University of Southern California Dissertations and Theses

Effects of the perceived and objectively assessed environment on physical activity in adults and children

(USC Thesis Other)

Effects of the perceived and objectively assessed environment on physical activity in adults and children

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

EFFECTS OF THE PERCEIVED AND OBJECTIVELY ASSESSED ENVIRONMENT
ON PHYSICAL ACTIVITY IN ADULTS AND CHILDREN

by

Casey Philip Durand

________________________________________________________________________

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PREVENTIVE MEDICINE)

August 2012

Copyright 2012 Casey Philip Durand
ii

DEDICATION
Completion of my PhD would not have been possible without a supportive family who
never questioned the wisdom of twenty-three years of uninterrupted education. Mom,
Dad, Kyle, Jamey, Dulcie and Jolie, thank you for everything, including the good natured
ribbing.
Also, thank you to my future wife Emmy Andrepont for putting up with two and a half
years of a long-distance relationship, my complaints about arcane academic matters, and
for giving me the incentive to finish my degree as soon as possible. I love you more than
anything.

iii

ACKNOWLEDGEMENTS
I would like to thank my advisor and committee chair Mary Ann Pentz for guidance,
advice and mentoring the past four years. I would also like to thank Genevieve Dunton
for serving as an informal second advisor and on the dissertation committee as well. I
have learned a tremendous amount from both of you. Thanks to Ricky Bluthenthal, Jimi
Huh and David Sloane for serving as committee members and providing valuable
feedback and advice on the development of this dissertation
I would also like to thank the excellent staff of the Healthy Places project at USC.
Without their dedication to the project, I would not have the high quality data I used for
the second and third papers. Robert Gomez, Keito Kawabata, Cesar Aranguri, Frank
Cedeno, Evelyn Saldana, Chris Castro, Michael Vasquez, and Kelly Tsai deserve a
tremendous amount of credit for making this happen. Finally, thank you Marny
Barovich, Laura Navarette, and Ryan Wilkerson for taking care of an uncountable
number of things that allowed me to get my PhD without losing my mind.

iv

TABLE OF CONTENTS
DEDICATION .............................................................................................................................................................. ii
ACKNOWLEDGEMENTS ....................................................................................................................................... iii
LIST OF TABLES ...................................................................................................................................................... vi
LIST OF FIGURES ................................................................................................................................................... vii
ABSTRACT ............................................................................................................................................................... viii
CHAPTER 1: SPECIFIC AIMS, BACKGROUND & SIGNIFICANCE ........................................................... 1
SPECIFIC AIMS ..................................................................................................................................................... 1
Study 1 ................................................................................................................................................................ 1
Study 2 ................................................................................................................................................................ 1
Study 3 ................................................................................................................................................................ 1
Problem Importance .......................................................................................................................................... 3
Etiology of Physical Inactivity........................................................................................................................ 4
Individual psychological correlates .......................................................................................................... 4
Social environment correlates .................................................................................................................... 7
Physical environment correlates ............................................................................................................. 10
Combined Individual, Social and Physical Environment Models ................................................. 16
PROPOSED STUDIES ............................................................................................................................................. 25
OVERVIEW ........................................................................................................................................................... 25
CHAPTER 2: CORRELATES OF DISCORDANCE BETWEEN PERCEIVED AND OBJECTIVE
MEASURES OF CRIME AND PHYSICAL DISORDER .................................................................................. 26
Abstract ............................................................................................................................................................ 26
Introduction ................................................................................................................................................... 28
Methods ............................................................................................................................................................ 33
Results .............................................................................................................................................................. 39
Discussion ....................................................................................................................................................... 41
Study One Figures ........................................................................................................................................ 50
CHAPTER 3: PERCEIVED AND OBJECTIVE NEIGHBORHOOD PROFILES AND THEIR
ASSOCIATION WITH ACTIVE COMMUTING TO SCHOOL ...................................................................... 53
v

Abstract ............................................................................................................................................................ 53
Introduction ................................................................................................................................................... 55
Methods ............................................................................................................................................................ 57
Results .............................................................................................................................................................. 62
Discussion ....................................................................................................................................................... 64
Study Two Figures ....................................................................................................................................... 73
CHAPTER 4: A LONGITUNDINAL EXAMINATION OF THE ECOLOGOCAL MODEL OF
PHYSICAL ACTIVITY IN ADULTS ..................................................................................................................... 75
Abstract ............................................................................................................................................................ 75
Introduction ................................................................................................................................................... 77
Methods ............................................................................................................................................................ 79
Results .............................................................................................................................................................. 89
Discussion ....................................................................................................................................................... 92
Study Three Figures and Tables............................................................................................................. 97
CHAPTER 5: OVERALL SUMMARY, CONCLUSIONS AND FUTURE DIRECTIONS ..................... 102
REFERENCES ........................................................................................................................................................ 108
APPENDIX: MODEL TESTING FOR LIGHT ACTIVITY ........................................................................... 120

vi

LIST OF TABLES
Table 1.1: Confirmatory factor analysis results ........................................................................................ 50
Table 1.2: Crime model results ........................................................................................................................ 51
Table 1.3: Physical disorder model results ................................................................................................. 52
Table 3.1: Final Psychosocial and Environmental Confirmatory Factor Analysis ...................... 99
Table 3.2: Structural Model Fit Statistics and Indices ......................................................................... 100
Table 3.3: Nested Model Comparisons ...................................................................................................... 100
Table 3.4: Parameter Estimates from Overall Best-Fitting Structural Model (#2) ................. 101
Table A1: Structural Model Fit Statistics and Indices for Light Activity ...................................... 120
Table A2: Nested Model Comparisons for Light Activity ................................................................... 120

vii

LIST OF FIGURES
Figure 2.1: NEWS Model Fit by Number of Classes ................................................................................. 73
Figure 2.2: NEWS Latent Profile Analysis .................................................................................................... 73
Figure 2.3: IMI Model Fit by Number of Classes ....................................................................................... 74
Figure 2.4: IMI Latent Profile Analysis .......................................................................................................... 74
Figure 3.1: Longitudinal Measurement Model for Latent Aesthetic Factor................................... 97
Figure 3.2: Conventional EIP, Reverse PIE & Reciprocal EIP Models .............................................. 97
Figure 3.3: Conventional IEP, Reverse PEI & Reciprocal IEP Models .............................................. 98
Figure 3.4: Fully Reciprocal Model ................................................................................................................. 98

viii

ABSTRACT
The purpose of this dissertation is to examine multiple methods of measuring the
neighborhood environment, and how those methods may differentially impact physical
activity in adults and children. The first paper examines the correlates of discordance
between perceived and objective measures of neighborhood crime safety and physical
disorder. The second paper uses neighborhood audits and resident reports to develop
distinct neighborhood typologies, and then correlates those typologies with active
commuting to school by children. The third paper uses longitudinal data to test potential
causal mechanisms linking psychosocial factors, perceptions of the built environment,
and physical activity in adults.
In the first paper, social control was consistently associated with perceiving the
neighborhood more positively than objective statistics indicated. Hispanic ethnicity, time
spent in the neighborhood, and acculturation were associated with perceiving the
neighborhood more negatively than objective data indicated.
In the second paper, both assessments methods revealed a profile characterized by
moderate levels on all environmental variables. This profile was associated with the
highest probability of actively commuting to school, while a profile characterized by low-
density single-family residential development was associated with the lowest probability.
In the third paper, the overall best fitting model was one in which the perceived built
environment affected intra- and inter-personal factors, which in turn affected MVPA.
However, there was no evidence of any mediation effects within this model.
ix

Taken together, the results of the three papers indicate that our understanding of the
environment is method dependent; different methods can lead to quite different
depictions of a given area. Further, perceptions of individuals within the same
geographic region can also vary, indicating that a unique “filtering” process occurs in
terms of processing the environment for each person. The exact nature of this filtering
process remains unclear. However, we can say that attempts to change the environment
to induce more physical activity may find more success if they can identify segments of
the population who are at particular risk for having activity hindering perceptions, and
develop interventions to target them accordingly.

1

CHAPTER 1: SPECIFIC AIMS, BACKGROUND & SIGNIFICANCE
SPECIFIC AIMS
The proposed dissertation will examine the multiple ways that psychological constructs,
perceptions of the neighborhood social environment, individual perceptions of the
physical neighborhood environment, and objective measures of the physical
neighborhood can impact physical activity in adults and children.
Study 1
1. Characterize the extent of mismatch between perceived and objective indicators
of neighborhood disorder and separately, indicators of neighborhood crime, in a
sample of adults.
2. Use multi-level structural equation modeling to examine the correlates of
mismatch between subjective and objective measures of neighborhood disorder
and crime, including factors capturing demographics, acculturation, and
perceptions of the neighborhood social environment.
Study 2
3. Among a sample of adults, use latent profile analysis to construct neighborhood
typologies in two models: one using objective indicators of a variety of
neighborhood characteristics, and one using analogous perceived indicators.
4. Use an individual’s classification in a particular typology to predict their child’s
active commuting behavior.
Study 3
2

5. Use structural equation modeling to develop multiple separate longitudinal
models for a sample of adults to determine causal ordering among psychosocial
factors, environmental perceptions and physical activity.
6. After choosing one model as the best overall fit, examine specific mediating
effects to understand which paths may hold potential to affect physical activity.

3

Problem Importance
Even after decades of public health campaigns, excess weight continues to be a problem
that plagues societies around the world. Current data from the United States indicates
that approximately 33% of adults are overweight, 34% are obese and 6% are extremely
obese. When viewed against data from previous decades, it is clear that the problem is
not improving. The 1988-94 National Health and Nutrition Examination Survey
(NHANES) reported the prevalence of obesity and extreme obesity was 23% and 3%,
respectively
90
. Likewise, obesity and overweight among children has increased
significantly in recent years. For 6-19 year olds, prevalence of obesity has increased
from approximately 11% in the 1988-94 NHANES to 17% in the 2003-06 NHANES
16
.
Among both adults and children, the distribution of obesity and overweight are not
uniform across racial and socioeconomic subgroups. Racial minorities and those of lower
socioeconomic status are more likely to be overweight and obese
63, 92, 106, 134
.
Obesity is a major public health concern because of the medical conditions that stem
from it. Obese individuals are at higher risk of metabolic complications such as
hypertension, elevated cholesterol, and type 2 diabetes
62, 102, 118, 139
. This in turn leads to
greater incidence of certain forms of cancer, heart disease, stroke, and all-cause
mortality
39
. Individuals who are obese as children are at greater risk of being obese
adults, and overweight children are more likely to have cardiovascular disease, diabetes,
and certain cancers as adults
31, 76
.
4

While multiple factors, including genetics and diet, influence body mass, a major
modifiable risk factor is physical inactivity. To gain maximal benefit from physical
activity, the United States Department of Health and Human Services currently
recommends adults achieve at least 150 minutes per week of moderate-intensity activity,
and children get 60 minutes per day of primarily moderate or vigorous activity
128
.
Population levels of activity, however, remain well below that. Among children 6-19,
41% met the guidelines in 2003-2006, and females were significantly less likely to meet
them compared to males
10
. Only 12% of high school age students fully met the
guidelines in 2010; the percentage was lower for Hispanic and black students
15
. Among
adults, 4% achieved the recommendations in 2003-2004; again, the amount of activity
varies by gender and ethnicity
17, 124
.
Etiology of Physical Inactivity
Individual psychological correlates
In attempting to understand the causes of physical inactivity, the historic focus has been
on psychological constructs of the individual and their usefulness as predictors of actual
behavior
46
. This focus is built on a broader notion that behavior is a product of conscious
cognitive processes
61, 137
. If these processes can be uncovered, interventions can be
designed to manipulate them and induce physical activity. Indeed, this approach has
found a wide audience in understanding many other health promoting or compromising
behaviors, including smoking, drug abuse, risky sexual behavior, and adherence to
treatment regimens
121
.
5

Psychological constructs are considered intrapersonal characteristics. Among the more
researched intrapersonal constructs are: self-efficacy, the confidence an individual has in
their ability to carry out a behavior; attitudes, the overall evaluation of a behavior;
outcome expectancies, the value an individual places on a given outcome; and intentions,
the likelihood of engaging in a behavior
45
.
Multiple theories have evolved that integrate some or all of these constructs to explain
differences in physical activity among individuals. These include the Theory of Reasoned
Action/Theory of Planned Behavior (TRA/TPB), the Transtheoretical Model (TTM), and
Social Cognitive Theory (SCT)
45
. TRA proposes that the proximal determinant of
behavior is intention, a construct that is a function of subjective norms and attitudes
regarding behavior. TPB extends the model by adding perceived behavioral control as a
predictor of intentions. According to TTM, individuals move through five stages
(precontemplation, contemplation, preparation, action, and maintenance) during the
process of behavior change. Shifts from stage to stage are influenced by a person’s self-
efficacy as well as their perceived pros and cons of changing. SCT is made up of
multiple constructs, including outcome expectations and expectancies, reinforcements,
behavioral capability, the external environment, and most centrally, self-efficacy
9
.
Through a concept called reciprocal determinism, where changes within the person, their
environment and their behavior occur simultaneously and influence each other in the
process, SCT is a more dynamic model of behavior than previous theories.
6

Despite the effort devoted their study, the evidence for the utility of these types of
psychological constructs is somewhat mixed. For adults, self-efficacy has been identified
as the strongest predictor of physical activity,
94

120

125
and not coincidentally, it is often a
component of formal theoretical models as well as a common intervention target. A
recent meta-analysis examining the effects of interventions to increase physical activity-
related self-efficacy found these types of interventions have a relatively modest average
effect size of 0.16
4
. Further, this analysis simply tells us how effective interventions are at
increasing self-efficacy; it is unclear whether these necessarily will change behavior,
which is the more important question. This question can only be tested through formal
mediation analysis, and in a review of mediators of physical activity interventions among
adults, Lewis et al found only one study which formally tested and found support for self-
efficacy as a mediator
68
. Even when formal mediation criteria are relaxed for additional
studies included in this review, it is still difficult to infer a mediation effect. Similarly, a
2010 review by Rhodes et al of mediators of physical activity change found that five
studies formally tested self-efficacy as a mediator, and only one found a significant
mediation effect
104
. Other common psychological factors, including attitudes, norms,
perceived behavioral control and intentions (those commonly associated with TRA/TPB)
have only weak associations with activity, and the construct of knowledge has almost no
evidence to support a predictive effect on activity
104, 125
. Finally, a recent longitudinal
study that examined 14 possible mediators (including all those listed above) of a
workplace intervention on physical activity in women found no evidence that any of the
constructs acted as mediators
98
.
7

Several explanations exist for the weak effects noted in the above studies. With respect
to interventions, a major issue may be theoretical fidelity. Little work has been done to
understand the extent to which programs are truly incorporating theoretical constructs
and designing their programmatic elements to specifically target the chosen constructs
110
.
Another explanation is that the proposed psychological constructs simply are not
effective predictors of physical activity, at least when studied in isolation. This assertion
is bolstered by the fact that the physical activity interventions in the two previously
discussed reviews covered a wide range of settings, strategies, and theoretical models in
diverse populations, yet the evidence for psychological mediation effects is minimal
Taken as a whole, the evidence surrounding psychological constructs has led researchers
to take alternative approaches to study physical activity.
Social environment correlates
In contrast to the psychological factors above, social environment characteristics are
interpersonal factors. Interpersonal factors are similar to the intrapersonal subdivision in
that they both involve conscious cognitive processes; however, interpersonal factors arise
from an individual’s interaction with cultural and societal norms, and other members of
their family, friends, social group or community
70
. Constructs in the social environment
subdivision include: subjective norms, an individual’s perception of whether others
approve or disapprove of a behavior; social norms, the way people typically act in
particular situations; social cohesion, trust and connectedness among citizens; social
capital, interpersonal trust between citizens, sense of community, and social
participation; social control, the capacity of a group to regulate its members; and
8

collective efficacy, social control among neighbors and their willingness to act for a
common good
22, 45, 71, 82, 117, 129
. There is some discrepancy of definition and variation in
how these terms are used, and thus they are not mutually exclusive
82
. For example, some
authors consider social capital to be a higher order construct that subsumes factors like
social control and collective efficacy
129
.
With the exception of subjective norms, which are sometimes categorized as an
intrapersonal factor, most of the above constructs have not been studied quite as
extensively in terms of their impact on physical activity as psychological constructs have
been
26
. Research that has been conducted suggests a positive association between
indicators of the social environment and activity. Ball et al found that women who
participated in local groups or events and who resided in neighborhoods where residents
reported trust in one another were more likely to engage in leisure-time physical activity
6
.
Mummery et al similarly found that individuals who perceived low levels of social
capital were more likely to be physically inactive when compared to those with who
perceived higher levels
89
. In a sample of Japanese adults, higher trust among neighbors
was associated with lower levels of physical inactivity
129
. Cleland et al found that
neighborhood social cohesion was positively associated with leisure time physical
activity in women, though this association held only in the partially adjusted analysis
21
.
The same relationships may hold for children. In a study of Chicago youth, Craddock et
al reported that low levels of baseline social cohesion predicted low levels of general
physical activity at the two-year follow-up
26
. This longitudinal study provides the
9

beginnings of some causal, rather than just correlational, evidence for the effect of the
social environment.
McNeill et al and Craddock et al have suggested that the social environment, particularly
cohesion and social capital, may impact physical activity by reinforcing healthy social
norms for activity, and because of the increased level of communication and contact in
socially functional neighborhoods, residents may have heightened awareness of health
promoting activities available to them
26, 82
. Additionally, cohesion and social capital
serve as informal control mechanisms that discourage illegal behavior (e.g. drug use,
prostitution, petty crime) that might prevent residents from being active in public spaces.
Collective action may also stem from neighborhoods with positive social environments,
empowering them to push for greater resource investments that enable physical activity
(e.g. park improvements, sidewalk replacement, etc.)
22, 26
. Foster et al, in their review of
the impact of crime on physical activity, proposed a conceptually similar model, in which
the physical and social environments, as well as individual demographics, act both
directly and indirectly on activity
41
. In the indirect path, their effects are mediated by the
objective and subjective assessment of neighborhood crime safety. No study has fully
tested this mediation model, but there is evidence supporting one half of the mediation
effect, the path from the social environment to crime safety
57, 117
. The evidence
supporting the path from neighborhood crime safety to physical activity is uncertain,
though this may be due more to weak measures and misspecification of the level of
influence (i.e. measuring crime at too broad a level) than any inherent lack of correlation
between crime safety and activity
41, 109
.
10

In sum, a growing body of evidence suggests that the relationships and interactions
among residents of a community and the resulting social environment could be an
important correlate of physical activity, and they may work by influencing mediators
such as health knowledge, political action, and crime safety and disorder. Further
research is needed to understand how specific domains of the social environment impact
these potential mediators and physical activity, as well as whether their effects are
stronger or weaker in certain sociodemographic groups. Methodological research is also
needed that refines our ability to measure these constructs so as to determine whether
they represent truly unique facets of the community.
Physical environment correlates
Recently, a new research paradigm has emerged that attempts to explicitly account for an
individual’s environment, and specifically, the built environment, which consists of all
man-made modifications to our surroundings, including buildings, streets, sidewalks and
parks. Broadly speaking, the built environment can be conceptualized as having two
facets: objective and perceived
43
. The distinction between these two views is important
from a measurement/methodological standpoint as well as a substantive standpoint. Each
could potentially provide unique explanations for how surroundings affect behaviors.
The objective environment is that which is measured using objective means, such as
Geographic Information Systems (GIS), neighborhood audits, archival or planning
records searches, or business directories
54
. The perceived environment is measured using
resident self-report questionnaires about their neighborhood, e.g. the Neighborhood
11

Environment Walkability Scale
114
. Both the perceived and objective environments can be
further subdivided into discrete components that may also be uniquely linked to activity.
Access to destinations is an assessment of proximity to a given feature and/or how
convenient it is to get there. Presence or absence is a simple measure of whether a given
feature is present. Quality or desirability of a feature, for example, the maintenance of
sidewalks, can also be assessed. Finally, the extent to which environmental aspects are
present that do not directly concern concrete features, such as safety from crime or traffic,
can be measured.
The last two divisions help illustrate the sometimes muddy distinction between that
which is perceived and that which is objective. Quality often has connotations with
subjectivity, but as will be shown later, it can be assessed using audit instruments
typically categorized as objective; the same is true for measures of safety. Perhaps the
most fundamental distinction lies in the respondent. Perceived measures might best be
described as those sourced from individuals who live in or regularly interact with the
environment under assessment, while objective measures are derived by individuals who
lack that degree of context with a specific area. “Perceived” and “objective” are loaded
words, and thus should be used with caution. No instrument termed objective should be
thought of as bearing the absolute truth, nor should perceived measures be regarded as
wrong because they disagree with an objective measure. Ultimately, the two forms of
measurement may simply be providing alternate, but equally valuable, versions of reality.
In that sense, measures of the environment lie on a scale of subjectivity; “perceived”
measures exist toward one end, while objective measures exist toward the opposite end.
12

Methodologically, objective and subjective assessment of a given place may not
necessarily provide the same picture. In fact, the two may be at odds with each other
43
.
These are important considerations for researchers in deciding how best to characterize
an area; methods may not be interchangeable. Substantively, one characterization of the
environment may be more strongly linked to physical activity. For example, perceptions
of the likelihood of crime victimization might be a stronger predictor of activity than
objective crime statistics
41
.
Another important distinction to make is between transport physical activity and
recreational physical activity
112
. Transport activity is activity which is undertaken for
purposes of getting somewhere, for example, walking to the grocery store, bank, school,
etc. In other words, it is a means to an end, and can be thought of as an alternative to car
travel. Recreational activity is activity which is undertaken for activity’s sake. This
could be taking a bike ride through the neighborhood or playing in a park. As with the
objective and perceived environment, there are substantive implications to these different
definitions of activity. Some environmental characteristics may be more strongly
associated with one type of activity (see below)
51, 56, 65, 66
.
Built environment and physical activity research is still a relatively young field, and as
such the bulk of studies examining this relationship largely use cross-sectional,
correlation design and analytic strategies
36
. Overall, the evidence from these studies is
mixed. The strongest association seems to be with transport physical activity. Several
reviews have examined the correlates of transport-related physical activity, and the
13

results indicate that population density, distance to destinations, land use mix and parks
and recreation facilities may be associated with this type of activity
113

55
. Recreational
physical activity may be associated with pedestrian infrastructure and aesthetics
113

93
.
Unfortunately, there is less evidence of whether the perceived or objective environment is
more important, for either overall or specific types of physical activity. Some empirical
research seems to indicate that both play a role, but the perceived environment may be
the most important, or at the very least, the most proximal predictor of physical activity.
For example, a 2010 study by Parra et al found that when assessing health-related
quality-of-life and self-rated health, multiple perceived features, such as traffic and public
space safety and street noise were associated with all outcomes, but only one objective
feature, public park density, was associated with self-rated health
95
.
What the studies above do not tell us is how the objective and perceived environment
relate to each other, if at all. Most of the studies which concurrently assess both domains
simply put them into one regression and report the adjusted effects and statistical
significance; this, however, does not provide any detail about the interaction between the
two. Instead, models must be tested that specifically hypothesize relationships and then
evaluate them. There are several studies that attempt to do this. Weden et al, using a
structural equation modeling approach, found that perceptions of the neighborhood (e.g.
upkeep, air quality, safety, litter, etc.) mediate the relationship between objective
measures (e.g. unemployment rates, income levels, age, etc.) and self-rated health
136
. To
be sure, this is not the same conceptualization of “objective environment” as discussed
above. However, it appears to be the only study that models environmental mediation,
14

and provides some support for the idea of a straight line mediation effect, wherein
objective features predict perceptions, which in turn predict health behaviors. While this
view may make intuitive sense, there are also multiple studies which dispute this
relationship. The case against this model is built primarily on studies which have shown
poor agreement between objective measures and resident perceptions of the same
neighborhood
81

43
. If mediation was the best explanatory model, this type of
disagreement should not be found. An alternate possibility is that the objective
environment is a moderator of the perception-behavior relationship. Again, there is very
little research testing this model, but one study by Durand et al found that parents’
perceptions of the neighborhood, including aesthetics, walking infrastructure, and traffic
and crime safety, were more strongly associated with active commuting among those
living in an objectively more walkable community, thereby lending support to the
moderation model
35
. The implication then is that the objective environment acts as a
facilitator of the perception-behavior pathway. The reality of course is that the true effect
of the perceived and objective environment on physical activity is likely to be much more
complex, and involve mediation, moderation, and feedback loops among the various
constructs that make up the larger environment. Only further research that moves beyond
simple correlations and has a longitudinal component will be able to reduce this
uncertainty. Ultimately, this is a crucial gap because without a better understanding of
this relationship, we will be unable to identify points of entry and causal pathways that
can serve as the basis for future intervention work.
15

A final limitation of existing research is the lack of information about whether
environmental characteristics, either perceived, objective or both, have a multiplicative
effective, such that together they are more than the sum of their parts. People do not live
and interact in isolation with one environmental component. Neighborhoods, commercial
districts and entire metropolitan areas are composed of numerous features that may
simultaneously affect a person’s behavior. After establishing the individual effects of
discrete components, it will next be necessary to identify combinations of components
that have the greatest impact on physical activity
36
. These combinations could then
provide guidance to planners and policy makers about optimal neighborhood design and
redevelopment strategies to promote greater physical activity and lower levels of auto-
centric travel.
One study has attempted to create neighborhood profiles and examine their impact on
physical activity among a sample of adults. Adams et al used latent profile analysis to
construct four neighborhood profiles, which ranged from areas with low walkability, low
transit access and few opportunities for recreational activities, to areas that have high
walkability and numerous opportunities for recreation
1
. Residents in the high walkable
profile reported over an hour more per week of walking for transportation and leisure
activity compared to those in the least walkable profile. This study provides some initial
evidence that it is possible to identify combinations of neighborhood characteristics that
are strongly associated with physical activity.
16

It is important to note that a major limitation of virtually all existing research exploring
the relationship between the built environment and physical activity is that of self-
selection, which could potentially bias results from studies that do not account for it
42, 48
.
For example, those who tend to exercise more may choose to live in a neighborhood that
offers more possibilities for activity (e.g. near transit, shopping, has parks, trails, etc.)
Thus, it is preexisting proclivity to be active, not the environment itself, which causes
physical activity. In this case, any measure of how the environment is associated with
activity would be artificially high. There is also the possibility that individuals are in
effect sorted into neighborhoods on the basis of longstanding economic or racial patterns
in a community, such that certain individuals are more likely to reside in a particular
neighborhood or part of town, an area that may or may not be more activity-friendly than
other areas. Again this could give an inaccurate picture of how the neighborhood affects
activity, as a lack of activity may be more a function of race or income, two factors well
known to be associated with physical activity, and not the neighborhood environment
itself. While methods have been proposed to deal with self-selection, most studies have
yet to incorporate them
14
.
Combined Individual, Social and Physical Environment Models
As noted above, there is evidence to suggest a significant positive association between at
least some measures and features of the environment and physical activity. There are
also studies, however, that contradict these findings. A review of the association between
environmental characteristics and physical activity by Durand et al found that overall, for
several features including walkability of neighborhoods, access to parks and other
17

gathering spaces, and transit availability, there is primarily non-significant evidence for
an environment-activity correlation, and some studies included in the review actually
found a negative association
36
. These contrasting findings may of course simply be
artifacts of the non-experimental design of most of the reviewed studies. Another
possible explanation is that the environment is not directly influencing behavior; rather,
this relationship may be indirect and mediated by cognitive processes
101
. Though there is
only moderate evidence that these processes mediate the relationship between traditional
educational-type interventions and health behaviors, they may play a greater role when
acted upon by environmental conditions. Social Cognitive Theory, noted earlier, the
ecological framework, and integrative transactional theory (ITT) posit relationships
between intra- and inter-personal factors and the environment
9, 96, 115
.
The ecological model proposes several levels of influence on the individual, including
intrapersonal, sociocultural, policy and the physical environment
115
. Similarly, ITT
specifies person, social situation and environmental influences on behavior
96
. Like SCT,
it proposes reciprocal relationships among levels, and an overall outward direction of
effects from person to social situation to environment. These approaches differ from
previous models of health promotion in that they specify interactions among the various
levels instead of studying each one in isolation. Studying the causes of behavior in this
fashion aids intervention work because it provides more than one point of entry and
multiple causal pathways for programs and policies; in fact, multiple intervention
strategies could be deployed at the same time, potentially increasing the chances of
achieving the desired outcomes.
18

What the ecological model does not make clear is the direction of influence. This can
only be elucidated through research that simultaneously accounts for these multiple
levels. Because of how ecological models have been presented, some have implicitly
assumed that they function similarly to what ITT proposes: the physical environment is
the most distal predictor of behavior, while intrapersonal factors are the most proximal
58
.
A number of examples that have tested this direction of effect are discussed next.
In a prospective study of the effects of neighborhood perceptions and functional
limitations on physical activity of older women, Morris et al found evidence for
mediating effect of self-efficacy
86
. However, the effects were small (indirect Betas of
~.03), such that the practical impact of this indirect effect are uncertain. Two studies by
Motl et al showed similarly small effects of self-efficacy as a mediator between the
perceived environment and physical activity in adolescent girls, though the effect was
stronger cross-sectionally (indirect Beta=0.22) than longitudinally (indirect Betas=0.02
and 0.03)
87, 88
. Prodaniuk et al reported on the possible effect of self-efficacy and
outcome expectations as a mediator of the relationship between the workplace
environment and workplace physical activity in adults
101
. Outcome expectations did not
show evidence of acting as a mediator, while self-efficacy was only weakly supported. In
a longitudinal follow-up study with this same sample, Plotnikoff found modest support
for self-efficacy as a mediator of the work environment-activity relationship
99
. Cross-
sectional indirect Betas at baseline, 6 and 12 months ranged from 0.05 to 0.09.
Longitudinally (work environment at time 1 predicting self-efficacy at time 2 predicting
activity at time 3), the indirect Beta was 0.04. McNeill et al assessed intrinsic motivation
19

and self-efficacy as mediators of the effect of neighborhood quality and availability of
activity facilities on walking, moderate and vigorous intensity physical activity
83
. They
appear to have found support for a path from environment to motivation to self-efficacy
to physical activity, but this was not quantified and did not appear to have been subjected
to a formal test of mediation. An analysis by Maddison et al examined constructs from
the Theory of Planned Behavior (attitudes, subjective norms, perceived behavioral
control and intentions) as mediators of the relationship between both the perceived and
objective physical environments and physical activity in adolescents
75
. Results showed
that none of the psychosocial variables mediated the effect of the environment on
activity. A similar study by Rhodes et al also proposed TPB constructs as mediators
between the perceived environment and walking in adults
105
. The effects of infrastructure
quality (indirect Beta=.09) and neighborhood aesthetics (indirect Beta=.05) on walking
appeared to be mediated by a path from attitudes to intentions. There was no evidence of
other psychosocial variables acting as mediators. In yet another study which used TPB as
a framework for mediation, de Bruijn et al did not find evidence of any mediation
between environmental measures and physical activity in a sample of Dutch
adolescents
29
. Also using TPB, Lemieux et al looked for mediating effects between
perceptions of the environment and active commuting among college students
67
. Results
indicated that the effect of neighborhood housing characteristics and the time it takes to
access services, work or school on intention to actively commute were mediated by
perceived behavioral control; it is not clear if this mediation effect extended to actual
activity, nor is the magnitude of the effect apparent. In an analysis of the association
20

between proximity to commercial physical activity facilities and vigorous physical
activity in high school age girls, Dowda et al found some evidence for a direct effect of
objectively-measured accessibility on activity, but no evidence that that self-efficacy
mediated the effect
33
. Using subjective assessments, they found an indirect effect between
perceived access to facilities and vigorous activity, mediated by self-efficacy; the effect
however, was small (indirect Beta=.01). Analyzing data separately by gender,
McCormack et al found evidence that perceived behavioral control mediated the
association between physical activity facility accessibility and vigorous activity in both
men and women, and the same relationship except with moderate activity in women
only
79
. Like several other studies here, this relationship was modest (indirect Betas of
approximately 0.08), and did not hold for the other seven perceived environment
characteristics assessed. Cerin et al examined self-efficacy as a mediator of the
relationship between perceived access to recreational facilities and walking, moderate
and vigorous physical activity
18
. No tests of mediation proved significant for any
outcome. Among a sample of adolescents, van der Horst et al tested the Theory of
Planned behavior as a mediator between measures of the home and neighborhood
environment and sports participation
131
. They found that attitude and intention
significantly mediated the relationship, and report the percent mediated ranged from 17%
to 82%. While seemingly large, these effects should be interpreted with caution, since it
is unclear what the absolute magnitude of the indirect effect really is. A more expansive
study by Ishii et al looked at multiple psychosocial variables, including pros and cons of
physical activity, self-efficacy, and social support as potential mediators, and examined
21

numerous pathways by which they may influence activity in a sample of Japanese
adults
53
. For three separate models in which walking, moderate intensity and vigorous
intensity activity were outcomes, the indirect effects of the environment through various
combinations of the psychosocial variables yielded small coefficients, the single largest
of which was .022. This study provides more insight than those previously discussed
because of the greater number of mediators examined, but again, the results are small.
A limitation of the studies above is that they are primarily cross-sectional. Therefore, it
is not possible to say definitively whether the pathways they examine are of value or not.
Thus, the environment-intrapersonal-activity path is still a possibility. An additional
possibility is essentially the reverse, where intrapersonal characteristics influence
perceptions of the environment, which in turn influence physical activity. This model has
not been directly tested, but there is some empirical research to suggest this may be the
case.
This view is supported by evidence from a study by Gebel et al showing that individuals
who reside in the same neighborhood can have markedly different perceptions of the
walkability of that neighborhood, and that those with more positive perceptions (meaning
greater awareness of environmental features thought to promote activity) are more likely
to engage in physical activity
43
. Something, therefore, is causing these differing
perceptions.
While demographic and socioeconomic factors no doubt play a role, the reversed model
(intrapersonal-environment-activity) would also propose that pre-existing psychosocial
22

variables are an integral part of an individual’s appraisal of their neighborhood vis-à-vis
physical activity. Further evidence for this model comes from a study by van Stralen et al
that assessed the impact of a psychosocially-oriented intervention combined with tailored
environmental information on physical activity
132
. They found support for perceptions of
the environment as a mediator between the intervention and physical activity. These
results provide some evidence that perceptions are the proximal determinant of physical
activity, and can be the product of psychosocial changes, similar to what the reversed
model here proposes.
The primary justification for this pathway, however, is theoretical. The reciprocal
determinism aspect of SCT implies that intrapersonal characteristics can influence the
environment just as much as the environment can influence intrapersonal characteristics
8
.
Intrapersonal constructs could cause an individual to place themselves in different
environments that are either more or less supportive for healthy behaviors
9
. This
placement may be literal (e.g. moving to a neighborhood, seeking a new social
environment) or figurative, in the sense that perceptions of their environment might be
altered. Support for this notion is found in studies of organizational psychology, where it
is thought that negative intrapersonal factors such as attitudes, beliefs, or orientation may
cause individuals to selectively process only the threatening or negative aspects of their
workplace, causing their perceptions of the work environment to be more negative than
an objective assessment might otherwise indicate
47
. Conversely, positive intrapersonal
characteristics may cause selective processing of supportive features of the environment,
enabling an individual to disregard unsupportive or negative aspects.
23

To make matters more complex, true reciprocal determinism would further contend that
current behavior is in fact a determinant of future behavior, because current behavior can
influence future intrapersonal and environmental determinants, which in turn affect future
behavior
8
. The interrelationship of these domains makes determining one causal path all
the more difficult. This creates the chicken-and-egg problem Bandura referred to: every
distal determinant ultimately has an even more distal determinant. As a concrete
example, it is conceivable current neighborhood walking behavior may create more
positive appraisals of the neighborhood social and physical environment, because the
individual is out and about, with greater opportunities to interact and form social ties to
their neighbors. This may lead to greater social support and norms for walking, and
increased self-efficacy for daily walking. Finally, this leads to greater amounts of
walking in the future.
Given the multiple ways these factors interact, it is crucial to consider alternative
plausible causal pathways. Researchers in fields as diverse as communications, nursing,
eating disorders, and occupational health have recently increased efforts to explore not
only conventional unidirectional models (i.e. models with no feedback loops), but also
reversed unidirectional models and reciprocal causation models (i.e. models with
feedback loops) in a longitudinal framework
37, 47, 119, 130, 138
. In some cases this has
confirmed conventional models of causality, but in others it has provided powerful
arguments for different directions of causal influence. As reversed and reciprocal
directions have not been explored in ecological models of physical activity, a major gap
24

exists in our understanding of how the person, their environment and behavior affect each
other.

25

PROPOSED STUDIES
OVERVIEW
At this point it is clear that a great many factors are associated with physical activity, but
it is much less clear how these diverse factors come together as a whole to impact
activity. Broadly, the three studies below aim to answer questions about how different
measurement methods can provide quite different pictures of those factors and their
subsequent associations with activity, as well as the causal structure through which
multiple levels of influence can affect physical activity. Each study grows in terms of
complexity and the kinds of questions to be answered; therefore, aside from the overall
theme connecting them, successive studies will build on the results of the prior study in
order to answer the most complex questions in the final analysis. The first study aims to
answer questions concerning the degree of similarity between perceived and objective
measures of crime and neighborhood disorder, and what factors could possibly account
for any discordance. The second study builds on the notion of differences between
assessment methods by examining the differential effects of the perceived and objective
neighborhood environment on active commuting to school by children. It also seeks to
identify combinations of neighborhood factors that together may be more than the sum of
their parts. The final study incorporates individual psychological factor with
environmental perceptions and physical activity data in a longitudinal framework. This
will help understand causal ordering and mediation effects, both important when
attempting to develop interventions or policy changes.
26

CHAPTER 2: CORRELATES OF DISCORDANCE BETWEEN PERCEIVED
AND OBJECTIVE MEASURES OF CRIME AND PHYSICAL DISORDER
Abstract
Background and objectives: Neighborhood environments can be characterized through
objective and perceived measures. Prior research has shown that measures of structural
features do not necessarily agree. Part of this disagreement may be due to interpersonal
and demographic characteristics of the individual. However, it is not clear if these same
relationships apply to less concrete aspects such as crime and physical disorder. The
objective of this paper is to examine whether demographic and neighborhood social
characteristics are correlated with discordance between perceived and objective measures
of crime and disorder
Methods: Secondary analysis of a 1390-respondent dataset. Demographics, perceived
levels of crime, physical disorder, and multiple aspects of the social environment were
self-reported by participants. Objective measures of crime incidence and code violations
were obtained from the City of Los Angeles. Multi-level models were run with perceived
crime or disorder as the outcome, and demographics and social environment variables as
predictors, stratified by objective levels of crime and disorder.
Results: Social control was consistently associated with perceiving the neighborhood
more positively than objective statistics indicated. Hispanic ethnicity, time spent in the
neighborhood, and acculturation were associated with perceiving the neighborhood more
negatively than objective data indicated.
27

Conclusions: Certain demographic sub-groups may be good targets for interventions to
change overly negative perceptions, possibly by facilitating a more positive social
environment. This may in turn influence behaviors such as physical activity. However,
the possibility that official statistics are biased should be investigated fully before
attempting to intervene at the individual level.

28

Introduction
As noted earlier, several studies have sought to examine correlates of mismatch between
measures of the perceived and objective neighborhood environments. In a 2009 study by
Gebel et al, five major features of the environment were considered, including overall
walkability, density, connectivity, land use mix, and retail density. They found that
adults of lower socioeconomic status, those who are overweight, and who walk less in
their neighborhood were more likely to misperceive their objectively determined high
walkable neighborhood as low walkable. A study of adult women by Ball et al examined
perceived and objective mismatch in terms of access to physical activity facilities.
Women who were in the oldest or youngest age groups and reported low income, self-
efficacy, activity levels, and resided for less time in their neighborhood were more likely
to be unaware of a local physical activity facility when one did in fact exist
7
.
Within the criminology and sociology fields, some studies have similarly examined the
degree of discordance between subjective and objective methods of assessing crime and
quality of life. Warr has previously found the relationship between objective and
perceived incidence of crime to be non-linear
135
. People overestimated the frequency of
rare crimes such as murder and rape, and underestimated the frequency of less-serious
crimes. This could be because individuals may judge frequency of events based on how
easily they can recall them. Ease of recollection may in turn be a function of
sensationalist media coverage of statistically unlikely but dramatic crime events like
murder or kidnappings. Lee and Marans compared self-reported feelings of safety with
objective crime rates
64
. The correlation between the measures indicated a fair amount of
29

mismatch. One reason for this was what the authors termed “scale discordance”, where
the frame of reference for an individual’s self-reported perception of safety is smaller in
size than the area captured by official statistics, thus leading to discordance. They further
found that blacks living in a high crime area were more likely to report strong feelings of
safety as compared to whites. This finding may be explained by the fact that
discrimination and economic factors make it more difficult for minorities to leave high
crime areas than whites. The result is that over the long-term, minorities adapt or adjust
to high crime situations and their feelings of safety increase, in spite of the objectively
assessed level of crime. In Chicago, Lewis and Maxfield compared objective crime rates
to resident reported perceived risk of victimization and found a lack of agreement
between the two
69
. Further analysis indicated that perceptions of incivilities, such as
loitering teenagers, graffiti, and abandoned buildings, modified the relationship between
objective crime and fear of crime. High perceived incivilities heightened the effect of
objective crime on fear, but lower perceived incivilities dampened the same effect. They
concluded that reducing neighborhood incivilities should be just as high a priority as
reducing objective crime in order to reduce concern and fear.
Several studies examining discordance between subjective and objective indicators of
urban quality of life (e.g. aesthetics, security, educational system, housing conditions,
etc.) have also found weak to modest correlations between the two methods
3, 80
. While
not formally testing these theories, the authors have speculated that the weak
relationships may be due to multiple factors, including aspiration levels. Residents with
high aspirations and expectations will likely report low satisfaction if they live in
30

objectively poor areas, whereas those with lower levels of aspiration will have relatively
higher satisfaction in the same conditions. These aspirations may be informed by prior
living conditions or comparison of their living conditions with other people whom the
resident perceives to be similar to them
3
.
The major implications of these studies is first that perceived and objective measures are
not simply interchangeable; each captures a unique dimension of the environment, and
researchers should consider the goals of their analysis before settling on a data collection
method. Second, there may be opportunities to intervene on certain personal and
environmental factors so that a greater balance is achieved between objective and
perceived indicators of the neighborhood; this balance may then induce greater
neighborhood-based physical activity.
A primary limitation of these studies is the range of variables considered as predictors of
mismatch. The quality of life studies did not formally examine any predictors for the low
correlations they observed. Lee and Marans considered scale of measures and race. In the
Gebel study, predictors included basic demographics, BMI, typical walking behavior, and
several psychosocial variables, such as self-efficacy and enjoyment with respect to
physical activity. The Ball study used very similar predictors. Although these are
important variables to consider, other variables may further explain divergences between
perceived and objective measures of the neighborhood. For example, many of the social
environment variables previously discussed (e.g. social cohesion or collective efficacy)
may explain why some people report perceptions of crime or disorder that are at odds
31

with what is found from objective measures. Consistent with this, some authors have
proposed conceptual models in which the neighborhood social environment is one of the
proximal determinants of perceptions of crime and disorder. Kamphuis et al found that
social cohesion was correlated with both perceptions of neighborhood attractiveness and
safety
57
. By examining these specific social factors, a better understanding of the
explanations for discordance regarding accommodation, aspiration and comparison may
be gained.
Research and Policy Implications
Ultimately, exploring these additional predictors as correlates of possible discordance in
other neighborhood characteristics is important for the following reasons. First, it would
allow us to see whether disorder and crime follow the same patterns as more concrete
characteristics, and in particular the type of discordance where residents perceive their
neighborhood as less favorable to activity than it really is. If these “negative perceivers”
exist, then it would raise questions about the suitability of using only objective measures
to assess the neighborhood in terms of disorder and crime. Though areas may not appear
problematic based on GIS or audit data, there nonetheless may be perceived issues related
to these factors that affect people’s proclivity to engage in activity.
Looking beyond just physical activity, this also has important implications for code
enforcement and police resource allocation decisions. If, for example, police patrol
routes are made solely on the basis of reported crime data, they may be missing hot spots
of crime or disorderly behavior in areas where people are less inclined to report these
32

lower level offenses, either because they have come to accept it as part of neighborhood
life or because of a lack of trust in, or negative interactions with, police. In this study,
these areas would be represented by high perceived crime/disorder and low objective
equivalents. Identifying social environment or demographic characteristics associated
with this type of mismatch could perhaps supplement the use of objective data to make
these resource decisions
64
.
Second, some demographic subgroups may be more or less likely to have discordant data.
Both the Gebel and Ball papers were conducted in Australia with relatively homogeneous
populations. A more diverse population that includes racial minorities, recent
immigrants, and those who are less acculturated could uncover whether certain groups
are likely to misperceive their neighborhood as unfavorable for activity. This would call
out for strategies to modify their negative perceptions; the further segmented the data, the
more targeted the interventions could be. It may also be the case that some groups
perceive their objectively determined unfavorable neighborhood as favorable, as
suggested by some of the criminology literature. Research exploring why these groups
are “positively misperceiving” their environment could identify community or individual
level characteristics that could be used as interventions in areas where the objective
environment is less supportive for activity.
Third, if certain indicators of the social environment prove to be associated with any
identified mismatch between the perceived and objective measures, it would provide
stronger evidence that upstream community factors should be targets of interventions to
33

increase neighborhood-based physical activity. Existing proposals for environmental
change to facilitate physical activity are heavy on modifications to the structure of the
neighborhood by, for example, installing more sidewalks, increasing the availability of a
variety of shops and services, or increasing access to public transit
107
. However, it may
be more feasible to work through the social aspect of the neighborhood, such as
increasing participation in community groups, and fostering awareness and trust among
neighbors. The same ends may be achieved (increased activity levels), but potentially at
less expense.
Research questions
The first study will be designed to answer the following questions:
1. To what extent are individual’s perceptions of their environment at odds with
objective measures with respect to neighborhood disorder and crime?
2. What individual level factors are correlated with discrepancies between the two
measures of the neighborhood?
Methods
Participants and Setting
This study is a secondary analysis of a longitudinal RAND Corporation dataset collected
with the intent of examining the effect of business improvement districts (BIDs) on
community factors associated with youth violence
44
. Participants, consisting of
adolescents and their parent, were interviewed from either neighborhoods with BIDs or
demographically similar neighborhoods without BIDs. Baseline data collection took
34

place in 2006 and 2007 on a sample of 760 households. An 18-month follow-up survey
was conducted with households who participated at baseline in 2008 and 2009. Data was
collected by study staff either over the phone or in person. Parents were asked to
complete a 110 item survey, while adolescents completed a 90 item survey. Major
content areas of the surveys included perceptions of neighborhood crime and violence,
parent and adolescent relationships, important neighborhood events, and youth
engagement with family, friends, and school, plus sociodemographic questions.
In addition to the above sample, the same surveys were administered with a second,
independent cross-section of parents and adolescents in 2009 and 2010. The same
methodology was used where participants were recruited from BID and non-BID
neighborhoods. This study uses the combined sample of the second wave of parent data
from the initial survey and the parent data from the second, independent cross section
survey, providing a total sample size of 1390.
Measures
Self-reported outcomes
Perceived neighborhood disorder was characterized through a series of five questions
asking whether a given item was a problem in their neighborhood. Items included litter,
graffiti, vacant and poorly maintained homes, and abandoned cars. This captures what
Sampson and Raudenbush have termed “physical disorder”
116
. The outcome of perceived
neighborhood safety was characterized through a question asking how safe it was to walk
around alone after dark on a four-point scale from completely safe to extremely
35

dangerous. Though a somewhat simple measure, it is not unusual in the criminology
literature to use measures of perceived safety similar to this
49, 50
.

Objective outcomes
Objective analogs of the perceived outcomes of neighborhood disorder were purchased
from the City of Los Angeles Department of Building and Safety. The Code
Enforcement Information System maintained by the Department lists all code
enforcement actions per month, and contains details including address of violation, date
case was initiated, and type of violation. This data only represents proactive code
enforcement cases in which inspectors undertake surveys of neighborhoods looking for
violations. Because the data does not include citizen-initiated cases, it reduces bias in the
data that may arise from frivolous complaints lodged by individuals with ulterior
motives. Cases from January 2007- April 2010 were aggregated, geocoded to a census
tract based on addresses in the files, and then collapsed at the tract level, so that each tract
has one number representing total code enforcement actions over the time period.
Analogs to the measures of perceived crime outcomes were obtained from the Los
Angeles Police Department. The crime statistics are counts of homicide, rape, robbery,
aggravated assault, auto theft, burglary, personal theft, and other types of theft. Data
from 2006-2008 are averaged to create an overall count of these crimes per census tract.
Using the United States Federal Bureau of Investigation’s (FBI) classification scheme the
crimes were subsequently categorized as either violent (homicide, rape, robbery and
aggravated assault) or non-violent property crimes (auto theft, burglary, personal and
36

other types of theft) to create a variable representing average violent and property crime
in the tracts
38
. Note that this does not adjust the crime statistics for population size in the
tract, as it is thought that the absolute amount of crime, and not a relative risk of crime, is
a more important driver of behavior
108
.
Correlates of discordance
One category of potential correlates includes individual sociodemographic
characteristics, consisting of the following measures: race/ethnicity; age; gender;
household income; employment status; whether they own or rent their home; length of
residence at current home; and several measures broadly related to immigration and
acculturation, including where participants were born and what languages they use at
home.
The second category of potential correlates relate to perceptions of the neighborhood, and
the social environment in particular. Informal social control was measured through a
series of 6 questions asking how likely it was that neighbors would take action in
response to community problems, and social cohesion was assessed by asking whether
people agreed or disagreed to statements about bonds among neighbors and willingness
to help each other
117
. Also assessed were items to get at an individual’s involvement in
terms of dealing with neighborhood problems, such as attending community groups or
meetings, and discussion with neighbors.
Neighborhood attachment was measured by three questions asking how attached
participants are to their neighborhood, how many adults they recognize and how many of
37

their friends live in their neighborhood. Questions were asked about whether any major
personal life events in the household or major events within the community had occurred
within the last year. Examples of the former include divorce, serious illness, and
unemployment for more than a month. Examples of the latter include closing or opening
of a school, major employer or recreational space.
Analytic Plan
Ideally, because a number of the predictors are measured through multiple items, we
would specify this as a latent variable structural equation model to control for
measurement error in the constructs. However, the complexities of a multi-level model
with a dichotomous outcome made that unfeasible in the context of this dataset.
Therefore, we used factor scores of the multi-item constructs derived from a confirmatory
factor analysis (CFA). Though this method is inferior to a truly latent variable model, it
is superior to using a simple average of items, because individual items’ contribution to
the factor score are weighted according to their loading derived from the CFA. The CFA
also insures variables are loading onto each hypothesized construct, checks for cross
loadings among variables, and reveals the degree of correlation between them. Overall
CFA fit will be assessed by examining the comparative fit index (CFI), root mean square
error of approximation (RMSEA), and 
2
. Well-fitting models should have a value of
approximately 0.95 on the CFI, and the RMSEA should be less than 0.06
52
. Ideally, the 
2

p-value should be non-significant, but in large samples it will almost always be
significant
60
. All models were fit in Mplus 6.12 using robust maximum likelihood (MLR)
estimation with Monte Carlo integration.
38

In order to analyze discrepancies in perceived and objective measures of crime, the
measure of perceived safety was dichotomized such that those who said their
neighborhood was completely or fairly safe were coded as “safe”, and those who said it
was somewhat or extremely dangerous were coded as “dangerous”. This dichotomous
measure is the ultimate outcome in the analysis. The objective crime data was then
divided into quartiles. This allowed us to specify separate structural models, with one
restricted to individuals who reside in census tracts in the highest quartile of crime, yet
report their neighborhood as being safe, and another with individuals who reside in
census tracts classified in the lowest quartile of crime, yet perceive their neighborhood to
be dangerous. Both models were run twice, once for property crimes and then for violent
crimes, for a total of four separate models. In order to control for clustering at the census
tract level, multi-level models with random intercepts only were specified. Since the
outcome is dichotomous, odds ratios and 95% confidence intervals will be reported. Note
that in the context of multi-level structural models with limited dependent variables
estimated using MLR, measures of model fit are not available.
For the objective physical disorder data, due to the large number of tracts with no
documented code violations, a decision was made to classify census tracts as those with
zero violations and those with one or more violation. Perceived physical disorder was
characterized through a latent factor measured by the variables previously described.
Using this latent variable as the outcome, multi-level structural models were estimated
separately for the two categories of code violations noted at the tract level (0 v. >0). As
the latent variable was continuous, model results are presented as linear regression
39

parameter estimates. Again, no fit statistics are available because the indicators of
perceived disorder are modeled as ordinal variables using the MLR estimator.
Results
Approximately 71% of participants were female, 56% were of Hispanic or Latino
descent, 34% were white non-Hispanic, average age was 48 (SD=8.3), 58% were not
born in the U.S., average tenure in the neighborhood was 15 years (SD=9.9), 51% owned
their residence, median income was between $30,000 and $40,000, 55% had at least
some college education, and 67% were employed at least part-time. Overall, 70% of
participants reported their neighborhood to be either completely or fairly safe.
The results of the CFA for the predictors measured by multiple items are presented in
table 1. This model showed good fit to the data (CFI=0.93; RMSEA=0.031 (90%
CI=0.029-0.034); 
2
=990.060, DF=419, p<0.0001). All factor loadings were at least
moderately sized and significant. Therefore, it was acceptable to move forward with
producing factor scores for the multi-item constructs.
The results of the crime structural models are displayed in table 2. The first two models
concern objective measures of violent crime. For the first model predicting subjectively
unsafe environment in an objectively safe census tract, one socioeconomic status variable
was significant. As expected time at home increases due to a lack of employment outside
the home, the odds of perceiving the environment as unsafe increases by 33%.
40

The second model predicts perceiving the neighborhood as safe among those who live in
the objectively least safe census tracts. Significant correlates here are social control and
only speaking English at home. For the former, increasing levels are associated with
greater odds of perceiving the neighborhood as safe, while the latter is associated with
lower odds of perceiving the neighborhood as safe.
The third and fourth models concern non-violent property crime. For the third model,
which predicts subjective feelings of danger in an objectively safe census tract, social
control and household income were both significantly associated such that increasing
levels were correlated with lower odds of perceiving the environment as dangerous.
Greater number of years lived in the neighborhood was associated with greater odds of
perceiving the neighborhood as dangerous.
Finally, for the fourth model, which predicts perceiving the environment as safe in an
objectively unsafe census tract, being Hispanic was significantly associated with lower
odds of perceiving the area as safe.
The results of the disorder structural models are displayed in table 3. The first model
concerns perceptions of disorder among those who live in census tracts with no
documented code violations from 2007-2010. Greater social cohesion and control were
significantly associated with perceiving less physical disorder in the neighborhood, while
Hispanics perceived greater levels of disorder compared to non-Hispanics. Those who
speak only English at home also perceived greater disorder than those who speak other
languages at home.
41

The second model concerns perceptions of lack of physical disorder among those who
live in a census tract with at least one documented code violation. Greater social control
was associated with perceiving less disorder, while those who are Hispanic, not born in
the U.S., and who speak only English at home perceived lower levels of lack of disorder.
That is, they were more likely to report that the indicators of disorder were to some
degree a problem.
Discussion
The goal of this analysis was to extend to new areas the important work done recently
seeking to understand why objective and subjective assessments of the environment can
be in conflict. Several demographic and social environment factors were found to be
correlated with a lack of concordance between official statistics of crime and disorder and
resident perceptions of the same.
Prior studies of this nature have tended to focus on large, discrete elements where the
truth is relatively easy to uncover, such as access to recreational centers. Crucially, we
have examined aspects of the environment where the truth is unlikely to ever be known.
As emphasized earlier, the objective and subjective measurements are probably on
different ends of a subjectivity continuum. An understanding that neither measure is
absolutely correct is important because it colors our interpretation of the results. Unlike
prior work, we cannot make absolute claims about the need to bring perceptions in line
with reality, since it could easily be the reverse: the objective statistics may in fact be a
poor facsimile of reality. This last point is especially complicating because it implies that
42

there are always two ways to interpret the data: either the individual has a biased view of
the reality best represented by official data, or that the official data is biased and
perceptions most faithfully represent reality.
With that stated, the results can be broadly classified as factors associated with a more
positive view of crime and/or disorder, and those associated with a more negative view.
This means that across models for a given factor, in some cases, the objective and
perceived measures will be discordant, and in other cases they will agree. Social control
is the primary example of the consistently positive view. As perceived social control
increases, individuals are more likely to perceive a higher violent crime area as safe, and
less likely to perceive a lower property crime area as unsafe. Similar relationships were
observed with perceived and objective measures of physical disorder. Less perceived
social control was associated with perceiving more physical disorder in areas with no
code violations, while more control was associated with perceiving less disorder in areas
with violations.
Social control represents the willingness of community members to act in the face of
perceived problems. We have shown that this characteristic is cross-sectionally
associated with perceiving fewer issues with crime and disorder. Because this
relationship holds regardless of what the objective statistics indicate, it does not appear
that this association can be explained by a situation where the community has acted to
eliminate all problems, which in turn cause lowered perceptions of community problems.
43

Rather, it may be the sense that the community will act if necessary that provides a
protective effect in terms of seeing problems.
It is interesting that of the factors that bear on the relationship between the individual and
their neighborhood (involvement, attachment, events, social cohesion and control), the
only significant correlates were cohesion and control, both of which use the
neighborhood as the referent, rather than the individual respondent. Thus, contextual
factors that affect all members of a group may be more important to consider than
individual level factors. Of course, we were unable to treat cohesion and control as truly
contextual effects (i.e. as level-2 predictors in the multi-level model) due to model
convergence problems, so this is an area for future exploration.
Ethnicity is an example of the more pessimistic view of the environment. For the
property crime model, Hispanics were less likely to perceive their objectively unsafe
neighborhood as safe. For disorder, they perceived more disorder in their neighborhood
with documented code violations. This means their objective and subjective assessment
were more congruent than their non-Hispanic counterparts. In contrast, among those with
no code violations in their census tract, Hispanics perceived higher levels of disorder,
meaning they were viewing their neighborhood in a potentially adverse manner, at least
insofar as perceptions of disorder influence physical activity.
Another consistently negative finding concerned those who speak only English at home.
This can be taken as a measure of acculturation. Households speaking exclusively
English are considered to be more acculturated than those that speak other languages in
44

the home. Respondents from English-exclusive homes had more congruent measures of
violent crime and disorder among those from census tracts with documented code
violations, while those from tracts with no violations perceived higher levels of disorder,
meaning the two assessments were not congruent.
Part of the reason for the relationship of the measure of acculturation with perceptions
may have to do with the respondents’ frame of reference. Those who are less
acculturated may be coming from a background in which a certain amount of crime and
disorder is the norm. While by objective standards we may see their current
neighborhood as disordered or dangerous, they may consider their current neighborhood
the most ordered, safe place they have lived in.
Sampson and Raudenbush and Hipp have previously found that non-whites perceive less
crime and disorder than whites who live in the same general area
49, 116
. Their explanation
for this is similar to the norms and point of comparison explanations noted above for
acculturation. It is unclear why we have found the opposite effect here. One explanation
is that what they observed was actually an effect of acculturation on disorder; had they
included a measure of it, the effect of Hispanic ethnicity may have been attenuated. Hipp
argued that aggregating residents to a relatively large cluster such as a census tract as we
have done leads to incorrect assumptions about the uniformity of crime and disorder
across households, thus distorting effect sizes and even direction
49
. Thus, we could
merely be seeing a methodological artifact.
45

A more involved explanation relates to our earlier point that so-called objective data
should never be mistaken for the truth. It is entirely possible that the culprit is the official
statistics concerning code violations. To the extent that they do not capture the true level
of disorder in the tract, resident perceptions will be artificially disjointed from them. If
the truth is that Hispanics live in the most disordered census tracts, their perceptions then
would actually be the most congruent with reality, implying there is little to no individual
level bias in their perceptions of disorder. This example serves to highlight the point that
we should never automatically assume that it is some individual characteristic that is the
cause of “faulty” perceptions. The real fault may lie with what we are calling objective
data.
With this stated, our results and that of Sampson and Raudenbush and Hipp are not so
different; they may merely reflect different ends of the mechanism linking official
statistics with perceptions. If those authors are correct that racial/ethnic minorities have a
higher threshold for perceiving crime and disorder as problematic, then it makes it less
likely they will report such things as “routine” levels of litter, graffiti, or petty theft. This
lack of reporting then causes official statistics to be downwardly biased relative to the
truth. Later on, when residents are surveyed, they may still perceive the same level of
crime and disorder, but relative to the distorted official statistics, it appears that they are
overestimating their prevalence, as we have found.
Finally, time spent in the neighborhood appears to be associated with negatively
misperceiving the neighborhood for the crime models. Employment status, coded so that
46

a stay at home parent has a higher value than someone employed full-time, as well as
number of years lived in the neighborhood, were associated with perceiving a safe
neighborhood as unsafe. This agrees with what is thought about the effect of time on
misperceiving other aspects of the neighborhood. Theoretically, the more time one
spends in the neighborhood, the more aware they become of their surroundings. This can
be beneficial, because it may make someone more aware of recreational spaces they may
not otherwise know are available to them, but it may also make them more aware of
activity hindering aspects, such as crime or physical disorder. Small-scale interventions,
such as walking groups, may provide an opportunity to “introduce” residents to the
activity promoting aspects of their neighborhood, but it should be understood that in
doing so they may subsequently notice more of the negative aspects as well.
Once the lack of truth in either perceived or objective measures is acknowledged, the
question becomes how to best balance the two sources of information for purposes of
developing policy or other interventions. Given the relatively consistent associations of
social control with crime and disorder, this construct may be the most promising to target.
Developing an environment conducive to resident action may be a positive policy-level
action to increase social control. Comfortable relationships among residents, police, and
code enforcement officials may increase the likelihood of truancy, graffiti, etc. being
reported. Importantly though, this will likely require consistent official intervention.
After all, if residents do not see public officials responding to complaints, they may lose
the desire to continue doing so.
47

Social control can also be promoted by strengthening community organizations that are
inclusive and visible to all residents. This last part is important, since as discussed, it
does not appear to be the individuals own involvement that leads to positive perceptions,
but their sense that the larger community will take action.
However, the potential for greater social control to provide a protective effect in a poor
environment must be balanced with the data suggesting some degree of increased risk,
especially in terms of crime data. It would be undesirable to modify perceptions to the
point where individuals who live in an area with a higher risk of victimization cease
taking appropriate protective measures due to a dampened perceived sense of risk.
Interventions designed to lower activity-hindering levels of perceived crime should
simultaneously convey sensible methods to reduce victimization. This might include
teaching participants to avoid walking alone, using high value objects in plain sight
(phones, designer purses), and not using headphones so as to maintain situational
awareness. In fact it is reasonable to believe that these strategies could be the
intervention itself, as it may lead participants to believe they are taking measures that will
reduce the risk of victimization, thus lowering their overall perception of the
neighborhood as an unsafe place
Time spent in the neighborhood is negatively correlated with perceptions of safety in the
safest neighborhoods. At first glance, it does not appear that efforts to reduce the crime
rate through policy mechanisms, such a greater police resources, will have much effect
since there is relatively little crime to begin with. Again though, it must be emphasized
48

that this could simply be due to a lack of reporting of crime. Greater investigation would
be needed to understand if there truly is a crime problem that can and should be
addressed by changing police staffing or patrol patterns. If this is not the case, then
changing perceptions of individuals who spend a great deal of time in the area would be a
possible intervention strategy. This is particularly true for individuals in low violent
crime areas. Educational materials and programs to emphasize just how infrequent these
felonies are could reduce the perceived lack of safety. Since all participants in this study
have children or are guardians of children, they may be incorporating fear for their
child’s safety into their own response. If so, actions to protect children in the
neighborhood, such as supervised routes to school, after-school activities, or better street
lighting at night, may increase their parent’s perceptions of overall safety.
Strengths and limitations
This analysis is strengthened by the relatively large sample size, the use of multiple
indicators for the predictor variables, and the incorporation of many social environment
variables not previously examined in other studies. Also, the geographically and
demographically diverse areas from which the sample was drawn makes it more likely
that these results can generalize to other urban contexts around the country.
There are several limitations to this analysis. First, the objective data is collected at the
census tract level, while the subjective data is at the individual level. While this data
structure is accounted for in the multilevel analysis, it is not ideal because the frame of
reference for the two methods is not necessarily the same, a problem noted in the Lee and
49

Marans and Hipp studies. It is likely that most individuals consider their neighborhood to
be a smaller area than the area of a typical census tract. Nonetheless, this is certainly a
better data structure than attempting to use even higher aggregations of crime statistics as
some studies do
109
. Also, given the cross-sectional nature of the data, we can only speak
about the results in terms of correlations, and cannot prove causal relationships.
Conclusions
Official statistics and perceptions of a given area will not always agree. There are
multiple reasons for this, including, as we have found here, social characteristics of the
area and the demographics of residents. If it is believed that crime and physical disorder
affect behavior, including physical activity, these sometimes dueling views of the
neighborhood need to be reconciled. In some cases, this may lead to the discovery that
the official statistics are underreporting the situation, and in other cases resident
perceptions are distorted. Options are available to address both possibilities. While this
complicates attempts to influence downstream behaviors, it may increase the chance that
we will address the underlying cause of behavior, resulting in more permanent, positive
change.

50

Study One Figures
Table 1.1: Confirmatory factor analysis results
Latent factor Manifest variable Unstandardized
loading (Standard
error)
Neighborhood
attachment
How many adults do you recognize 1.0 (0.00)
How attached do you feel -1.342 ( 0.095)
How many of your friends live in your
neighborhood
0.98 (0.067)
Life events Death in family 1.0 (0.00)
Serious illness 1.697 (0.489)
Moved/changed residence 0.559 (0.117)
Period of unemployment 0.492 (0.124)
Birth of a child 0.466 (0.124)
Neighborhood events Closing/opening of a school 1.0 (0.00)
Closing/opening of a major employer 0.556 (0.137)
Closing/opening of a recreational space 0.700 (0.198)
Major road work or construction 0.540 (0.186)
Other 0.741 (0.198)
Neighborhood
involvement
Participate in local groups 1.0 (0.00)
Get together with neighbors & discuss
community concerns
0.173 (0.025)
Attend community meetings 0.911 (0.126)
Social cohesion People willing to help neighbors 1.0 (0.00)
Close-knit neighborhood 1.095 (0.058)
People can be trusted 0.977 (0.047)
People don’t get along -0.591 (0.055)
People don’t share same values -0.392 (0.060)
Parents know children’s friends 0.963 (0.052)
Adults know local children 0.724 (0.057)
Parents know each other 0.938 (0.051)
People willing to do favors like watch
children or home
1.095 (0.049)
Social control Act if children skipping school 1.0 (0.00)
Act if children spray-painting 1.153 (0.049)
Act if children show disrespect 0.890 (0.038)
Act if fight broke out in public 1.054 (0.045)
Act if youth gang was selling
drugs/intimidating people
1.170 (0.050)
Act if local school threatened with
closure due to budget
0.575 (0.048)

51

Table 1.2: Crime model results
Odds ratios (95% Confidence interval)

Violent crime Property crime
Strata
Objectively safe Objectively
unsafe
Objectively safe Objectively
unsafe
Independent
variable
Subjectively unsafe
N=346; T=79
Subjectively safe
N=337; T=98
Subjectively
unsafe
N=343; T=85
Subjectively
safe
N=335;
T=96
Neighborhood
attachment
1.35 (0.02-113.38) 1.23 (0.11-13.5) 0.15 (0.02-1.46) 6.94 (0.76-
63.03)
Life events 1.65 (0.83-3.27) 1.22 (0.78-1.92) 1.06(0.77-1.46) 1.11 (0.76-
1.60)
Neighborhood
events
0.62 (0.23 -1.67) 0.95 (0.53-1.68) 0.74 (0.37-1.46) 0.93 (0.54-
1.62)
Neighborhood
involvement
0.96 (0.55-1.66) 1.12 (0.81-1.55) 0.86 (0.61-1.22) 1.17 (0.87-
1.59)
Social cohesion 0.12 (0.00-6.41) 6.55 (0.73-59.19) 1.11 (0.16-7.86) 1.35 (0.17-
10.83)
Social control 0.47 (0.1-2.19) 2.67 (1.15-6.19)* 0.42 (0.21-0.83)* 1.70 (0.84-
3.45)
Hispanic 2.55 (0.45-14.62) 0.39 (0.13-1.14) 2.10 (0.73-6.04) 0.51 (0.26-
0.98)*
Not born in U.S. 0.30 (0.03-2.81) 1.8 (0.69-4.7) 0.82 (0.31-2.16) 1.62 (0.91-
2.86)
Household
income
0.89 (0.75-1.05) 1.18 (0.99-1.41) 0.89 (0.79-0.99)* 1.12 (0.99-
1.27)
Employment
status
1.33 (1.02-1.73)* 0.99 (0.83-1.17) 1.17 (0.97-1.42) 0.98 (0.83-
1.16)
Rent home 1.18 (0.3-4.62) 1.16 (0.54-2.52) 1.49 (0.62-3.6) 0.81 (0.46-
1.44)
Age 1.01 (0.89-1.15) 1.04 (0.98-1.09) 1.01 (0.97-1.06) 0.98 (0.95-
1.02)
Years lived in
neighborhood
1.03 (0.96-1.10) 1.01 (0.97-1.06) 1.05 (1.02-1.08)* 1.0 (0.96-
1.03)
Female 1.6 (0.55-4.67) 1.69 (0.78-3.67) 1.66 (0.82-3.37) 0.84 (0.46-
1.55)
Only speak
English at home
0.61 (0.14-2.69) 0.33 (0.11-1.01)* 0.73 (0.32-1.62) 0.74 (0.37-
1.47)
Note: N=number of respondents; T= number of census tracts; *p<=0.05

52

Table 1.3: Physical disorder model results
Parameter estimates
(standard error)
Strata No objective
disorder
Objective
disorder
Independent
variable
Subjective
disorder
N=825; T=212
Lack of
subjective
disorder
N=525; T=144
Neighborhood
attachment
0.228 (0.983) 1.087 (0.958)
Life events -0.239 ( 0.140) 0.154 (0.165)
Neighborhood
events
-0.120 (0.236) 0.250 (0.262)
Neighborhood
involvement
-0.139 ( 0.119) 0.055 (0.131)
Social cohesion -1.643 (0.854)* 0.663 (0.830)
Social control -0.857 (0.276)* 1.402 (0.331)*
Hispanic 1.590 (0.327)* -0.780 ( 0.362)*
Not born in U.S. 0.555 (0.343) -0.931 (0.342)*
Household income -0.062 (0.036) 0.074 (0.055)
Employment
status
0.058 (0.055) -0.071 (0.071)
Rent home 0.362 (0.283) -0.112 (0.272)
Age 0.008 (0.014) 0.022 (0.018)
Years lived in
neighborhood
0.020 (0.012) -0.014 (0.013)
Female 0.409 (0.230) 0.167 (0.298)
Only speak English
at home
0.766 ( 0.281)* -0.570 (0.288)*
Note: N=number of respondents; T= number of census
tracts; *p<=0.05

53

CHAPTER 3: PERCEIVED AND OBJECTIVE NEIGHBORHOOD
PROFILES AND THEIR ASSOCIATION WITH ACTIVE COMMUTING TO
SCHOOL
Abstract
Background and objectives: Active commuting to school is a potentially important way
to increase physical activity among children. Features of the built environment may
support or hinder the ability to actively commute, but existing research shows conflicting
findings about how individual factors impact this behavior. This may be due to a failure
to simultaneously consider the effects of multiple features of the environment. The
objectives of this study are to develop mutually exclusive neighborhood typologies using
both perceived and objective environmental assessments, and relate these typologies to
active commuting.
Methods: Data from 278 families participating in the Healthy Places trial was used.
Parent perceptions of the built environment were assessed by the Neighborhood
Environment Walkability Survey. The Irvine Minnesota Inventory was used to
objectively assess participants’ neighborhoods. Separate latent profile analyses were
completed for each of the two assessment methods, including active commuting as an
indicator.
Results: By both assessment methods, a profile characterized by moderate levels on all
environmental variables was associated with the highest probability of actively
commuting to school, while a profile characterized by low-density single-family
residential development was associated with the lowest probability. More than twice as
54

many individuals where classified in the latter profile by the objective method than by the
subjective method.
Conclusions: Moderate levels of many environmental aspects may provide the most
supportive environment for active commuting. This implies that there are feasible
opportunities to modify existing neighborhoods and plan future developments to promote
active commuting. Changing only perceptions of the environment may also be a way to
encourage this behavior.

55

Introduction
Research exploring correlates of active commuting has grown considerably in the last
several years because active commuting is seen as a potentially important way to
integrate physical activity in children’s lives
12, 24, 127
. However, the evidence is not clear
about what exact factors are associated with the behavior, nor is it clear whether it is the
perceptions of the child’s parent or the objective neighborhood that is the more important
determinant
27, 100
. The discrepant findings may be because neighborhood features tend to
be considered independently while partitioning out the effect of other environmental
influences. For example, a statistical model may examine the impact of sidewalks on
active commuting, while controlling for aesthetics and traffic safety. While this can help
isolate the influence of one neighborhood feature, it is not reflective of the way parents or
children actually experience their neighborhood. The simultaneous influence of multiple
features needs to be considered to understand the true way neighborhood design can
impact behavior. As noted, there may be a multiplicative relationship among diverse
features when combined, an effect that is lost in traditional regression modeling. The
second study will build on existing knowledge and previously identified research gaps by
constructing unique neighborhood typologies composed of multiple features that can be
correlated with measures of active commuting to school by children. It will also examine
how typologies based on perceptions of the neighborhood differ from those constructed
with more objective measures. This will aid in our understanding of whether one
assessment method is more strongly associated with active commuting.
56

While the first proposed study focuses on two very specific elements of the neighborhood
(crime and disorder), and the discordance between measurement methods, the second
study will expand the number and type of neighborhood characteristics that are assessed
objectively and subjectively. Also, the second study extends the ideas in the first by
examining not simply what explains discrepancy in measurement methods, but how those
methods may differentially impact a meaningful outcome, in this case active commuting
to school.
Research and Policy Implications
Though this analysis cannot definitively answer whether objective or perceived measures
of the neighborhood are the primary drivers of physical activity, it could help shape the
nature of interventions designed to increase active commuting to school. The results may
imply that some features of the neighborhood could be better addressed through structural
changes assessed objectively (e.g. improving street crossings) on the basis of their
strength of association with active commuting. Other, more subjective aspects may be
better modified through programmatic activities, such as reshaping parental fear of crime
to be more in line with reality. Still other aspects may be clearly linked to both perceived
and objective measures, such that modifications to both are necessary. Also, the process
of identifying neighborhood profiles will potentially help policy makers decide what
combination of changes to the environment may provide the greatest return as far as
improving physical activity. Instead of implementing changes piece-meal, planners may
use this knowledge to direct funding towards several smaller improvements that together
have a bigger impact than any one of them alone.
57

Research questions
Study two will be designed to answer the following questions:
1. With respect to features of the environment that may promote physical activity,
what unique neighborhood typologies can be constructed on the basis of both
perceived and objective assessments of those features?
2. Do typologies differ by objective or perceived assessment methods?
3. How does the probability of active commuting to school vary as a function of
membership in a particular profile?
Methods
Participants and Setting
Healthy PLACES (Promoting Livable Active Community Environments) is a four year,
controlled trial that has enrolled families from neighborhoods in southern California
representing a diverse range of urban characteristics, including street layouts, density,
housing types and land use mixture. The goal of the study is to understand how features
of these neighborhoods influence behaviors such as physical activity and eating. In this
analysis, baseline data from 278 families is used.
Families in eligible cities were recruited by informational fliers distributed at schools,
community events and shopping centers, advertisements in local papers, postcards mailed
to homes, and door-to-door recruitment by study staff. Interested parents were asked to
call the study office, where there completed a brief eligibility screen. To be eligible,
families must reside in one of several cities in San Bernardino County, CA; have a child
58

in 3rd
th
-8
th
grade, and a household income of less than $165,000. One parent and one
child from each family were enrolled. Informed consent was obtained from participants,
and the study was approved by the University of Southern California Institutional Review
Board.
Measures
Data collection took place during scheduled appointments at local community centers, or
if needed, individual home visits. At the first visit, participants were given electronic
equipment (accelerometers and GPS trackers) along with instructions for when and how
to wear them and had their height and weight measured. At a follow-up appointment one
week later, participants returned equipment and filled out several paper surveys,
including questions about their neighborhood, psychosocial constructs, friends and social
network, demographics, day-to-day activities, physical activity, and diet. Upon
successful completion, each family was compensated $100 for their time.
The Neighborhood Environment Walkability Scale was used to measure perceptions of
the neighborhood environment. The NEWS is a self-report instrument measuring
perceptions of 14 neighborhood factors: residential density, land use mix-diversity, land
use mix-access, street connectivity, infrastructure and safety for walking, aesthetics,
traffic hazards, crime, lack of parking, lack of cul-de-sacs, hilliness, physical barriers,
walkways connecting cul-de-sacs, and social interaction while walking. All but the last
six factors are composites of multiple questions. For most questions, response choices
are on a four point scale, with one being “strongly disagree” and four being “strongly
59

agree.” The NEWS has previously been validated and shown to be reliable in adults
19,
114
.
The objective neighborhood was assessed using the Irvine-Minnesota Inventory (IMI)
28
.
The IMI is a neighborhood audit instrument designed to capture features of the micro-
scale street environment. This level of specificity is important in order to maintain
comparability with the NEWS, which measures the neighborhood at a similarly fine-
grained unit of analysis. Alternative objective methods, including other audit instruments
and GIS databases, do not have the capacity to assess micro-scale features such as types
of traffic or pedestrian signal systems, traffic calming devices, or the presence of
abandoned buildings. Using an established rating scheme, trained auditors evaluated
multiple street segments in a 500 meter network buffer around each participant’s home.
For purposes of the audit, this constituted their neighborhood. Buffers were drawn
around each participant’s home, and three randomly selected street segments were
audited. For each element of the IMI, scores were averaged across the three segments to
create an overall score for that element within a given buffer. Reliability of the IMI has
previously been shown
13
.
Individual items that best exemplified 11 physical environment features linked to active
commuting in prior studies were selected from the two instruments. The features were
prevalence of detached single family housing; land use mix diversity; presence of
sidewalks; condition of sidewalks; presence of crosswalks; how interesting and attractive
the neighborhood is; traffic volume and speed; presence of cul-de-sacs; and flatness of
60

the land. Note that composite factors from the NEWS could not be used because the IMI
lacks a similar factor structure; lack of overlapping factor structure could invalidate
comparisons across perceived and objective analyses. Therefore, individual items were
used instead.
Active commuting behavior was assessed by one item on the child survey asking how
they typically traveled to and from school. Response choices were walking, skating,
biking, car, or bus. By selecting one of the first three choices, the child was classified as
an active commuter. Car or bus riders were classified as non-active commuters.
Distance to a destination is known to exert a considerable effect on the choice of travel
modes. In order to control for the effect of distance between home and school in this
analysis, addresses of home and school for each participant were geocoded, and the
shortest network distance between them was calculated.
Analytic Plan
To determine neighborhood typologies that could be used to predict physical activity, a
technique called latent profile analysis (LPA) will be used. LPA is a person-centered
method to identify homogeneous sub-groups (the profiles) within a sample on the basis of
similar patterns of responses to a series of indicator items
60
. LPA assumes that manifest
indicator variables are a function of an underlying categorical latent variable. This stands
in contrast to how latent variables are typically considered in structural equation
modeling, where the latent variable is assumed to be continuous. On the basis of their
61

measures on the manifest variables, individuals can be categorized into one of the
mutually exclusive categories formed by the LPA
The number of profiles for an LPA is driven by both empirical and theoretical
considerations. For each LPA, the number of profiles tested will begin at one. The
number will be increased until, on the basis of a likelihood ratio test and entropy (a
measure of classification quality), there is no value in adding an additional one.
Information criteria, including Akaike Information Criteria (AIC), Bayseian Information
Criteria (BIC) and sample size adjusted BIC (SS-BIC) will also be used to compare the
fit between x and x+1 profiles. The model with lower values on these criteria provides
the superior fit
60
. Emphasis will be placed on (SS-BIC) which has been shown in a
simulation study to be the superior statistical measure of model fit
91
. While no statistical
significance test is available for information criteria, a general guide is a 5 to 10 point
difference in the information criteria between models indicates better fit
103
. In terms of
theoretical considerations, the goal is to find the point of diminishing returns in adding
profiles. Every additional profile provides a finer grained understanding of the
constituent profiles. At some point though, this added level of specificity does not aid in
understanding or differentiating the profiles from a practical point of view. Thus,
statistical measures of fit should be used carefully along with considerations of
theoretical meaningfulness and interpretability of profiles.
Two series of LPAs will be conducted, since the interest here is on examining whether
unique profiles are determined on the basis of either perceived or objective manifest
62

variables as the input, and whether those profiles are differentially associated with active
commuting. The above procedures will be done first with the NEWS indicators, and then
with the IMI indicators. All statistical analyses were completed in Mplus v. 6.12 using
the robust maximum likelihood estimator. The active commuting variable is modeled
simultaneously with the environmental variables, and is simply treated as another
observed indicator of the profiles
97
. The control variables, including child age, gender,
Hispanic, body mass, socioeconomic status, and distance to school are modeled as
predictors of the underlying categorical latent variable.
Results
Participating children were 51% male, 48% Hispanic, the mean age was 11.8 (SD=1.5),
15% were overweight (BMI between 85
th
and 95
th
percentile for age), 23% were obese
(BMI over 95
th
percentile), and 36% received free or reduced price lunch at school.
Parents were 88% female, 61% Hispanic, and the mean age was 39.5 (SD=5.9).
For the LPA built with the perceived data (the NEWS), figure 1 shows the measures of fit
for models with between one and five profiles. The information criteria drop
precipitously between the first and second models; between three and five profiles the
decline is less dramatic. For all models, entropy is quite good, though highest for model
3. The likelihood ratio test approaches significance until the difference between the third
and fourth models. Statistical considerations alone appear to indicate that perhaps the
three-profile model is the best. However, as noted earlier, we must also consider whether
adding more profiles than this will give us additional insight into the profiles. In this case,
63

adding an additional class appears to provide a helpful degree of elaboration. Therefore,
a four-profile model is chose as the best one. This profile is show graphically in figure 2.
The four profiles can be described as follows: Profile 1 is characterized by high density
(due to the low level of detached single family residential and moderate level of land use
diversity), but low walking infrastructure and aesthetics. 13% of the sample had the
highest probability of being classified in profile 1. Profile 2 is characterized by
consistently moderate levels of virtually all variables; it is neither the highest nor lowest
on any indicator. 13% of the sample had the highest probability of being classified in
profile 2. Profile 3 is characterized by flat land, high walking infrastructure and high
land use diversity. 30% of the sample had the highest probability of being classified in
profile 3. Finally, profile 4 is characterized by the highest level of low-density
residential, the lowest level of land use diversity, and the highest levels of walking
infrastructure, aesthetics, traffic safety and cul-de-sacs. 44% of the sample had the
highest probability of being classified in profile 4.
In terms of active commuting, profile 2 had the highest probability of active commuting
at 0.388. Profile 1 had the second highest, at 0.305; profile 3 had the third highest at
0.274; and profile 4 had the lowest probability of active commuting, at 0.244. Pairwise
comparisons indicate that all probabilities are significantly different from each other.
Moving to the objectively assessed LPA (the IMI), the measures of model fit were
broadly similar to those from the NEWS model. As seen in figure 3, three or four profiles
appears to be the best based on statistical considerations, but the greater specificity
64

flowing from one additional profile resulted in selecting the four profile solution as the
best, just as with the NEWS data. Figure 4 graphically depicts the profiles. Profile 1 is
analogous to profile 2 from the earlier LPA in that it is consistently moderate on all
variables. In this case, 11% of the sample had the highest probability of being classified
in profile 1. Profile 2 is consistently low on most variables, with the exception of traffic
safety and single-family residential. 5% of the sample had the highest probability of
being classified in profile 2. Profile 3 is characterized by high walking infrastructure,
such as sidewalk presence and condition and crosswalks, but is the lowest on the two
measures of traffic safety. 8% of the sample had the highest probability of being
classified in profile 3. Profile 4 is analogous to profile 4 from the NEWS data in that it is
characterized as low-density, primarily residential, with high levels of walking
infrastructure and traffic safety. 76% of the sample had the highest probability of being
classified in profile 4.
Probability of active commuting for each of the profiles was as follows. Profile 1 had the
highest probability, at 0.533, profile 3 had the second highest probability at 0.303, profile
4 had the third highest at 0.26, and profile 2 had the lowest at 0.071. Pairwise
comparisons indicate that the only significant difference in probabilities is between
profiles 1 and 4.
Discussion
The goals of this analysis were to develop unique neighborhood profiles by
simultaneously considering multiple aspects of the physical environment; examine how
65

profiles differ by using either perceived or objective measures; and determine how each
profile related to active commuting to school.
Interestingly, two similar profiles emerged across the NEWS and IMI LPAs: consistently
moderate; and low density residential/high walking infrastructure and traffic safety.
Further, the highest probability of active commuting to school was noted in both of the
consistently moderate profiles. While we of course cannot infer causal relationships from
this cross-sectional data, there are several important implications of these results. The
first is that perhaps making moderate changes to many aspects of the neighborhood may
be optimal from the standpoint of increasing active commuting to school. Wholesale
changes to the way communities are designed and built may not be necessary. This is an
important finding because it holds promise for both new and existing communities. For
neighborhoods developed in the last four or five decades, even if unlimited funds were
available, it would simply be impractical or impossible to retrofit the entire neighborhood
to be denser and closer to destinations, with more connected street patterns and fewer cul-
de-sacs. More likely are smaller changes, such as installing sidewalks where none
existed before, lowering speed limits or installing traffic calming devices, and
landscaping or other cosmetic enhancements to improve aesthetics. These are just the
types of small but broadly applied changes that our results suggest would increase the
probability of active commuting.
Small-scale change is even more important when cost is considered. For example, the
Florida Department of Transportation estimates the cost of installing a mile of concrete
66

sidewalks on one side of the street at almost $153,000, whereas a mile of sidewalks on
both sides costs about $302,000
40
. If a moderate approach is taken to retrofitting
neighborhoods without sidewalks, it might be best to only place the sidewalks on one
side of the street, and apply the remaining funds to several other improvements noted
above.
These results are equally important for new communities. The reality is that in the
United States at least, there is still large demand for low-density, single family
development away from urban centers
11, 59
. Developers simply respond to market
demand and continue building communities of this sort. It is unreasonable to believe that
private land developers will ignore these forces and make a complete shift to high
density, urban infill developments. Fortunately, as our results imply, a total shift may not
be necessary. Instead, adjustments to the way new communities are built, such as
reducing the number of cul-de-sacs, moderately increasing density, small shifts to make
the street network more connected, and more attractive urban design would be
acceptable. Note that this does not require completely scrapping the development model
that currently exists and has proven to be quite popular. Relatedly, these moderate shifts
may be easier to integrate into municipal zoning codes and ordinances than large, radical
changes, particularly because the final decision on amendments to zoning and ordinances
typically lies with an elected body, such as a city council, and not with non-political
staffers. Codifying moderate-scale changes is important because they would then apply
equally to all new projects within a given jurisdiction, increasing the likelihood of
population-level changes in healthy behaviors like active commuting to school.
67

Another interesting point is that both the NEWS and IMI uncovered a low density
residential/high walking infrastructure and traffic safety profile (#4). In the case of the
NEWS, this profile was characterized by the lowest probability of active commuting, and
for the IMI, while overall it did not have the lowest probability of active commuting, the
only significant difference in probability was between this profile and the consistently
moderate one discussed above. For the IMI, the profile with the lowest probability of
actively commuting (#2) was characterized by very similar levels of residential density
and land use mix, as well as high levels of traffic safety. While the probability of active
commuting in profile 2 was not significantly different from other profiles, it is consistent
with results from profile 4.
This is somewhat at odds with what might be expected because these profiles are so high
on factors that are thought to promote active commuting; in fact, the only negative
aspects have to do with density and diversity. At first glance it may seem that a lack of
density and land use diversity can overpower many other positive features and hinder
active commuting to school. However, the case may be that these are just indicators of
neighborhoods that are among the most isolated, in the sense that they are located
relatively far from other development. It is not necessarily isolation from schools that is
driving this relationship, as we have controlled for that distance, but more that outlying
neighborhoods could be characterized by a lack of connectivity to other development,
including schools. As an example, two children may live a half-mile from their
respective schools, but a child in an older, more urbanized neighborhood could be
expected to have infrastructure that developed over time to connect his neighborhood to
68

the school. However, surrounding developments may not have progressed to the point
where a child in a newer, edge-of-the-city neighborhood has means of active travel from
his neighborhood to the school. This is to say that connectivity and infrastructure
between neighborhoods is just as important as those factors within the neighborhood.
Children in this fourth profile appear to have excellent conditions within their
neighborhood to be physically active, but it may be more difficult to safely travel by
walking or biking outside its confines. This is an entirely plausible situation since for
both the NEWS and IMI we are only seeking to characterize the immediate
neighborhood, not the broader community. Future research should consider the
environment on the entire pathway between home and destination, and not assume what
characterizes the immediate neighborhood generalizes to the entire trip.
It is interesting to note that a substantially higher portion of participants had the highest
probability of being classified into this low density/high walking infrastructure profile by
the IMI than by the NEWS (76% v. 44%). Given this discrepancy, it is clear that people
do not “see” their neighborhood in the same way that the objective method does. Several
reasons may account for this. First is the concept of scale discordance, in which
individuals conceptualize the boundaries of their neighborhood differently than the
objective method does. This is almost certainly occurring to some extent here, given that
the 500 meter buffer is not likely to fit most people’s definition of the neighborhood. The
second is that the IMI is more of “snapshot” of the neighborhood on the particular day the
auditors were there. The NEWS on the other hand is more like a “movie”, in that
participants have a much greater deal of experience living and interacting with the
69

neighborhood on a daily basis, and can incorporate perhaps many years of this experience
when self-reporting responses.
It is important to emphasize that these issues are not limitations of the analysis; in fact,
they bear directly on the question of whether the perceived and objective environment are
the same thing, and if not, why. The differential classification into profiles across
perceived and objective LPAs indicates that these and perhaps other methodological
artifacts make direct substitution of one method for the other impossible without arriving
at different conclusions about the nature of an individual’s neighborhood.
Despite the discrepancy in terms of the proportion of people who are classified into the
different profiles across the perceived and objective analyses, both methods converge
onto basically the same types of environments that are the most and least supportive of
active commuting in the context of the LPA. In that sense, the methods are substitutable
to some degree when the goal is to identify and characterize the kinds of neighborhoods
(or more accurately, profiles) that exist in a given area. This general understanding
would be useful in the initial stages of determining what environmental policies or
interventions would be appropriate in a community
1
.
Up to now, we have focused on the implications of the results from the standpoint of
making structural changes to the environment, e.g. adding sidewalks or traffic calming
devices. However, it may be possible to achieve increased active commuting to school
by only shifting parent’s perceptions of the neighborhood, while making little to no
structural changes. Support for this idea comes from the 99 parents who by the objective
70

method had the least supportive neighborhood for active commuting, yet by the perceived
method saw their neighborhood as something other than the least walkable. We cannot
say precisely what caused the objective-subjective difference, but in addition to the
methodological differences noted above, it may be due to intrapersonal factors. The first
paper in this dissertation showed that informal social control is correlated with perceiving
an objectively negative area in a more positive manner. Resiliency stemming from strong
neighborhood institutions and social conditions has previously been offered as an
explanation as to why certain individuals can thrive in non-supportive or even hostile
environments, so there may be an element of that here
133
. Self-efficacy has also been
linked to perceiving a less activity-friendly environment as one conducive to activity, as
has being more physically active
43
. However, those studies focused on adult’s
perceptions and their own activity, not how their perceptions linked to their child’s
activity as we do here. More research is needed to understand whether acting on these
factors in parents creates change in active commuting by children.
Finally, it is important to realize that we have identified more and less supportive
environments that relate specifically to active commuting to school by children. What
relationship these profiles have to other forms of physical activity, including in-
neighborhood activity, as well as all forms of activity by adults, is not known at this
point. This is an important area for future research, as ideally we would uncover
environments that maximize all activity by adults and children. Some aspects of the built
environment, such as mixed land use, may promote activity by adults, but hinder activity
by children. Therefore, the complete implications of the types of environmental change
71

noted here, particularly structural change, should be understood before proceeding with
any modifications.
Strengths and Limitations
Strengths of this study include the micro-scale nature of the data, which allows us the
capture and incorporate into the LPAs many of the details of the environment that may
come together to impact physical activity. Also, active commuting is a context specific
activity: At least some portion of the trip must take place in the neighborhood. This
provides a stronger theoretical rationale for the link between the neighborhood
environment and activity than if we had used an a-contextual measure of activity, such as
accelerometers.
Limitations of this study are as follows. The first is that the data is cross-sectional; as
such we cannot attribute causality to any feature or combination of features of the
neighborhood. The second is that latent profiles are somewhat empirically driven. We
cannot guarantee that any neighborhoods actually exist in our sample that combine all the
variables at the levels described, or that we could create one that exactly mirrors any of
the profiles. Third, the measure of active commuting is a simple dichotomous variable. It
cannot quantify actual minutes spent walking or biking or sitting in a car or bus; that
knowledge would provide a more detailed understanding of how the neighborhood can
impact longer-term health outcomes like BMI or cardiovascular health through physical
activity. Fourth, as noted earlier, it is possible that self-selection bias is leading to more
active families to choose more activity-friendly neighborhoods. Like most studies, we do
72

not have the ability to control for that here; results should be interpreted with this in
mind. Finally, we have not accounted for children’s perceptions of their neighborhood.
Thus we are assuming that it is the perceptions of the parent that are the primary driver of
whether children actively commute to school or not.
Conclusions
In seeking to better understand the relationship between environment and physical
activity, it is important to model the environment more closely to the way people actually
experience it, which is to say holistically rather than piecemeal. This also helps to give a
better sense of which aspects may be working together such that they are multiplicative
rather than simply additive in terms of their impact on activity. Using this type of
approach, we found that active travel to school may be encouraged by living in a
neighborhood with many moderately walkable characteristics. It may be hindered by the
types of low-density, primarily residential neighborhood that can be found at the edge of
the development line. While similar profiles emerged from both the perceived and
objective data, individuals did not retain their same classification across LPAs. Further
work is needed that assesses the environment outside the neighborhood, and that
examines longitudinal models of change.

73

Study Two Figures
Figure 2.1: NEWS Model Fit by Number of Classes

Figure 2.2: NEWS Latent Profile Analysis

0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
1.02
0
2000
4000
6000
8000
10000
12000
14000
16000
1 2 3 4 5
Entropy
SS-BIC
No. of Classes
SS-BIC
Entropy
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
1
2
3
4
74

Figure 2.3: IMI Model Fit by Number of Classes

Figure 2.4: IMI Latent Profile Analysis

0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
1.02
-1000
0
1000
2000
3000
4000
5000
6000
1 2 3 4 5
Entropy
SS-BIC
No. of Classes
SS-BIC
Entropy
0
0.5
1
1.5
2
2.5
1
2
3
4
75

CHAPTER 4: A LONGITUNDINAL EXAMINATION OF THE
ECOLOGOCAL MODEL OF PHYSICAL ACTIVITY IN ADULTS
Abstract
Background and objectives: The ecological model incorporates multiple levels of
influence, including intra- and inter-personal constructs and environmental factors, and
has been proposed as a causal model of physical activity. However, the causal ordering
among the various factors is not clear, primarily because there has been little examination
of alternative models of causality and longitudinal research. The objectives of this study
are to establish causal ordering among the levels of influence and determine if any
significant mediation paths exist.
Methods: Two waves of data from 312 adults participating in the Healthy Places trial
were used. Participants answered questions about self-efficacy, attitudes, subjective
norms, and social cohesion, and completed the Neighborhood Environment Walkability
Survey to assess the environment. Accelerometers were worn for one week to assess
moderate-to-vigorous physical activity. Structural equation modeling was used to test a
series of nested models proposing various causal relationships among the levels of
influence. Specific paths within the best fitting model were tested for evidence of
mediation. Demographic factors were controlled for.
Results: The overall best fitting model was one in which the perceived built environment
affected intra- and inter-personal factors, which in turn affected MVPA. However, there
was no evidence of any mediation effects within this model.
76

Conclusions: While the best fitting model was consistent with theoretical expectations,
there was little support for mediation effects, raising questions about the value of the
tested model for both explanatory and intervention development purposes. Other
constructs, particularly with respect to intra-personal factors, and relationships, such as
moderation, may provide a more robust causal model of physical activity.

77

Introduction
As previously discussed, models exist that have attempted to account for multiple levels
of influence on physical activity behavior, and their primary hypothesized direction of
effect is from the environment to intrapersonal characteristics to behavior. The goal of
this final study is to extend prior research by examining not only the standard
unidirectional model, but also multiple alternative unidirectional models, where the
specified factor ordering is reversed, and reciprocal models that contain feedback loops, a
process that may help uncover causal ordering of the variables and allow for testing of
mediation pathways.
A major limitation of prior research is the use of primarily cross-sectional data.
Attempting to establish causality, especially in terms of mediating effects, with only one
wave of data has been shown to markedly bias estimates of effect
78, 123
. This study will
attempt to improve on existing work by utilizing two waves of data to establish temporal
ordering of variables. Additionally, it will take a structural equation modeling approach.
This is important for three reasons. First, it allows for the simultaneous estimation of
multiple hypothesized pathways. Estimation of individual pathways one at a time, as in a
standard regression framework, does not fully account for the co-occurring effects of
other pathways. Second, it allows for the incorporation of measurement error in model
estimates through the use of multiple indicators for each latent construct, again,
something that cannot be accomplished in regression analysis. Third, it allows for
systematic testing of competing models through tests of model fit
60
. This will aid in
78

selecting the model that best fits the pattern of observed data, which is especially
important due to the number of models that will be tested.
The third proposed study builds on the first two in several ways. First, it incorporates
important non-environmental constructs that may impact physical activity, specifically
individual-level psychological constructs. Second, it proposes and tests mediation
pathways, rather than assuming a direct effect of the independent variable on the
outcome. Finally, and perhaps most importantly, it utilizes longitudinal rather than cross-
sectional data to more accurately explore the complex relationships proposed in the
models below.
Research and Policy Implications
For researchers, understanding the relationship among individual factors, the
environment and physical activity is important, as this knowledge can provide a
framework for designing interventions to promote activity. As with any behavior,
knowing the causal ordering of factors will enable researchers to direct their efforts at the
intervening variables most likely to create change. The results of this study will be
particularly important in this regard because, as noted earlier, current research seems to
be leaning towards one particular causal chain without fully exploring the alternative
models to be tested here. Before moving to expensive programmatic interventions, the
causal model should be more clearly established.
For policy makers, knowledge of causal ordering will aid in understanding whether it is
reasonable to expect that environmental change can meaningfully impact activity on its
79

own, or if the effect of the environment works in concert with individual-level factors
they may have much less control over. This analysis will also be able to identify
individual environmental components that have promising effects on physical activity,
either directly or indirectly, since it will treat the components as discrete factors, as
opposed to one overarching “environment” construct. As cities begin to incorporate
health concerns into their planning activities, it is important they act on full information
concerning whether and how their attempts to make the environment more activity-
friendly will produce increases in activity commensurate with the effort (in terms of both
time and money) to make those changes.
Research questions
Study three will be designed to answer the following questions:
1. Given measures of intrapersonal factors, the neighborhood environment, and
physical activity, does a unidirectional or reciprocal model better sort the causal
order of the factors?
2. Once order is determined, can particular mediating pathways be identified
between any of the factors above?
Methods
Participants and Setting
Healthy PLACES (Promoting Livable Active Community Environments) is a four year,
controlled trial that has enrolled families from neighborhoods in southern California
representing a diverse range of urban characteristics, including street layouts, density,
80

housing types and land use mixture. The goal of the study is to understand how features
of these neighborhoods influence behaviors such as physical activity and eating.
Families in eligible cities were recruited by informational fliers distributed at schools,
community events and shopping centers, advertisements in local papers, postcards mailed
to homes, and door-to-door recruitment by study staff. Interested parents were asked to
call the study office, where there completed a brief eligibility screen. To be eligible,
families must reside in one of several cities in San Bernardino County, CA; have a child
in 3rd
th
-8
th
grade, and a household income of less than $165,000. One parent and one
child from each family were enrolled. Informed consent was obtained from participants,
and the study was approved by the University of Southern California Institutional Review
Board. In this analysis, baseline and follow-up data collected approximately one year
later from 312 adults is used.
Measures
Psychosocial variables were self-efficacy, attitudes, subjective norms, and social
cohesion. Self-efficacy was measured by three questions asking how sure participants
were they could exercise more, be active in their free time even on a busy day, and ask
their best friend to be physically active with them. Items were scored on a four point
scale, from not sure to definitely sure, and the Cronbach’s alpha was 0.75. Attitudes were
measured by three questions using a seven point semantic differential scale with a
common stem
25
: “I feel that participating in regular physical activity is…” A) extremely
useless-extremely useful; B) extremely harmful-extremely beneficial; C) extremely
81

unenjoyable-extremely enjoyable. Cronbach’s alpha for this scale was 0.78. Subjective
norms were measured by two questions asking whether people who are important to them
think they should participate in regular physical activity, and whether people who are
important to them support them participating in regular physical activity
30
. These were
scored on a seven point scale, from strongly disagree to strongly agree, and the
Cronbach’s alpha was 0.71. Social cohesion was measured by two questions asking how
satisfied participants were with the number of friends and number of people they know in
their neighborhood, scored from strongly dissatisfied to strongly satisfied, and the
Cronbach’s alpha was 0.93
111
. The Neighborhood Environment Walkability Scale was
again used to measure perceptions of the neighborhood environment (see study 2).
Recall that of the 14 NEWS factors, eight are measured by multiple variables, and six are
single item questions. For this study we have elected to utilize only the six factors that
were previously validated in a multi-level confirmatory factor analysis (land use mix-
access, street connectivity, infrastructure for walking/biking, aesthetics, traffic safety, and
crime safety). The reason for this is that it allows us to keep the structural models almost
fully latent, a desirable quality.
The physical activity outcome variable was measured by using uniaxial Actigraph
accelerometers. Participants were asked to wear the device on their hip for seven days,
removing it only when sleeping, bathing, or swimming. Upon returning the device, study
staff downloaded and processed the data using Meterplus software. The accelerometer
records data in 30-second epochs. Based on the number of counts (an indication of
activity) in each epoch, it is classified as sedentary, low, light, moderate, or vigorous
82

intensity activity. Classification thresholds were determined using established cut
points
126
. The epochs recorded in each category on valid days (defined as at least 10
hours of wear time) were summed, divided by two (to convert epochs to minutes), and
then divided by the number of valid days. This produces average daily valid minutes in
each of the activity intensity categories. Only individuals with at least three valid days
were used in the analysis. For purposes of analysis, moderate and vigorous activity was
summed to create a moderate-to-vigorous activity (MVPA) category.
Analytic Plan
The first step is to construct confirmatory factor analysis (CFA) models. Following the
procedures of Eveland et al, separate CFAs will be developed for the set of psychosocial
and environmental latent factors
37
. The CFAs consist of the component factors measured
at time one and time two, configured so that the measurement errors of the same indicator
variable were correlated between the two time points, while also estimating a path
between the factors to assess temporal stability of the factor
72
. Because measurement of
each factor should be invariant over time, the loadings of each indicator are constrained
to be equal at time one and time two. The fit of this model is assessed using conventional
fit statistics (see below). If the constrained model does not fit worse than the
unconstrained model, the CFA demonstrates weak invariance, the minimum level of
invariance necessary for a longitudinal model
72

After establishing the measurement model, structural models will be developed for each
of the hypothesized relationships. As there are only two waves of data, all the following
83

models were based on the “half-longitudinal” analysis methods proposed by Cole and
Maxwell
23
. Ideally, when assessing mediation longitudinally, we would have at least
three waves of data, which would allow for temporal separation between the predictor
and mediator and between the mediator and the outcome. However, two waves are
acceptable when the causal process can be considered ongoing, rather than a function of a
discrete event setting it off (as in a deliberate intervention)
23
. All models here would be
consistent with the assumption of an ongoing process.
Model specification
Ultimately, there are nine separate models that will be tested. The first four models are
based on the idea that intrapersonal constructs mediate the relationship between the
perceived environment and physical activity. In the first model (stability model), only
autoregressive paths are specified from each time one (T1) factor to the same factor at
time two (T2). In figure 2, these are indicated by the paths “LagE”, “LagI”, and
“LagPA”. This model implies that the only predictor of any factor at T2 is the same
factor at T1, and that there are no mediating or reciprocal relationships among the factors.
It should be noted that in figures 2-4 only three factors are specified: environment,
intrapersonal, and physical activity. However, these are only conceptual second order
factors subsuming all the individual first order factors under them. This is done for ease
of presentation of multiple complex models. In the actual analysis, all environmental and
psychosocial factors listed in the measures section above are modeled individually (i.e.
walking infrastructure, aesthetics, self-efficacy, attitudes, etc. are all modeled
84

individually). This produces a total of 10 latent variables at each of time 1 and time 2
(six NEWS factors, and four psychosocial factors), plus one manifest physical activity
variable.
The second model (conventional model) builds on the stability model by further adding
paths from environmental factors at T1 to intrapersonal factors at T2, and from
intrapersonal factors at T1 to physical activity at T2. In figure 2, these paths are labeled
“A” and “B.” This model implies that the environmental perceptions predicts
intrapersonal factors, and those factors predict physical activity. This is termed the
conventional model because it is in line with what most prior research has tested.
The third model (reverse model) is similar to the conventional model, except that it
reverses causal order, such that there are paths from intrapersonal factors at T1 to
environmental perceptions at T2, and from physical activity at T1 to intrapersonal factors
at T2. In figure 2, these paths are labeled “C” and “D”. This model implies that physical
activity is the most distal variable, and predicts intrapersonal factors, which in turn
predict perceptions of the environment. Like both previous models, the autoregressive
paths are specified.
The fourth model (reciprocal model) combines the first three models. In figure 2, this
includes all labeled paths. This model implies a feedback loop, where environmental
perceptions can cause physical activity by working through intrapersonal factors, and
physical activity can cause environmental perceptions by also working through
intrapersonal factors.
85

Models five through eight are similar to the first four models, except that the positions of
intrapersonal factors and environmental perceptions are reversed. Whereas models one to
four are essentially variations on the previously published models discussed in the
background section that assume intrapersonal factors are always proximal to physical
activity, the next four models are consistent with the alternative structure proposed,
where environmental perceptions are always proximal to physical activity.
Model five assumes only lagged regression paths between the factors measured at time
one, and are indicated by the paths in figure 3 with the “Lag” label.
The sixth model (alternative conventional model) builds on the stability model by further
adding paths from intrapersonal factors at T1 to environmental factors at T2, and from
environmental factors at T1 to physical activity at T2. In figure 3, these paths are labeled
“A” and “B.” This model implies that intrapersonal factors predict environment
perceptions, which in turn predict physical activity.
The seventh model (reverse model) specifies paths from environmental perceptions at T1
to intrapersonal factors at T2, and from physical activity at T1 to environmental
perceptions at T2. In figure 3, these paths are labeled “C” and “D”. This model implies
that physical activity is the most distal variable, and predicts environmental perceptions,
which in turn intrapersonal factors. Like both previous models, the autoregressive paths
are specified.
86

The eighth model (reciprocal model) combines the previous three models. In figure 3,
this includes all labeled paths. This model implies a feedback loop, where intrapersonal
factors can cause physical activity by working through environmental factors, and
physical activity can cause intrapersonal factors by also working through environmental
perceptions.
The ninth and final model is displayed in figure 4. This is the fully reciprocal model,
where each factor at time 2 is caused by all the factors at time 1. This depiction is the
model most faithful to Bandura’s reciprocal causality model, since there is no explicit
causal ordering.
In each of the above models, correlations were specified among all exogenous variables,
and among all disturbances. To account for method variance over time, correlations were
specified between the measurement errors for the same indicator at T1 and T2, just as in
the measurement model. To control for possible confounding influences, all models
include variables representing age, gender, income, body mass index and ethnicity.
Model assessment and selection
As there is no single accepted index to assess the fit of the model to the data, multiple
indices were used, including the comparative fit index (CFI), the standardized root mean
square residual (SRMR), root mean square error of approximation (RMSEA), and 
2
.
Well-fitting models should have a value of approximately 0.95 on the CFI, SRMR should
be less than 0.08, and the RMSEA should be less than 0.06
52
. Ideally, the 
2
p-value
should be non-significant, but in large samples it will almost always be significant
60
.
87

After assessing the fit of each model individually, the next step is to compare models to
each other and choose the model that provides the best fit while placing a priority on
model parsimony. The 
2
difference test allows us to compare models when they are
nested. Taking models one to four as set A and models five to eight as set B, the stability,
conventional, and reverse models within each set can be individually compared to the
reciprocal model because they are each nested within it. This first set of tests will allow
us to determine whether, for each set, the reciprocal model should be kept; if all three
other models fit worse than the reciprocal, then it can be judged the best model.
However, if any of the three do not fit significantly worse than the reciprocal, then the
reciprocal can be ruled out. The stability model is also nested within the conventional
and reverse models. Using a similar process of 
2
difference testing, we can determine
whether the simpler stability model is a better fit than the more complex models, or if it
can be ruled out due to poor fit. If at the end the stability and reciprocal models are ruled
out, we cannot judge between the conventional and reverse models using the 
2
difference test because they are not nested. Therefore, we will compare them using the
measures of relative fit noted earlier and also examine information criteria, including
Akaike Information Criteria (AIC), Bayseian Information Criteria (BIC) and sample size
adjusted BIC (SS-BIC). The model with lower values on these criteria provides the
superior fit
60
. While no statistical significance test is available for information criteria, a
general guide is a 5 to 10 point difference in the information criteria between models
indicates better fit
103
.
88

After choosing the single best model from within sets A and B, the two models selected
can them be compared against the fully reciprocal model nine. This is possible regardless
of which model is selected from each set, since the first eight models are all by definition
nested within model nine. This final testing will allow us to rule in or rule out model
nine. If its fit is superior to less complex models, it will be kept as the best model. If
however it fits significantly worse than the models selected from within sets A and B, it
will be ruled out and a comparison between the remaining two models will be done using
relative fit indices and information criteria to determine the best single model.
In the event that either the conventional, reverse or reciprocal models from within set A
or B are identified as providing the best fit, indirect, or mediated, effects can be
calculated by multiplying the coefficients labeled “A” and “B” or “C” and “D” in figures
2 & 3. This procedure uses the product of coefficients method for calculating indirect
effects, which has been shown to have greater statistical power than the Baron and Kenny
method
74
. Rather than relying on normal-theory significance tests, bias-corrected
bootstrap confidence intervals can be created for indirect effects by repeatedly drawing
observations with replacement from the data to create a “new” dataset of equal sample
size to the original and running the model with this new dataset. Over many iterations,
10,000 in this case, the parameter estimates from each new dataset form an empirical
sampling distribution from which confidence intervals, and thus tests of statistical
significance, can be derived
77
. All statistical analyses were completed in Mplus v. 6.12
using the robust maximum likelihood estimator.
89

Results
Of the 312 participants, 87% were female, 56% were Hispanic, and mean age at time 1
measurement was 39 (SD=6.04). The median income at time 1 was between $60,000 and
$70,000, while at time 2 it was between $50,000 and $60,000. At time 1, median daily
MVPA was 20 minutes; the time 2 median was 21 minutes. Time 1 median BMI was
28.8, while at time 2 it was 27.9.
Confirmatory factor analyses
The CFA for the psychosocial factors showed excellent fit (CFI=0.95; SRMR=0.07;
RMSEA=0.043 (95% CI=0.032-0.053); 
2
=239.796, DF=152, p<0.0001). No significant
decrement in fit was noticed by restraining factor loadings to be equal over time, thereby
demonstrating weak factorial invariance. The psychosocial CFA can be seen in table 1.
Initially, our goal was to re-create the established factor structure of the NEWS with six
latent factors; however replicating this structure longitudinally resulted in a poor fit even
without invariant loadings (CFI=0.76; SRMR=0.083; RMSEA=0.062 (95% CI=0.059-
0.064); 
2
=3793.57, DF=1732, p<0.0001). Rather than attempting to make questionable
ad hoc modifications to improve fit, such as within-time error correlations, we decided to
re-examine the data and its structure by first conducting an exploratory factor analysis
(EFA) with geomin rotation.
One benefit of the EFA is that it can help reduce the number of latent factors to a more
manageable number. Also, it makes clear whether certain variables have important cross-
loadings that were ignored as part of original CFA, which restrained all cross-loadings to
90

be zero. Conventional CFA wisdom holds that cross-loadings are to be avoided, as they
may make the definition of the latent variables less clear. However, as Asparouhov and
Muthen have recently demonstrated, ignoring significant cross-loadings, even those small
in magnitude, can artificially inflate the correlations between latent factors, negatively
impacting structural paths
5
. Allowing cross-loadings is an acknowledgement of the
complex inter-relationships among micro-scale measures of the environment. Therefore
the EFA was used to identify a more parsimonious factor structure and potential item
cross loadings that could be modeled in a follow-up CFA.
The revised CFA included a smaller number of latent factors, as well as several cross-
loadings. This model reduced the number of factors to four, capturing aesthetics, land
use access, walking infrastructure, and crime and traffic hazards. The fit was good
(CFI=0.895; SRMR=0.066; RMSEA=0.044 (95% CI=0.04-0.049); 
2
=1278.159,
DF=793, p<0.0001), and no significant difference was noted by constraining loadings to
be equal across time. The final environmental CFA used as the basis for the structural
models to follow can be seen in table 1.
Structural models
The nine a priori structural models were run next. The fit statistics and indices for all
models are shown in table 2. As would be expected, the fit indices are quite similar, and
in the case of RMSEA and SRMR, show excellent fit to the data. The CFI is below
conventional measures of acceptability, but it is known to have properties that make it a
poor choice for assessing fit in the presence of numerous either high or low observed
correlations in the data, as is the case in this dataset.
91

Results of the formal 
2
difference tests are shown in table 3. For set A, the non-
significant comparison between models 4 and 2 indicates that model 2 does not fit the
data any worse, and for parsimony concerns, we choose that model initially. Then,
comparing model 2 to the even more parsimonious model 1, we find that the stability
model does in fact worsen model fit. Based on these comparisons, we choose model 2 as
the best model from set A. In set B, we find that both models 6 and 7 do not worsen
model fit compared to model 8, leading us to reject model 8 and test 6 and 7 against the
parsimonious stability model (5). Unlike in set A though, we find that model 5 does not
fit significantly worse than 6 or 7, so we choose 5 as the best fitting model for set B.
Finally, we compare models 2 and 5 against the fully reciprocal model. Model 2 does not
fit significantly worse, leading us to reject model 9; however compared to model 5, 9
does fit better. Thus we are left with 2 as the best fitting model in one comparison and 9
in the other. This then reduces to the test we just conducted comparing 9 versus 2. On
the basis of that test, we choose model 2 as the overall best fitting model. Parameter
estimates from model 2 are displayed in table 4.
Finally, we must examine whether any indirect (or mediated) effects exist within the
framework of model 2. Of the 16 possible mediation paths, running from each
environmental factor to each psychosocial factor to MVPA, all are small in magnitude
(range: -3.3 to 2.5 minutes per day change in MVPA per unit increase in the
environmental factor, mediated through the psychosocial factor) and none approached
statistical significance.
92

Discussion
This analysis is among the most comprehensive assessments of the ecological model and
its impact on physical activity to date. Ultimately, we did not find support for any
mediation pathways within the model. While the models presented here are not
exhaustive, they do raise questions about the relationship among the perceived
neighborhood environment, psychosocial factors, and physical activity.
There are a number of issues to consider. One possible explanation for the lack of any
mediated findings is the restricted way in which we allowed the factors to influence each
other. Factors within the environmental and psychosocial blocks were only allowed to
correlate within time; they were not allowed to predict one or more other factors from the
same block over time. The reason for this approach is a lack of theory about how these
factors should relate to each other causally within the context of the ecological model.
Therefore, the most parsimonious possible approach was taken. Existing models of
behavior which focus exclusively on psychosocial factors do propose causal relations
among these factors. For example, the Theory of Planned Behavior proposes that
attitudes and subjective norms are indirect causes of behavior, mediated by intentions (a
construct not included here)
85
. It is possible there is a more complex relationship among
factors than specified here.
Another reason for the lack of mediated paths may have to do with the fact that we only
included perceived measures of the environment. As we have seen in the first two
papers, perceived and objective aspects of the environment are not necessarily congruent.
93

Including objective measures, perhaps as a distal predictor of the perceived environment,
or as an independent predictor of MVPA, might lead to different conclusions. Given that
both perceived and objective measures of the environment have been correlated with
MVPA, an area for future research is to understand how the two methods of assessment
relate to each other and other factors within the ecological model.
As we noted previously, it is possible perceived and objective measures of the
environment relate to each other through a moderation process. This idea can plausibly
be extended to the full ecological model such that instead of a mediation effect, the
environment, psychosocial factors and MVPA interact, rather than sequentially cause
each other. A recent cross-sectional study demonstrated that psychosocial factors interact
to some degree with aspects of the environment thought to influence recreational activity.
However, there was no evidence this applied to either overall MVPA or transport
walking
32
. Also, moderation may be upwardly biased in cross-sectional data the same
way mediation is, so this data must be confirmed longitudinally before accepting it as a
reasonable alternative expression of the relationships between levels of influence.
Some may critique our use of an a-contextual accelerometer-derived measure of physical
activity, rather than an instrument designed to specifically capture neighborhood-based
activity, either through self-report or linking accelerometer data to GPS. This is a
legitimate point to raise, and should be tested in future research, but as with the other
points discussed thus far, it is unclear whether this would have a substantive effect on our
conclusions. To further probe this idea, identical model testing procedures outlined in the
94

methods section were followed, but this time substituting minutes of light accelerometer-
derived physical activity (1-<3 METS) for MVPA. This level of activity can include light
walking of the sort that may be done in the neighborhood for either leisure or
transportation purposes, for example walking to a neighbor’s house or bus stop
2
. In this
case, the stability model was the best model, indicating the total absence of mediation
effects, and more generally, any across-time relationships among any of the factors, save
for their own value at time 1 (details available in Appendix 3). It is reasonable to expect
that if a detailed measure of neighborhood-based physical activity would lead to
substantively different conclusions, we should see at least some semblance of a mediation
effect with one of these two measures of physical activity, even if they lack context.
The points discussed above relate to model modifications that may be appropriate to
investigate in the future. However, they remain within the basic context of an ecological
model in which there are relationships among the environment, theory-based
psychosocial factors, and physical activity. The lack of any promising mediation effects
in the models tested here indicate that there may be a need to fundamentally rethink the
way the ecological model has been conceived up to now.
Part of the reason why none of the tested models hold up may be that they assume the
effect of the environment on physical activity always involves psychosocial factors in
some form. Some models assume psychosocials inform the environmental perceptions,
and others that the environment works only indirectly through psychosocials. Contrary to
what appears to be current belief, the environment may be acting directly on behavior
95

through automatic processes such as associative memory and implicit attitudes, with little
to no influence from factors like self-efficacy or subjective norms
121, 122
. In fact, some of
the earliest proponents of the ecological model allowed for a direct influence of the
environment on behavior
115
. Once these processes develop, they can exert an influence on
behavior that does not require active cognition in the same way that traditional
psychosocial variables do. While at this point largely untested with physical activity,
such alternative mechanisms should be examined in future research to further crystallize
what the amorphous “intrapersonal” level of the ecological model truly means.
Strengths and Limitations
A primary strength of this analysis is the longitudinal data structure. This allows us to
satisfy one of the chief criteria of testing mediation, temporal precedence of variables.
Another strength is the use of structural equation modeling to examine all environmental
and psychosocial constructs as latent factors that take measurement error into account.
Also, we have tested numerous plausible causal models, a common recommendation but
rare across all applications of SEM.
Several limitations should be noted. Without four waves of data, it is not possible to
directly examine how changes in all the variables effect changes in variables that follow.
Two-wave mediation allows these effects to be indirectly inferred, but not with the same
certainty as more waves would provide. The second is that while we have controlled for
multiple theoretically important confounders in our models, it is virtually impossible to
totally rid any statistical model (even one from a randomized design) of the omitted
96

variable problem. This is less of an issue here because we have not claimed to have found
any causal relationships, but it is a point that should be acknowledged. Third, as with the
second study, we again cannot account for self-selection bias and its possible impact on
the results because individuals are not randomly allocated into neighborhoods.
Conclusions
The expanding body of built environment research incorporating complex, multi-level
models is intriguing and potentially rewarding because it may uncover the subtle
relationships between an individual and their environment that are necessary to
understand if we are to develop effective environmentally-oriented interventions and
policies to promote physical activity. Empirical support for these models is necessary,
and like most model development processes, it will likely follow an iterative approach,
with each tested model providing insight into how future models should be structured.
The results of our model testing indicates a need to refine what components make up
each level of the ecological model, and precisely how those levels relate to each other.

97

Study Three Figures and Tables
Figure 3.1: Longitudinal Measurement Model for Latent Aesthetic Factor

Figure 3.2: Conventional EIP, Reverse PIE & Reciprocal EIP Models

Aesthetics_T1
V1
1
V2
2
V3
3
Aesthetics_T2
4
V4
5
V5
6
V6
7
Figure 1: Longitudinal Measurement Model for Latent Aesthetics Factor
Note: Pairs of correlated errors represent the same variable measured at time 1 and time 2,
i.e. (V1,V4) (V2,V5) (V3,V6)
Env_T1
Intra_T1
PA_T1
Env_T2 1
Intra_T2 2
PA_T2 3
LagE
LagI
LagPA
Figure 2: Conventional EIP, Reverse PIE & Reciprocal EIP Models
C
A
B
Note that factors in ellipses represent conceptual second order factors.
Analysis is actually done with individiual first order factors. See text for further description
Model 1=Lag paths only; Model 2=Lag + A & B paths
Model 3= Lag + C & D paths; Model 4=All paths
D
Time 1 Time 2
98

Figure 3.3: Conventional IEP, Reverse PEI & Reciprocal IEP Models

Figure 3.4: Fully Reciprocal Model

Intra_T1
Env_T1
PA_T1
Intra_T2 1
Env_T2 2
PA_T2 3
LagI
LagE
LagPA
Figure 3: Conventional IEP, Reverse PEI & Reciprocal IEP Models
C
A
B
Note that factors in ellipses represent conceptual second order factors.
Analysis is actually done with individiual first order factors. See text for further description
Model 5=Lag paths only; Model 6=Lag + A & B paths
Model 7=Lag + C & D paths; Model 8=All paths
D
Time 1 Time 2
Env_T1
Intra_T1
PA_T1
Env_T2 1
Intra_T2 2
PA_T2 3
LagE
LagI
LagPA
Figure 4: Fully Reciprocal Model
C
A
B
Note that factors in ellipses represent conceptual second order factors.
Analysis is actually done with individiual first order factors. See text for further description
All paths depicted are involved in fully reciprocal model.
D
Time 1 Time 2
E
F
99

Table 3.1: Final Psychosocial and Environmental Confirmatory Factor Analysis
Latent factor Manifest variable Unstandardized
loading (Standard
error)
Self-efficacy Exercise more 1.0 (0.00)
PA in free time 0.961 (0.067)
PA with best friend 0.661 (0.067)
Attitudes PA is useless 1.0 (0.00)
PA is harmful 0.920 (0.096)
PA is unenjoyable 0.869 (0.060)
Subjective norms Important people think I should be PA 0.923 (0.068)
Important people support my PA 0.923 (0.068)
Social Number of friends in neighborhood 1.065 (0.048)
Number of people you know in neighborhood 1.065 (0.048)
Land use mix-
access
Stores within walking distance 1.0 (0.00)
Many places within walking distance 1.050 (0.048)
Easy to walk to transit 0.938 (0.050)
Many alternative routes in neighborhood 0.300 (0.051)
Crosswalks and pedestrian signals 0.323 (0.053)
Many interesting things to look at 0.218 (0.044)
Walking
infrastructure
Distance between intersections is short 1.0 (0.00)
Many alternative routes in neighborhood 1.002 (0.313)
Sidewalks on most streets 2.601 (0.741)
Cars between sidewalk & road 2.894 (0.864)
Grass/dirt between sidewalk & road 2.449 (0.804)
Neighborhood well lit at night 3.462 (1.071)
Walkers can be seen by people in their home 3.363 (1.000)
Crosswalks and pedestrian signals 1.915 (0.660)
Trees along streets 1.816 (0.591)
Traffic on nearby streets makes it unpleasant to
walk
-1.970 (0.734)
Traffic speed on nearby streets is slow 2.606 (0.806)
Aesthetics Attractive sights 1.0 (0.00)
Many interesting things to look at 0.782 (0.054)
Attractive buildings 0.779 (0.058)
Easy to walk to transit -0.261 (0.063)
Crime & traffic
hazards
High crime rate in neighborhood 1.0 (0.00)
Neighborhood well lit at night -0.128 (0.058)
Crime makes it unsafe to walk during day 0.962 (0.051)
Crime makes it unsafe to walk at night 1.162 (0.062)
Traffic on nearby streets makes it unpleasant to
walk
0.396 (0.185)
Drivers exceed posted speed limits 0.207 (0.065)
Note: Paths constrained to equality across time 1 & 2 within factors. All loadings p<0.05.

100

Table 3.2: Structural Model Fit Statistics and Indices
Model
set
Model
#
AIC SS-BIC Chi2 DF Scaling
factor
from
H0
RMSEA(CI) CFI SRMR
A 1 62785.592 63015.270 3515.283 2298 2.545 0.041(0.038-
0.044)
0.885 0.061
A 2 62787.766 63028.871 3475.671 2278 2.470 0.041(0.038-
0.044)
0.886 0.061
A 3 62807.298 63048.403 3493.027 2278 2.464 0.041(0.039-
0.044)
0.885 0.061
A 4 62809.625 63062.157 3453.691 2258 2.397 0.041(0.038-
0.044)
0.887 0.060
B 5 62785.592 63015.270 3515.283 2298 2.545 0.041(0.038-
0.044)
0.885 0.061
B 6 62808.155 63049.260 3493.592 2278 2.464 0.041(0.039-
0.044)
0.885 0.061
B 7 62808.192 63049.297 3497.640 2278 2.470 0.041(0.039-
0.044)
0.884 0.061
B 8 62830.877 63083.409 3476.416 2258 2.396 0.042(0.039-
0.044)
0.884 0.060
9 62819.556 63076.659 3443.779 2250 2.366 0.041(0.038-
0.044)
0.887 0.060

Table 3.3: Nested Model Comparisons
Model set Model comparison
∆ χ
2
( ∆ D F) P value
A 4 v. 3 32.739(20) 0.036
A 4 v. 2 28.605(20) 0.096
A 4 v. 1 61.277(40) 0.017
A 3 v. 1 28.849(20) 0.091
A 2 v. 1 32.603(20) 0.037
B 8 v. 7 28.609(20) 0.096
B 8 v. 6 27.483(20) 0.122
B 8 v. 5 56.307(40) 0.045
B 7 v. 5 27.709(20) 0.116
B 6 v. 5 28.731(20) 0.093
9 v. best set A model
(#2)
40.355(28) 0.061
9 v. best set B model
(#5)
73.010(48) 0.011

101

Table 3.4: Parameter Estimates from Overall Best-Fitting Structural Model (#2)
Dependent factor
(Time 2)
Independent factor (Time 1) Unstandardized
parameter
estimate
(Standard Error)
Self-efficacy Self-efficacy 0.646 (0.070)
Land use mix-access -0.074 (0.047)
Walking infrastructure -0.027 (0.395)
Aesthetics 0.022 (0.072)
Crime & traffic hazards -0.090 (0.077)
Attitudes Attitudes 0.398 (0.090)
Land use mix-access 0.009 (0.059)
Walking infrastructure 0.223 (0.486)
Aesthetics 0.066 (0.080)
Crime & traffic hazards 0.025 (0.079)
Subjective norms Subjective norms 0.559 (0.083)
Land use mix-access 0.107 (0.066)
Walking infrastructure 0.182 (0.622)
Aesthetics -0.041 (0.112)
Crime & traffic hazards -0.090 (0.108)
Social Social 0.557(0.065)
Land use mix-access 0.005(0.063)
Walking infrastructure 0.025 (0.497)
Aesthetics 0.016 (0.083)
Crime & traffic hazards -0.092 (0.091)
MVPA MVPA 0.489 (0.176)
Self-efficacy 5.014 (5.610)
Attitudes 11.236 (7.029)
Subjective norms -18.029 (10.179)
Social 7.950 (4.285)

102

CHAPTER 5: OVERALL SUMMARY, CONCLUSIONS AND FUTURE
DIRECTIONS
The motivation behind the three dissertation papers was to provide a better understanding
of the relationship between perceptions of the environment, their objective counterparts,
and distinct types of physical activity in adults and children.
In study one, the objective was to understand whether any social or demographic factors
correlated with discordance between objective data and perceived measures of crime and
physical disorder. We found that some factors, such as social control, were associated
with a more positive view of the area, while others, such as time spent in the
neighborhood and being Hispanic, were associated with a more negative view, relative to
objective measures.
While the first study looks at possible reasons for environmental discrepancies, the
second study extended this idea to behavior. Distinct neighborhood profiles were
developed using both perceived and objective measures of the neighborhood; these
profiles were subsequently related to active commuting to school. Both measurement
methods converged on similar profiles that had the highest and lowest probability of
active commuting, yet at the individual level, participants did not always fall into the
same profile across methods.
Finally, the third study combined the perceived environment with another set of
theoretically important psychosocial factors to try to understand the actual causal
mechanisms of physical activity, and in particular, whether the environment was acting
103

directly or indirectly on activity. We allowed for numerous possible causal pathways, but
found no support for any longitudinal mediating pathways in the model, even in the
model most consistent with the frequently cited ecological model of behavior.
Taken together, we can conclude that our understanding of the environment is a function
of the method we choose to assess it with. As mentioned previously, this goes beyond
what may seem like a mere methodological issue; it is not possible to simply “adjust” for
differences between methods. As a result, there are real consequences to our
understanding of what drives physical activity that flow from the decisions we make
about how to measure it. However, this is not especially enlightening, as this point has
been tacitly recognized before. The real contribution of the dissertation is showing how
uniquely the environment is experienced by each person. Physical activity is in part a
function of the environment, whether defined as subjective or objective, as we saw in the
second paper. The objective environment, though, is filtered differently for each person,
which is clear from the differing reports among individuals with basically the same
objective environment in the first paper. Unfortunately, how that filtering process occurs
remains unknown. We had hoped to uncover that relationship in the third paper; though
we were not able to say what the relationship was, we were able to say to some degree
what it is not. In fact, the failure to settle on any of the proposed models could signal that
a unifying theory about the individual-environment-physical activity relationship may not
be possible given the different ways each person experiences their environment.
104

However, this does not obviate the need to have a theoretical understanding of physical
activity behavior; this uniqueness simply points up the necessity of engaging more deeply
in the kinds of “audience segmentation” that Gebel et al have discussed
43
. It may not be
sufficient to target whole cities or even neighborhoods. The most effective interventions
may lie in targeting certain individuals in specific areas. In some ways this brings us
back to many of the original theories and models about physical activity that placed
primary emphasis on the individual. Context certainly matters, but it is individualized, so
that we cannot speak of a particular context that uniformly affects everyone who is
seemingly subject to it.
Future directions
Multiple areas deserving of further investigation were discussed in the individual studies.
However, there are two topics equally applicable to all the studies, and built environment
research as a whole, that should be discussed. The first relates to the aforementioned idea
of audience segmentation. Virtually no work on the built environment and physical
activity takes account of preferences at the individual level. Whether perceived or
objective measures are used, the usual tactic is to assess the distance to or count of very
general categories of destinations, like grocery stores, banks, recreation facilities,
workplaces, etc. The problem is that if an individual’s preference is to shop at a specific
place several miles from their home, no amount of environmental change is likely to
affect that loyalty. For example, if someone has a long-term relationship with Bank A,
located 3 miles from their home, and they have no desire to switch, the fact that Bank B
is located two blocks away is largely irrelevant. They may have positive or negative
105

feelings about the quality of the walking environment along those two blocks, but that
holds little influence over their mode of transport to the bank; brand loyalty does.
Because of the lack of investigation of this factor, it is impossible to say exactly what
effect it has had on current research. However, it is not unreasonable to believe that it
may in part be responsible for the conflicting findings noted in the research about how
specific aspects of the environment affect transport-related physical activity. Research in
progress indicates that local shopping behavior is associated with preferences for local
stores, socioeconomic status, and the neighborhood social environment
34
. This level of
detail about why individuals shop where they do will help identify neighborhoods that are
predisposed to transport-related physical activity, making them prime targets for some
form of environmental modification to facilitate the transition from motorized to active
travel.
One way to account for this is that questionnaires, instead of asking how close is the
nearest bank might ask how close is the bank you typically use, along with one or two
questions asking how likely they would be to switch to a closer, but different brand, bank.
Asking questions in this manner would remove some of the effects of preference. The
latter questions would help further segment the population, because those who are
extremely unlikely to switch probably would not benefit from any method to change their
transportation behavior short of a location of their bank opening nearby. Objective
measures, such as GIS or audits, would have no way of capturing type of information,
though questions about locations that are actually used could be combined with GIS data
to get an objective measure of distance to or count of destinations.
106

Another area for further investigation that has received more attention is the use of
experimental designs in which the environment, programmatic activities, or both are
manipulated. There are several benefits to these designs, including increased ability to
define the level or critical mass necessary to change behavior. For example, how much
density or land use mix is needed to cause more frequent walking to destinations, all
other things equal? This can be accomplished by comparing multiple natural
communities that differ to some degree on density or land use. Discoveries of the level of
a characteristic needed will be a tremendous breakthrough because it will help drive
policy decisions, including the detailed planning and zoning codes that ultimately govern
development.
The difficulty with natural experiments is that there are limitations on how much control
the investigator has on randomization and manipulation of the environment. However,
several promising experiments of this sort have emerged recently, including the Healthy
Places study, and though they will never approach laboratory-grade rigor, they will still
provide a more informed view of how the environment affects behavior
73
. Self-selection
was mentioned as a limitation several times in this dissertation, and studies that are able
to assign the environmental condition to some extent, even if imperfect, will be useful in
separating this bias from the overall effect.
Programmatic activities are of course much easier to randomize and manipulate, and
there are more examples of this type of intervention, especially for programs designed to
increase active commuting to school. Parent or teacher/school education, sometimes
107

combined with environmental modifications, has shown potential to increase this form of
activity
20, 84
.

108

REFERENCES
1. Adams MA, Sallis JF, Kerr J, et al. Neighborhood environment profiles related to
physical activity and weight status: A latent profile analysis. Prev Med. 2011;52:326-331.
2. Ainsworth BE, Haskell WL, Whitt MC, et al. Compendium of physical activities: an
update of activity codes and MET intensities. Medicine & Science in Sports & Exercise.
2000;32:S498-S504.
3. Angur MG, Widgery R, Angur SG. Congruence among Objective and Subjective
Quality-of-Life (QOL) Indicators. Alliance Journal of Business Research. :47-54.
4. Ashford S, Edmunds J, French DP. What is the best way to change self-efficacy to
promote lifestyle and recreational physical activity? A systematic review with meta-
analysis. British Journal of Health Psychology. 2010;15:265-288.
5. Asparouhov T, Muthen BO. Exploratory structural equation modeling. Structural
Equation Modeling. 2009;16:397-438.
6. Ball K, Cleland VJ, Timperio AF, Salmon J, Giles-Corti B, Crawford DA. Love thy
neighbour? Associations of social capital and crime with physical activity amongst
women. Soc Sci Med. 2010;71:807-814.
7. Ball K, Jeffery RW, Crawford DA, Roberts RJ, Salmon J, Timperio AF. Mismatch
between perceived and objective measures of physical activity environments. Prev Med.
2008;47:294-298.
8. Bandura A. The Self System in Reciprocal Determinism. American Psychologist.
1978;33:344-358.
9. Baranowski T, Perry CL, Parcel GS. How individuals, environments, and health
behavior interact: Social cognitive theory. In: Glanz K, Rimer BK, Lewis FM, eds.
Health Behavior and Health Education. 3rd ed. San Francisco: Jossey-Bass; 2002:165-
184.
10. Belcher BR, Berrigan D, Dodd KW, Emken BA, Chou C, Spruijt-Metz D. Physical
Activity in US Youth: Effect of Race/Ethnicity, Age, Gender, and Weight Status.
Medicine & Science in Sports & Exercise. 2010;42:2211-2221.
11. Belden Russonello & Stewart LLC. The 2011 Community Preference Survey: What
Americans are Looking for When Deciding Where to Live. Washington, D.C: National
Association of Realtors; 2011. Available from:
http://www.stablecommunities.org/library/2011-community-preference-survey-what-
americans-are-looking-when-deciding-where-live.
109

12. Boarnet MG, Anderson CL, Day K, McMillan T, Alfonzo M. Evaluation of the
California Safe Routes to School legislation: urban form changes and children's active
transportation to school. Am J Prev Med. 2005;28:134-140.
13. Boarnet MG, Day K, Alfonzo M, Forsyth A, Oakes M. The Irvine–Minnesota
Inventory to Measure Built Environments: Reliability Tests. Am J Prev Med.
2006;30:153-153.
14. Boone-Heinonen J, Guilkey DK, Evenson KR, Gordon-Larsen P. Residential self-
selection bias in the estimation of built environment effects on physical activity between
adolescence and young adulthood. International Journal of Behavioral Nutrition and
Physical Activity. 2010:70.
15. Centers for Disease Control and Prevention. Physical Activity Levels of High School
Students — United States, 2010. MMWR. 2011;60:773-777.
16. Centers for Disease Control and Prevention. Obesity Prevalence. Atlanta, GA:
Centers for Disease Control and Prevention; 2009 Accessed 4-17-09.
17. Centers for Disease Control and Prevention. Prevalence of regular physical activity
among adults--United States, 2001 and 2005. MMWR. 2007;56:1209-1212.
18. Cerin E, Vandelanotte C, Leslie E, Merom D. Recreational Facilities and Leisure-
Time Physical Activity: An Analysis of Moderators and Self-Efficacy as a Mediator.
Health Psychology. 2008;27:S126-S135.
19. Cerin E, Saelens BE, Sallis JF, Frank LD. Neighborhood Environment Walkability
Scale: validity and development of a short form. Medicine & Science in Sports &
Exercise. 2006;38:1682-1691.
20. Chillon P, Evenson KR, Vaughn A, Ward DS. A systematic review of interventions
for promoting active transportation to school. International Journal of Behavioral
Nutrition and Physical Activity. 2011;8:10.
21. Cleland V, Ball K, Hume C, Timperio A, King AC, Crawford D. Individual, social
and environmental correlates of physical activity among women living in
socioeconomically disadvantaged neighbourhoods. Soc Sci Med. 2010;70:2011-2018.
22. Cohen DA, Finch BK, Bower A, Sastry N. Collective efficacy and obesity: The
potential influence of social factors on health. Soc Sci Med. 2006;62:769-778.
23. Cole DA, Maxwell SE. Testing Mediational Models With Longitudinal Data:
Questions and Tips in the Use of Structural Equation Modeling. J Abnorm Psychol.
2003;112:558-558-577.
110

24. Committee on Environmental Health. The built environment: Designing communities
to promote physical activity in children. Pediatrics. 2009;123:1591-1598.
25. Courneya KS, McAuley E. Cognitive Mediators of the Social Influence-Exercise
Adherence Relationship: A Test of the Theory of Planned Behavior. Journal of
Behavioral Medicine. 1995;18:499-515.
26. Cradock AL, Kawachi I, Colditz GA, Gortmaker SL, Buka SL. Neighborhood social
cohesion and youth participation in physical activity in Chicago. Soc Sci Med.
2009;68:427-435.
27. Davison KK, Werder JL, Lawson CT. Children's active commuting to school: current
knowledge and future directions. Preventing Chronic Disease. 2008;5:A100.
28. Day K, Boarnet M, Alfonzo M, Forsyth A. The Irvine-Minnesota inventory to
measure built environments: development. Am J Prev Med. 2006;30:144-152.
29. de Bruijn G, Kremers SPJ, Lensvelt-Mulders G, de Vries H, van Mechelen W, Brug
J. Modeling Individual and Physical Environmental Factors with Adolescent Physical
Activity. Am J Prev Med. 2006;30:507-512.
30. Dean RN, Farrell JM, Kelley ML, Taylor MJ, Rhodes RE. Testing the Efficacy of the
Theory of Planned Behavior to Explain Strength Training in Older Adults. Journal of
Aging and Physical Activity. 2006;15:1-12.
31. Dietz WH. Childhood weight affects adult morbidity and mortality. J Nutr.
1998;128:411S-414S.
32. Ding D, Sallis JF, Conway TL, et al. Interactive effects of built environment and
psychosocial attributes on physical activity: a test of ecological models. Annals of
Behavioral Medicine. 2012;43:s152.
33. Dowda M, Dishman R, Porter D, Saunders R, Pate R. Commercial Facilities, Social
Cognitive Variables, and Physical Activity of 12th Grade Girls. Annals of Behavioral
Medicine. Springer New York; 2009;37:77-87.
34. Durand CP, Bluthenthal R, MacDonald J. Correlates of local shopping behavior. . In
preparation.
35. Durand CP, Dunton GF, Spruijt-Metz D, Pentz MA. Does Community Type
Moderate the Relationship Between Parent Perceptions of the Neighborhood and Physical
Activity in Children? AJHP.
111

36. Durand CP, Andalib M, Dunton GF, Wolch J, Pentz MA. A systematic review of
built environment factors related to physical activity and obesity risk: implications for
smart growth urban planning. Obesity Reviews. 2011;12:e173-e182.
37. Eveland WP, Hayes AF, Shah DV, Kwak N. Understanding the Relationship
Between Communication and Political Knowledge: A Model Comparison Approach
Using Panel Data. Political Communication. 2005;22:423-446.
38. Federal Bureau of Investigation. Crime in the United States 2010. Washington, DC:
U.S. Government Printing Office; 2010. Available from: http://www.fbi.gov/about-
us/cjis/ucr/crime-in-the-u.s/2010/crime-in-the-u.s.-2010.
39. Field AE, Coakley EH, Must A, et al. Impact of Overweight on the Risk of
Developing Common Chronic Diseases During a 10-Year Period. Arch Intern Med.
2001;161:1581-1586.
40. Florida Department of Transportation Office of Policy Planning. Annual Roadway
Construction Costs. Tallahassee, FL: State of Florida; 2011. Available from:
http://www.dot.state.fl.us/planning/policy/costs/costs-D3.pdf.
41. Foster S, Giles-Corti B. The built environment, neighborhood crime and constrained
physical activity: An exploration of inconsistent findings. Prev Med. 2008;47:241-251.
42. Frank LD, Saelens BE, Powell KE, Chapman JE. Stepping towards causation: do
built environments or neighborhood and travel preferences explain physical activity,
driving, and obesity?. Soc Sci Med. 2007;65:1898-1914.
43. Gebel K, Bauman A, Owen N. Correlates of Non-Concordance between Perceived
and Objective Measures of Walkability. Annals of Behavioral Medicine. 2009;37:228-
238.
44. Gil-Trejo L, Otis E. RAND 2009-2010 Second Cross Section Survey: Method and
Technical Approach to Data Collection. Social Science Research Center, California State
University, Fullerton; 2010.
45. Glanz K, Rimer BK, Lewis FM, eds. Health Behavior and Health Education: Theory,
Research and Practice. 3rd ed. San Francisco: Jossey-Bass; 2002.
46. Glanz K, Rimer BK, Lewis FM, eds. Models of Individual Health Behavior. 3rd ed.
San Francisco, CA: Jossey-Bass; 2002Health Behavior and Health Education.
47. Grunberg L, Moore S, Sikora P, Greenberg E. Downsizing and Alcohol Use: A
Cross-lagged Longitudinal Analysis. Working Paper. Institute of Behavioral Science,
University of Colorado at Boulder; 2006;PEC2006-0002.
112

48. Handy S, Cao XY, Mokhtarian PL. Self-selection in the relationship between the built
environment and walking - Empirical evidence from northern California. Journal of the
American Planning Association. 2006;72:55-74.
49. Hipp JR. Resident Perceptions of Crime and Disorder: How Much is "Bias," and How
Much is Social Environment Differences? Criminology. 2010;48:475-508.
50. Hipp JR. Assessing Crime as a Problem: The Relationship Between Residents'
Perception of Crime and Official Crime Rates Over 25 Years. Crime & Delinquency.
2010.
51. Hoehner CM, Brennan Ramirez LK, Elliott MB, Handy SL, Brownson RC. Perceived
and objective environmental measures and physical activity among urban adults. Am J
Prev Med. 2005;28:105-116.
52. Hu L, Bentler PM. Cutoff criterion for fit indexes in covariance structure analysis:
conventional criteria versus new alternatives. Structural Equation Modeling. 1999;6:1-
55.
53. Ishii K, Shibata A, Oka K. Environmental, psychological, and social influences on
physical activity among Japanese adults: structural equation modeling analysis.
International Journal of Behavioral Nutrition and Physical Activity. 2010;7:61.
54. Jago R, Baranowski T, Harris M. Relationships Between GIS Environmental Features
and Adolescent Male Physical Activity: GIS Coding Differences. Journal of Physical
Activity & Health. 2006;3:230-242.
55. Kaczynski AT, Henderson KA. Parks and recreation settings and active living: a
review of associations with physical activity function and intensity. Journal of Physical
Activity & Health. 2008;5:619-632.
56. Kaczynski AT. Neighborhood Walkability Perceptions: Associations With Amount of
Neighborhood-Based Physical Activity by Intensity and Purpose. Journal of Physical
Activity & Health. 2010;7:3-10.
57. Kamphuis CBM, Mackenbach JP, Giskes K, Huisman M, Brug J, van Lenthe FJ.
Why Do Poor People Perceive Poor Neighborhoods? The Role of Objective
Neighborhood Features and Psychosocial Factors. Health and Place. 2010;16:744-754.
58. Kamphuis C, van Lenthe F, Giskes K, Huisman M, Brug J, Mackenbach J.
Socioeconomic differences in lack of recreational walking among older adults: the role of
neighbourhood and individual factors. International Journal of Behavioral Nutrition and
Physical Activity. 2009;6:1.
113

59. Kinder Institute for Urban Research. The Kinder Houston Area Survey-2012.
Houston: Rice University; 2012. Available from:
http://has.rice.edu/uploadedFiles/Houston_Area_Survey/2012%20Kinder%20Houston%2
0Area%20Survey.pdf.
60. Kline RB. Principles and Practice of Structural Equation Modeling. 2nd ed. New
York, NY: The Guilford Press; 2005.
61. Kremers SPJ, de Bruijn GJ, Visscher TLS, van Mechelen W, de Vries NK, Brug J.
Environmental influences on energy balance-related behaviors: A dual-process view.
International Journal of Behavioral Nutrition and Physical Activity. 2006;3.
62. Lavie CJ, Milani RV, Ventura HO. Obesity and Cardiovascular Disease: Risk Factor,
Paradox, and Impact of Weight Loss. J Am Coll Cardiol. 2009;53:1925-1932.
63. Lee NE, De AK, Simon PA. School-based physical fitness testing identifies large
disparities in childhood overweight in Los Angeles. J Am Diet Assoc. 2006;106:118-121.
64. Lee T, Marans RW. Objective and Subjective Indicators: Effects of Scale
Discordance on Interrelationships. Social Indicators Research. 1980;8:47-64.
65. Lee C. Environment and active living: The roles of health risk and economic factors.
American Journal of Health Promotion. 2007;21:293-304.
66. Lee C, Moudon AV. Correlates of Walking for Transportation or Recreation
Purposes. Journal of Physical Activity & Health. 2006;3:S77-S98.
67. Lemieux M, Godin G. How well do cognitive and environmental variables predict
active commuting? International Journal of Behavioral Nutrition and Physical Activity.
2009;6:12.
68. Lewis BA, Marcus BH, Pate RR, Dunn AL. Psychosocial Mediators of Physical
Activity Behavior Among Adults and Children. American Journal of Preventive
Medicine. 2002;23:26-35.
69. Lewis DA, Maxfield MG. Fear in the Neighborhoods: an Investigation of the Impact
of Crime. Journal of Research in Crime and Delinquency. 1980;17:160-189.
70. Lewis FM. Perspectives on models of interpersonal health behavior. In: Glanz K,
Rimer BK, Lewis FM, eds. Health Behavior and Health Education. 3rd ed. San
Francisco, CA: Jossey-Bass; 2002.
114

71. Linden-Bostrom M, Persson C, Eriksson C. Neighbourhood characteristics, social
capital and self-rated health - A population-based survey in Sweden. BMC Public Health.
2010;10:628.
72. Little TD, Preacher KJ, Selig JP, Card NA. New developments in latent variable
panel analyses of longitudinal data. International Journal of Behavioral Development.
2007;31:357-365.
73. Ludwig J, Sanbonmatsu L, Gennetian L, et al. Neighborhoods, Obesity, and Diabetes
— A Randomized Social Experiment. N Engl J Med. 2011;365:1509-1519.
74. MacKinnon DP, Lockwood CM, Hoffman JM, West SG, Sheets V. A comparison of
methods to test mediation and other intervening variable effects. Psychol Methods.
2002;7:83-104.
75. Maddison R, Hoorn S, Jiang Y, et al. The environment and physical activity: The
influence of psychosocial, perceived and built environmental factors. International
Journal of Behavioral Nutrition and Physical Activity. 2009;6:19.
76. Maffeis C, TatÃ² L. Long-Term Effects of Childhood Obesity on Morbidity and
Mortality. Horm Res. 2001;55:42-45.
77. Mallinckrodt B, Abraham WT, Wei M, Russell DW. Advances in testing the
statistical significance of mediation effects. Journal of Counseling Psychology.
2006;53:372-378.
78. Maxwell SE, Cole DA. Bias in Cross-Sectional Analyses of Longitudinal Mediation.
Psychol Methods. 2007;12:23-23-44.
79. McCormack GR, Spence JC, Berry T, Doyle-Baker P. Does Perceived Behavioral
Control Mediate the Association Between Perceptions of Neighborhood Walkability and
Moderate- and Vigorous-Intensity Leisure-Time Physical Activity? Journal of Physical
Activity & Health. 2009;6:657-666.
80. McCrea R, Shyy T, Stimson R. What is the Strength of the Link Between Objective
and Subjective Indicators of Urban Quality of Life? Applied Research in Quality of Life.
2006;1:79-96.
81. McGinn AP, Evenson KR, Herring AH, Huston SL, Rodriguez DA. Exploring
associations between physical activity and perceived and objective measures of the built
environment. Journal of Urban Health. 2007;84:162-184.
82. McNeill LH, Kreuter MW, Subramanian SV. Social environment and physical
activity: a review of concepts and evidence. Soc Sci Med. 2006;63:1011-1022.
115

83. McNeill LH, Wyrwich KW, Brownson RC, Clark EM, Kreuter MW. Individual,
social environmental, and physical environmental influences on physical activity among
Black and White adults: A structural equation analysis. Annals of Behavioral Medicine.
2006;31:36-44.
84. Mendoza JA, Watson K, Baranowski T, Nicklas TA, Uscanga DK, Hanfling MJ. The
Walking School Bus and Children's Physical Activity: A Pilot Cluster Randomized
Controlled Trial. Pediatrics. 2011;128:e537-e544.
85. Montano DE, Kasprzyk D. The theory of reasoned action and the theory of planned
behavior. In: Glanz K, Rimer BK, Lewis FM, eds. Health Behavior and Health
Education. 3rd ed. San Francisco: Jossey-Bass; 2002:67-98.
86. Morris K, McAuley E, Motl R. Neighborhood satisfaction, functional limitations, and
self-efficacy influences on physical activity in older women. International Journal of
Behavioral Nutrition and Physical Activity. 2008;5:13.
87. Motl RW, Dishman RK, Ward DS, Saunders RP, Dowda M, Felton G.
Perceived physical environment and physical activity across one year among adolescent
girls: Self-efficacy as a possible mediator? Journal of Adolescent Health. 2005;37:403-
408.
88. Motl RW, Dishman RK, Saunders RP, Dowda M, Pate RR. Perceptions of Physical
and Social Environment Variables and Self-Efficacy as Correlates of Self-Reported
Physical Activity Among Adolescent Girls. Journal of Pediatric Psychology. ;32:6-12.
89. Mummery WK, Lauder W, Schofield G, Caperchione C. Associations between
physical inactivity and a measure of social capital in a sample of Queensland adults.
Journal of Science and Medicine in Sport. 2008;11:308-315.
90. National Center for Health Statistics. Prevalence of overweight, obesity and extreme
obesity among adults: United States, trends 1976-80 through 2005-06. Hyattsville, MD:
National Center for Health Statistics; 2008 Accessed 4-17-09.
91. Nylund KL, Asparouhov T, Muthen BO. Deciding on the number of classes in latent
class analysis and growth mixture modeling. A Monte Carlo simulation study. Structural
Equation Modeling. 2007;14:535-569.
92. Ogden CL. Disparities in obesity prevalence in the United States: black women at
risk. Am J Clin Nutr. 2009;89:1001-1002.
93. Owen N, Humpel N, Leslie E, Bauman A, Sallis J. Understanding environmental
influences on walking: review and research agenda. Am J Prev Med. 2004;27:67-76.
116

94. Pan S, Cameron C, DesMeules M, Morrison H, Craig C, Jiang X. Individual, social,
environmental, and physical environmental correlates with physical activity among
Canadians: a cross-sectional study. BMC Public Health. 2009;9:21.
95. Parra DC, Gomez LF, Sarmiento OL, et al. Perceived and objective neighborhood
environment attributes and health related quality of life among the elderly in Bogotá,
Colombia. Soc Sci Med. 2010;70:1070-1076.
96. Pentz MA. Prevention aimed at individuals: An integrative transactional perspective.
In: McCrady BS, Epstein EE, eds. Addictions: A Comprehensive Guidebook for
Practitioners. New York: Oxford University Press; 1999:555-572.
97. Petras H, Masyn K. General growth mixture analysis with antecedents and
consequences of change. In: Piquero A, Weisburd D, eds. Handbook of Quantitative
Criminology. 1st ed. New York: Springer; 2010:69-100.
98. Plotnikoff R, Pickering M, Rhodes R, Courneya K, Spence J. A test of cognitive
mediation in a 12-month physical activity workplace intervention: does it explain
behaviour change in women? International Journal of Behavioral Nutrition and Physical
Activity. 2010;7:32.
99. Plotnikoff RC, Pickering MA, Flaman LM, Spence JC. The Role of Self-Efficacy on
the Relationship Between the Workplace Environment and Physical Activity: A
Longitudinal Mediation Analysis. Health Education & Behavior. 2010;37:170-185.
100. Pont K, Ziviani J, Wadley D, Bennett S, Abbott R. Environmental correlates of
children’s active transportation: A systematic literature review. Health and Place.
2009;15:849-862.
101. Prodaniuk T, Plotnikoff R, Spence J, Wilson P. The influence of self-efficacy and
outcome expectations on the relationship between perceived environment and physical
activity in the workplace. International Journal of Behavioral Nutrition and Physical
Activity. 2004;1:7.
102. Prospective Studies Collaboration. Body-mass index and cause-specific mortality in
900 000 adults: collaborative analyses of 57 prospective studies. The Lancet.
2009;373:1083-1096.
103. Raftery AE. Bayesian Model Selection in Social Research. Sociological
Methodology. 1995;25:111-163.
104. Rhodes R, Pfaeffli L. Mediators of physical activity behaviour change among adult
non-clinical populations: a review update. International Journal of Behavioral Nutrition
and Physical Activity. 2010;7:37.
117

105. Rhodes R, Courneya K, Blanchard C, Plotnikoff R. Prediction of leisure-time
walking: an integration of social cognitive, perceived environmental, and personality
factors. International Journal of Behavioral Nutrition and Physical Activity. 2007;4:51.
106. Robinson WR, Gordon-Larsen P, Kaufman JS, Suchindran CM, Stevens J. The
female-male disparity in obesity prevalence among black American young adults:
contributions of sociodemographic characteristics of the childhood family. Am J Clin
Nutr. 2009;89:1204-1212.
107. Rodriguez DA, Aytur S, Forsyth A, Oakes JM, Clifton KJ. Relation of modifiable
neighborhood attributes to walking. Prev Med. 2008;47:260-264.
108. Roman CG, Chalfin A. Fear of walking outdoors. A multilevel ecologic analysis of
crime and disorder. Am J Prev Med. 2008;34:306-312.
109. Roman CG, Chalfin A. Fear of Walking Outdoors: A Multilevel Ecologic Analysis
of Crime and Disorder. Am J Prev Med. 2008;34:306-312.
110. Rovniak LS, Hovell MF, Wojcik JR, Winett RA, Martinez-Donate AP. Enhancing
theoretical fidelity: an e-mail-based walking program demonstration. American Journal
of Health Promotion. 2005;20:85-95.
111. Saelens BE, Sallis JF, Black J, Chen D. Measuring perceived neighborhood
environment factors related to walking/cycling. Annals of Behavioral Medicine.
2002;24:S139.
112. Saelens B, Sallis J, Frank L. Environmental correlates of walking and cycling:
findings from the transportation, urban design and planning literatures. Ann Behav Med.
2003;25:80-91.
113. Saelens BE, Handy SL. Built environment correlates of walking: a review. Medicine
& Science in Sports & Exercise. 2008;40:S550-66.
114. Saelens BE, Sallis JF, Black JB, Chen D. Neighborhood-based differences in
physical activity: an environment scale evaluation. Am J Public Health. 2003;93:1552-
1558.
115. Sallis JF, Owen N. Ecological models of health behavior. In: Glanz K, Rimer BK,
Lewis FM, eds. Health Behavior and Health Education. 3rd ed. San Francisco: Jossey-
Bass; 2002:462-484.
116. Sampson RJ, Raudenbush SW. Seeing Disorder: Neighborhood Stigma and the
Social Construction of "Broken Windows". Social Psychology Quarterly. 2004;67:319-
342.
118

117. Sampson RJ, Raudenbush SW, Earls F. Neighborhoods and violent crime: a
multilevel study of collective efficacy. Science. 1997;277:918-924.
118. Schubert CM, Rogers NL, Remsberg KE, et al. Lipids, lipoproteins, lifestyle,
adiposity and fat-free mass during middle age: the Fels Longitudinal Study. Int J Obes
Relat Metab Disord. 2005;30:251-260.
119. Seidel A, Presnell K, Rosenfield D. Mediators in the dissonance eating disorder
prevention program. Behav Res Ther. 2009;47:645-653.
120. Sherwood NE, Jeffery RW. THE BEHAVIORAL DETERMINANTS OF
EXERCISE: Implications for Physical Activity Interventions. Annu Rev Nutr.
2000;20:21-44.
121. Stacy AW, Ames SW, Wiers RW, & Krank M. Associative memory in appetitive
behavior: Framework and relevance to epidemiology and prevention. In: L. M. Scheier,
ed. Handbook of Drug use Etiology. Washington, D.C.: APA Books; 2010.
122. Stacy AW, Ames SL, Knowlton BJ. Neurologically Plausible Distinctions in
Cognition Relevant to Drug Use Etiology and Prevention. Substance Use & Misuse.
2004;39:1571-1623.
123. Taris TW, Kompier MA. Games researchers play--extreme-groups analysis and
mediation analysis in longitudinal occupational health research. Scand J Work Environ
Health. 2006;32:463-472.
124. Troiano RP, Berrigan D, Dodd KW, Masse LC, Tilert T, Mcdowell M. Physical
Activity in the United States Measured by Accelerometer. Medicine & Science in Sports
& Exercise. 2008;40:181-188.
125. Trost SG, Owen N, Bauman AE. Correlates of adults participation in physical
activity: review and update. Med Sci Sports Exerc. 2002;34:1996-2001.
126. Tudor-Locke C, Brashear MM, Johnson WD, Katzmarzyk PT. Accelerometer
profiles of physical activity and inactivity in normal weight, overweight, and obese U.S.
men and women. Int J Behav Nutr Phys Act. 2010;7:60.
127. Tudor-Locke C, Ainsworth BE, Popkin BM. Active commuting to school: an
overlooked source of childrens' physical activity? Sports Medicine. 2001;31:309-313.
128. U.S. Department of Health and Human Services. 2008 Physical Activity Guidelines
for Americans. Washington, D.C.: U.S. Department of Health and Human Services;
2008;U0036 Accessed 4-17-09.
119

129. Ueshima K, Fujiwara T, Takao S, et al. Does Social Capital Promote Physical
Activity? A Population-Based Study in Japan. PLoS ONE. 2010;5:e12135.
130. Van Der Heijden BIJM, Demerouti E, Bakker AB, The NEXT Study Group
coordinated by Hans-Martin Hasselhorn. Work-home interference among nurses:
reciprocal relationships with job demands and health. J Adv Nurs. 2008;62:572-584.
131. van der Horst K, Oenema A, te Velde SJ, Brug J. Do Individual Cognitions Mediate
the Association of Socio-cultural and Physical Environmental Factors with Adolescent
Sports Participation? Public Health Nutrition. 2010;13:1746-1754.
132. van Stralen M, de Vries H, Mudde A, Bolman C, Lechner L. The working
mechanisms of an environmentally tailored physical activity intervention for older adults:
a randomized controlled trial. International Journal of Behavioral Nutrition and Physical
Activity. 2009;6:83.
133. Wandersman A, Nation M. Urban neighborhoods and mental health: Psychological
contributions to understanding toxicity, resilience, and interventions. American
Psychologist. 1998;53:647-656.
134. Wang Y, Beydoun MA. The obesity epidemic in the United States--gender, age,
socioeconomic, racial/ethnic, and geographic characteristics: a systematic review and
meta-regression analysis. Epidemiol Rev. 2007;29:6-28.
135. Warr M. Fear of crime in the united states: Avenues for research and policy. In:
Duffee D, ed. Measurement and Analysis of Crime and Justice. Vol 4. Washington, D.C.:
National Institute of Justice; 2000:451-489.
136. Weden MM, Carpiano RM, Robert SA. Subjective and Objective Neighborhood
Characteristics and Adult Health. Social Science and Medicine. 2008:1256-1270.
137. Wiers RW, de Jong PJ, Havermans R, Jelicic M. How to Change Implicit Drug Use-
Related Cognitions in Prevention: A Transdisciplinary Integration of Findings from
Experimental Psychopathology, Social Cognition, Memory, and Experimental Learning
Psychology. Substance Use & Misuse. 2004;39:1625-1684.
138. Zapf D, Dormann C, Frese M. Longitudinal studies in organizational stress research:
A review of the literature with reference to methodological issues. J Occup Health
Psychol. 1996;1:145-145-169.
139. Zhang R, Reisin E. Obesity-hypertension: the effects on cardiovascular and renal
systems. American Journal of Hypertension. 2000;13:1308-1314.

120

APPENDIX: MODEL TESTING FOR LIGHT ACTIVITY
Table A1: Structural Model Fit Statistics and Indices for Light Activity
Model
set
Model
#
AIC SS-BIC Chi2 DF Scaling
factor
from
H0
RMSEA (CI) CFI SRMR
A 1 62794.677 63024.356 3419.562 2298 2.360 0.04
(0.037-
0.042)
0.892 0.061
A 2 62819.543 63060.648 3407.862 2278 2.300 0.04
(0.037-
0.043)
0.891 0.061
A 3 62813.208 63054.313 3395.769 2278 2.292 0.04
(0.037-
0.042)
0.893 0.061
A 4 62840.062 63092.594 3386.321 2258 2.237 0.04
(0.037-
0.043)
0.892 0.060
B 5 62794.677 63024.356 3419.562 2298 2.360 0.04
(0.037-
0.042)
0.892 0.061
B 6 62816.916 63058.022 3400.745 2278 2.293 0.04
(0.037-
0.042)
0.892 0.061
B 7 62813.955 63055.060 3401.911 2278 2.300 0.04
(0.037-
0.043)
0.892 0.060
B 8 62836.093 63088.625 3382.682 2258 2.238 0.04
(0.037-
0.043)
0.892 0.060
9 62847.503 63104.605 3379.077 2250 2.216 0.04
(0.037-
0.043)
0.892 0.060

Table A2: Nested Model Comparisons for Light Activity
Model set Model comparison ∆ χ
2
( ∆ D F ) P value
A 4 v. 3 24.455(20) 0.223
A 4 v. 2 27.928(20) 0.111
A 4 v. 1 53.203(40) 0.079
A 3 v. 1 28.408(20) 0.100
A 2 v. 1 25.243(20) 0.192
B 8 v. 7 27.307(20) 0.127
B 8 v. 6 26.753(20) 0.142
B 8 v. 5 54.042(40) 0.068
B 7 v. 5 26.732(20) 0.143
B 6 v. 5 27.250(20) 0.128
9 v. best set A model (#1) 63.900(48) 0.062
9 v. best set B model (#5) 63.900(48) 0.062

Abstract (if available)

Abstract The purpose of this dissertation is to examine multiple methods of measuring the neighborhood environment, and how those methods may differentially impact physical activity in adults and children. The first paper examines the correlates of discordance between perceived and objective measures of neighborhood crime safety and physical disorder. The second paper uses neighborhood audits and resident reports to develop distinct neighborhood typologies, and then correlates those typologies with active commuting to school by children. The third paper uses longitudinal data to test potential causal mechanisms linking psychosocial factors, perceptions of the built environment, and physical activity in adults. ❧ In the first paper, social control was consistently associated with perceiving the neighborhood more positively than objective statistics indicated. Hispanic ethnicity, time spent in the neighborhood, and acculturation were associated with perceiving the neighborhood more negatively than objective data indicated. ❧ In the second paper, both assessments methods revealed a profile characterized by moderate levels on all environmental variables. This profile was associated with the highest probability of actively commuting to school, while a profile characterized by low-density single-family residential development was associated with the lowest probability. ❧ In the third paper, the overall best fitting model was one in which the perceived built environment affected intra- and inter-personal factors, which in turn affected MVPA. However, there was no evidence of any mediation effects within this model. ❧ Taken together, the results of the three papers indicate that our understanding of the environment is method dependent

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Understanding the methodological limitations In the ecological momentary assessment of physical activity

PDF

Effects of parent stress on weight-related parenting practices and child obesity risk

PDF

Understanding the dynamic relationships between physical activity and affective states using real-time data capture techniques

PDF

Mixed methods investigation of user engagement with a smoking cessation app

PDF

Influences of specific environmental domains on childhood obesity and related behaviors

PDF

The acute and longitudinal associations between sedentary behaviors, affective states, and emotional disorder symptoms among youth

PDF

Investigating daily effects of activity space-based built environment exposures on physical activity behaviors in Hispanic women during pregnancy and early postpartum

PDF

Psychosocial and cultural factors in the primary prevention of melanoma targeted to multiethnic children

PDF

The Internet activities, gratifications, and health consequences related to compulsive Internet use

PDF

The built environment, tour complexity, and active travel

PDF

Applications of mindfulness toward promoting psychosocial health in diverse communities

PDF

Understanding normative influence of neighborhoods: a multilevel approach to promoting Latinas’ cervical cancer prevention behaviors in urban ethnic communities

PDF

Effects of sugar and fiber consumption in minority adolescents and self-tracking as a potential dietary intervention tool

PDF

Developing and testing novel strategies to detect inattentive responding in ecological momentary assessment studies

PDF

The impact of the environment on childhood allergic rhinitis: findings from the Children’s Health Study

PDF

The acute relationship between affective states and physiological stress response, and the moderating role of moderate-to-vigorous physical activity

PDF

Free-living and in-lab effects of sedentary time on cardiac autonomic nervous system function in youth with overweight and obesity

PDF

Factors and correlates of sexual behaviors among young adults from continuation high schools

PDF

Problematic alcohol use in Hispanic emerging adults: the role of perceived discrimination, cultural identity, and salient cultural values

PDF

A network analysis of online and offline social influence processes in relation to adolescent smoking and alcohol use

Asset Metadata

Creator Durand, Casey Philip (author)

Core Title Effects of the perceived and objectively assessed environment on physical activity in adults and children

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Preventive Medicine (Health Behavior)

Publication Date 08/09/2012

Defense Date 06/14/2012

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag built environment,environment,OAI-PMH Harvest,objective,perceived,physical activity,smart growth,structural equation modeling

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Pentz, Mary Ann (committee chair), Bluthenthal, Ricky (committee member), Dunton, Genevieve F. (committee member), Huh, Jimi (committee member), Sloane, David C. (committee member)

Creator Email casey.p.durand@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-89464

Unique identifier UC11289459

Identifier usctheses-c3-89464 (legacy record id)

Legacy Identifier etd-DurandCase-1150.pdf

Dmrecord 89464

Document Type Dissertation

Rights Durand, Casey Philip

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA