design: The influence of urban form on travel.
New York, NY: Oxford University Press.
Boone-Heinonen, Janne, Gordon-Larsen, Penny, Guilkey, David K., Jacobs, David R. Jr, & Popkin,
Barry M. (2011). Environment and physical activity dynamics: The role of residential self-selection.
Psychology of Sport and Exercise, 12(1), 54–60.
Brown, Barbara B., & Werner, Carol M. (2007). A new rail stop: Tracking moderate physical
activity bouts and ridership. American Journal of Preventive Medicine, 33(4), 306–309.
Brownson, Ross C., Boehmer, Tegan K., & Luke, Douglas A. (2004). Declining rates of physical
activity in the United States: What are the contributors? Annual Review of Public Health, 26(1),
421–443.
Cao, Xinyu, Mokhtarian, Patricia L., & Handy, Susan L. (2011). Examining the impacts of
residential self-selection on travel behaviour: Methodology and empirical findings. In E.
Venezia (Ed.), Urban sustainable mobility (pp. 15–100). Milan: FrancoAngeli.
Cao, Xinyu (Jason), Mokhtarian, P. L., & Handy, S. L. (2009). Examining the impacts of residential
self-selection on travel behaviour: A focus on empirical findings. Transport Reviews, 29(3),
359–395.
Cervero, Robert, & Duncan, Michael (2003). Walking, bicycling, and urban landscapes: Evidence
from the San Francisco Bay Area. American Journal of Public Health, 93(9), 1478–1483.
Chatman, Daniel G. (2009). Residential choice, the built environment, and nonwork travel:
Evidence using new data and methods. Environment and Planning A, 41(5), 1072–1089.
CTOD. (2008). Station area planning: How to make great transit-oriented places. Oakland, CA:
Reconnecting America’s Center for Transit-Oriented Development. Document Number
TOD202.
Ding, Ding, & Gebel, Klaus (2012). Built environment, physical activity, and obesity: What have we
learned from reviewing the literature? Health & Place, 18(1), 100–105.
Durand, Casey P., Andalib, Mohammad, Dunton, Genevieve F., Wolch, Jennifer, & Pentz, Mary
Ann (2011). A systematic review of built environment factors related to physical activity and
obesity risk: Implications for smart growth urban planning. Obesity Reviews, 12(5), e173–e182.
Ewing, Reid, & Cervero, Robert (2001). Travel and the built environment: A synthesis.
Transportation Research Record: Journal of the Transportation Research Board, 1780, 87–114.
Ewing, Reid, & Cervero, Robert (2010). Travel and the built environment—A meta-analysis.
Journal of the American Planning Association, 76(3), 265–294.
Finer, Nicholas. (2006). Medical consequences of obesity. Medicine, 34(12), 510–514.
Frank, Lawrence D., Engelke, Peter, & Schmid, Thomas (2003). Health and community design: The
impacts of the built environment on physical activity. Washington, DC: Island Press.
Urban Geography 255

Giles-Corti, Billie, Bull, Fiona, Knuiman, Matthew, McCormack, Gavin, Van Niel, Kimberly,
Timperio, Anna, . . . Boruff, B. (2013). The influence of urban design on neighbourhood
walking following residential relocation: Longitudinal results from the RESIDE study. Social
Science & Medicine, 77, 20–30.
Handy, Susan, Cao, Xinyu, & Mokhtarian, Patricia L. (2006). Self-selection in the relationship
between the built environment and walking: Empirical evidence from northern California.
Journal of the American Planning Association, 72(1), 55–74.
Handy, Susan L., Boarnet, Marlon G., Ewing, Reid, & Killingsworth, Richard E. (2002). How the
built environment affects physical activity: Views from urban planning. American Journal of
Preventive Medicine, 23(2), 64–73.
Krizek, Kevin J. (2003). Residential relocation and changes in urban travel: Does neighborhood-
scale urban form matter? Journal of the American Planning Association, 69(3), 265–281.
Marshall, Wesley, & Garrick, Norman (2010). Effect of street network design on walking and
biking. Transportation Research Record: Journal of the Transportation Research Board, 2198,
103–115.
McCormack, Gavin, & Shiell, Alan (2011). In search of causality: A systematic review of the
relationship between the built environment and physical activity among adults. International
Journal of Behavioral Nutrition and Physical Activity, 8(1), 125.
Mokhtarian, Patricia L., & Cao, Xinyu (2008). Examining the impacts of residential self-selection
on travel behavior: A focus on methodologies. Transportation Research Part B:
Methodological, 42(3), 204–228.
Mumford, Karen G., Contant, Cheryl K., Weissman, Jennifer, Wolf, Jean, & Glanz, Karen (2011).
Changes in physical activity and travel behaviors in residents of a mixed-use development.
American Journal of Preventive Medicine, 41(5), 504–507.
Næss, Petter (2009). Residential self‐selection and appropriate control variables in land use: Travel
studies. Transport Reviews, 29(3), 293–324.
Rodríguez, Daniel A., Khattak, Assad J., & Evenson, Kelly R. (2006). Can new urbanism encourage
physical activity?: Comparing a new urbanist neighborhood with conventional suburbs. Journal
of the American Planning Association, 72(1), 43–54.
Saelens, Brian E., & Handy, Susan L. (2008). Built environment correlates of walking: A review.
Medicine & Science in Sports & Exercise, 40(7), S550–S566.
Singleton, Royce, & Straits, Bruce C. (2005). Approaches to social research (4th ed.). New York,
NY: Oxford University Press.
Sommer, Barbara B., & Sommer, Robert (1997). A practical guide to behavioral research: Tools
and techniques (4th ed.). New York, NY: Oxford University Press.
Stevens, Robert, & Brown, Barbara (2011). Walkable new urban LEED_Neighborhood-
Development (LEED-ND) community design and children’s physical activity: Selection, envir-
onmental, or catalyst effects? International Journal of Behavioral Nutrition and Physical
Activity, 8(1), 139.
Veall, Michael R., & Zimmermann, Klaus F. (1996). Pseudo-R2 measures for some common limited
dependent variable models. Sonderforschungsbereich 386 (pp. 1–33). München: Ludwig-
Maximilians-Universität.
Wells, Nancy M., Evans, G. W., & Yang, Y. (2010). Environments and health: Planning decisions as
public-health decisions. Journal of Architectural and Planning Research, 27(2), 124–143.
WHO. (2010). Global recommendations on physical activity for health. Geneva: World Health
Organization.

You might also like