University of Southern California Dissertations and Theses

Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

(USC Thesis Other)

Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

UNIVERSITY OF SOUTHERN CALIFORNIA

Healthy Mobility: Untangling the relationships between the built
environment, travel behavior, and environmental health

by

E-Sok Andy Hong

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA

In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
URBAN PLANNING AND DEVELOPMENT

Committee in charge:
Professor Marlon G. Boarnet, Chair
Professor Lisa Schweitzer
Professor Scott Fruin

August 2016

Healthy Mobility: Untangling the relationships between the built environment, travel behavior,
and environmental health

Copyright © 2016 by E-Sok Andy Hong

All rights reserved. Printed in the United States of America. No part of this dissertation may be
used or reproduced in any manner whatsoever without written permission except in the case of
brief quotations embodied in critical articles or reviews.

Healthy Mobility
Untangling the relationships between the built environment,
travel behavior, and environmental health

E-Sok Andy Hong

i
First and foremost, I praise and give thanks to God for all his blessings throughout my work. I
wish to express sincere appreciation to Professor Marlon Boarnet for his tremendous support and
patience and to Professor Scott Fruin for his enthusiastic support and help for data collection and
analysis. I would like to particularly give special thanks to Professor Lisa Schweitzer for providing
valuable advice and insights, which significantly helped my research come to fruition. I would
also like to thank Suresh Ratnam for his extensive volunteering work in mobile measurements for
the CicLAvia study, and I thank all the volunteers, Mercede Ramjerd, Yichang Chen, Ivan Torres,
Jenny Song, and Dan Huynh, who provided tremendous help for the CicLAvia sampling work.
Most importantly, none of this would have been possible without the love, encouragement, and
devotion of my wife, Rosa, and my daughter, Emily, both of whom are a constant source of my
joy and happiness. I am extremely grateful to my family, especially, my father, Sang Eun Hong,
my mother, Mi Hyun Cho, my father-in-law, Sung Kyu Kim, and my mother-in-law, Hee Ja Choi,
who have given me with the love, prayers, patience, and encouragement throughout this endeavor.
Finally, my thanks go to all my colleagues, friends, and significant others who have supported me
to complete this work directly or indirectly.

Los Angeles, California
E-Sok Andy Hong
August, 2016

ii

The built environment forms the basis of physical and social environment where individuals and
groups perform all functions of life. However, researchers increasingly understand that the current
built environment and transportation systems can create unhealthy and unsustainable condition.
Low-density and segregated land uses, and poor urban design features deter walking and bicycling,
and increase reliance on fossil fuel and automobile. Further, poor families living in a racially
segregated neighborhood often lack healthy food options and face more environmental burdens.
These problems vis-à-vis the built environment and health are the quintessential example of wicked
problems where the relationships are complex and operate through multiple confounders. Given
this challenge, my dissertation, comprised of three empirical essays, expands on our understanding
of the relationships between the built environment, transportation, and health by using
experimental research design and original data collected on the individual level.
The first essay uses panel data to investigate the before-and-after impact of a new light rail
transit line on active travel behavior. The panel design provides an opportunity for stronger causal
inference than is possible in the much more common study designs that use cross-sectional data.
It also provides an opportunity to examine how an individual’s previous activity behavior
influences the role that new light rail transit access plays in promoting active travel behavior. The
results show that, when not controlling for subject’s before-opening walking or physical activity,
there was no significant relationship between treatment group status and after-opening walking or
physical activity levels. However, when controlling for an interaction between baseline walking
or physical activity levels and treatment group membership, I found that living within a half-mile
of a transit station was associated with an increase in walking and physical activity for the subjects

iii
who were more sedentary before the new light rail opened. The results imply that future policy
and research should consider the possibility that sedentary populations may be more responsive to
new transit investments, and a different market segment approach based on people’s activity levels
would be needed to promote healthy travel choices.
The second essay examines the effects of mode switching behavior from car to light rail
transit (LRT) on air pollution exposure. Few studies have systematically examined under what
circumstances traffic exposure changes when switching from car to LRT. In this study, we
conducted a real-time air pollution measurement while one technician was driving a car and
another technician was taking an LRT at the same time for 20 weekdays (a total of 80 trips).
Simultaneous measurement allowed us to control for daily variations in meteorological condition.
Further, we conducted a robustness check to test the effects of other vehicle-specific factors, such
as fan strength, vehicle speed, and vehicle age. The results indicate that traffic exposure was
significantly altered by ventilation status, and traffic microenvironment and other vehicle-specific
factors were not strong enough to outweigh the effect of ventilation. The results suggest that mode
shift from automobile to LRT will be particularly beneficial to those owning an older vehicle with
a sub-par ventilation system. Moreover, the results imply that early retirement of old vehicle stocks
and transit subsidy programs targeted to low income households could help poor families avoid
exposure to higher traffic emissions.
In the third essay, I examine how air pollution levels change during a car-free street event,
called CicLAvia. Few studies have examined how car-free street events affect localized air
pollutant concentrations. Using an instrumented mobile monitoring platform, we measured
concentrations of particulate matter less than 10 and 2.5 microns in size, ultrafine particle number,
black carbon, and particle-bound polycyclic aromatic compounds in three different urban locations

iv

in Los Angeles on Sundays before, during and after CicLAvia Sunday. Meteorological conditions
were similar across the Sundays measured, and most pollutants did not show statistically
significant changes despite an increase in traffic congestion generated by the event. These results
suggest that more careful traffic management are needed to reduce disruptions to normal traffic
flow during special events like CicLAvia. Furthermore, the results suggest that continuous
technological improvements in gasoline-powered vehicles now allow minimal air quality impacts
from changes in traffic congestion. This implies that technology-driven solutions could be
effective at managing urban air pollution problems in the long run.
In conclusion, the three essays provide empirical evidence to show that the health effects
of the built environment are not only affected by the “place” but also by the "people” that interact
with the built environment. These findings led me to conclude that the environmental health
impacts are the results of the dynamic interactions between the built environment and human
behavior. The findings also suggest that health impact assessment of transportation systems
requires a close examination of multiple pathways that link the built environment and health.
Uncovering these pathways would require new research framework and more fine-grained data
collection methods that are capable of capturing human behavior and environmental patterns that
occur at multiple spatial and temporal scales. The purpose of my dissertation is to help develop a
new methodology and to inform existing policy and planning to consider a broad range of
behavioral and environmental factors related to health outcomes. To this end, I hope this
dissertation will help us move forward with a paradigm shift to healthy mobility.

v

1. Built environment and health: A brief history ....................................................................... 3
2. Conceptual approach of this dissertation ............................................................................... 5
3. Built environment and travel behavior ................................................................................... 8
3.1. Land use and travel behavior ............................................................................................ 9
3.2. Physical activity and active living research .................................................................... 10
3.3. Advances in transportation modeling ............................................................................. 11
3.4. Summary and research needs ......................................................................................... 13
4. Travel behavior and air pollution exposure ......................................................................... 14
4.1. Mode choice and air pollution exposure ......................................................................... 15
4.2. Confounders of personal exposure ................................................................................. 16
4.3. Physical activity and air pollution exposure ................................................................... 17
4.4. Summary and research needs ......................................................................................... 19
5. Built environment and air pollution ..................................................................................... 19
5.1. Air quality modeling and emissions research ................................................................. 20
5.2. Empirical studies using regression-based approach ....................................................... 21
5.3. Accountability research .................................................................................................. 22
5.4. Summary and research needs ......................................................................................... 23
6. Contribution of the dissertation to the current knowledge ................................................... 24
7. The structure of the dissertation ........................................................................................... 26

vi

1. Introduction .......................................................................................................................... 29
2. Literature review .................................................................................................................. 31
2.1. Public transit and active travel behavior......................................................................... 31
2.2. Built environment and physical activity ......................................................................... 32
2.3. Past behavior and habitual process ................................................................................. 32
3. Methodology ........................................................................................................................ 34
3.1. Study area ....................................................................................................................... 34
3.2. Sampling procedure ........................................................................................................ 36
3.3. Measures ......................................................................................................................... 38
3.4. Statistical analysis........................................................................................................... 40
4. Results .................................................................................................................................. 41
4.1. Sample characteristics .................................................................................................... 41
4.2. Walk trip model results ................................................................................................... 43
4.3. Physical activity model results ....................................................................................... 45
4.4. Robustness checks .......................................................................................................... 48
5. Discussion ............................................................................................................................ 52
6. Conclusion and policy implications ..................................................................................... 55
7. Appendix - accelerometer data processing ......................................................................... 58

1. Introduction .......................................................................................................................... 67
2. Methodology ........................................................................................................................ 69
2.1. Site location and description ........................................................................................... 69
2.2. Study design ................................................................................................................... 71
2.3. Study protocol and validation ......................................................................................... 73
2.4. Instrumentation ............................................................................................................... 75
2.5. Data post-processing and instrument validation ............................................................. 77
2.6. Data analysis ................................................................................................................... 79
3. Results .................................................................................................................................. 80
3.1. Measured concentrations ................................................................................................ 80

vii
3.2. ANOVA results .............................................................................................................. 88
3.3. Robustness check ............................................................................................................ 90
4. Discussion ............................................................................................................................ 95
5. Conclusions .......................................................................................................................... 97
6. Supporting material ............................................................................................................ 100

1. Introduction ........................................................................................................................ 123
2. Methodology ...................................................................................................................... 127
2.1. Study area ..................................................................................................................... 127
2.2. Research design ............................................................................................................ 129
2.3. Mobile monitoring ........................................................................................................ 134
2.4. Traffic data ................................................................................................................... 136
2.5. Meteorological data ...................................................................................................... 138
2.6. Data post-processing..................................................................................................... 139
2.7. Data analysis ................................................................................................................. 139
3. Results ................................................................................................................................ 141
3.1. Meteorological trends ................................................................................................... 141
3.2. Overall trends in traffic and pollutants ......................................................................... 144
3.3. Disaggregate analysis of pollutants by traffic counter location ................................... 147
3.4. Spatial patterns of PM10 and UFP ................................................................................. 151
4. Discussion .......................................................................................................................... 153
5. Conclusion ......................................................................................................................... 156
6. Supporting material ............................................................................................................ 159

1. Summary of the findings .................................................................................................... 182
2. Recommendations for future policy and research .............................................................. 184

viii

Figure 1. Conceptual approach of this dissertation ......................................................................... 6
Figure 2. Summary of the major themes in the literature ............................................................... 8
Figure 3. A framework for linking air pollution sources to health effects ................................... 14

Figure 1. Neighborhoods around Western Station (A); and La Cienega Station (B) before and after
the Expo Line, Los Angeles, CA .................................................................................................. 35
Figure 2. The Expo Line study area: Los Angeles, California, 2011–2013 ................................. 36
Figure 3. Differential impact of past behavior on light rail treatment .......................................... 48
Figure 4. Comparison between the original model and the outlier models .................................. 50
Figure 5. Monte Carlo simulation of the random treatment models ............................................. 52

Figure 1. Map of sampling routes, the vehicle route and the train route ...................................... 71
Figure 2. Experimental condition matrix ...................................................................................... 72
Figure 3. Comparison of collocated measurements for (a) PM2.5, (b) BC, (c) UFP #, (d) CO2 ... 79
Figure 4. In-transit and in-vehicle PM2.5 concentration by experimental condition ..................... 83
Figure 5. In-transit and in-vehicle BC concentration by experimental condition......................... 84
Figure 6. In-transit and in-vehicle ultrafine concentration by experimental condition ................ 86

ix
Figure 7. In-transit and in-vehicle CO2 concentration by experimental condition ....................... 87
Figure 8. ANOVA results comparing different travel conditions ................................................. 89
Figure 9. Effect of fan strength on in-vehicle UFP compared to in-transit UFP .......................... 92
Figure 10. Effect of vehicle speed on in-vehicle UFP compared to in-transit UFP ..................... 94
Figure 11. Effect of vehicle age on in-vehicle UFP compared to in-transit UFP ......................... 95

Figure 1. (a) Location of three CicLAvia events; (b) A photo taken during a typical Sunday in
downtown Los Angeles; and (c) A photo taken during the DTLA CicLAvia Sunday ............... 128
Figure 2. Sampling routes of Pasadena CicLAvia ...................................................................... 131
Figure 3. Sampling routes of Culver City CicLAvia .................................................................. 132
Figure 4. Sampling routes of DTLA CicLAvia .......................................................................... 133
Figure 5. Instrument setup (a) mobile monitoring platform; (b) backpack instrument .............. 135
Figure 6. Spatial matching between traffic data and pollutant measurements ........................... 137
Figure 7. Hourly time series of selected weather parameter ....................................................... 143
Figure 8. Hourly time series of traffic and pollutant measurement from the mobile platform... 146
Figure 9. Boxplots of selected pollutants by traffic counter location in Culver City ................. 149
Figure 10. Boxplots of selected pollutants by traffic counter location in DTLA ....................... 150
Figure 11. Heatmaps for the Culver City location ...................................................................... 152
Figure 12. Heatmaps for the DTLA location .............................................................................. 153

x

Table 1. Descriptive statistics of study participants ...................................................................... 43
Table 2. Poisson and negative binomial models of total walk trip counts .................................... 45
Table 3. Linear regression models of moderate-and-vigorous physical activity .......................... 46

Table 1. Summary of instruments used in the study ..................................................................... 75
Table 2. Descriptive summary of the pollutant measurements ..................................................... 82
Table 3. I/O ratios for RC and OA under various conditions ....................................................... 91

Table 1. Monitoring instruments employed in the field measurement ....................................... 135

1

2

s a regular bicycle commuter, I always wondered why it has to be so stressful and
unpleasant to bike to work. One of the most disturbing things that faces bicyclists
on a daily basis is exposure to vehicular emissions while riding on a busy street.
Despite health benefits of increased physical activity through regular bicycling, exposure to traffic
exhausts while bicycling makes urban bicycling less healthy and enjoyable for people with
respiratory symptoms. The problem of elevated exposure to traffic pollutants is not only relevant
to urban bicyclists but to users of all other active transportation, including pedestrians,
skateboarders, and transit riders. This paradox of urban commuting presents significant challenges
to many cities as they develop and implement plans to increase walking and bicycling. This
dissertation grew out of an inquiry into this paradox of urban commuting, and further expanded to
broaden our understanding of the links between the built environment and health. Specifically, my
inquiry focuses on how the built environment creates interactions between human activity and air
A

3
pollution, and ultimately affects health outcomes. Below, I start with a brief overview of research
on the built environment and health, and elaborate on three important pathways that modify the
relationship between the built environment and health.

The built environment is an important determinant of human health and wellbeing. The
built environment forms the basis of physical and social environments, such as such as housing,
schools, offices, parks, and grocery stores, where individuals and groups perform all functions of
life. However, researchers increasingly understand that the current built environment and
transportation systems can create unhealthy and unsustainable condition. Low-density and
segregated land uses, coupled with poor urban design features, deter walking and bicycling, and
increase reliance on fossil fuel and automobile. Rising rates of childhood obesity over the past
several decades are largely attributable to the combined effects of social and physical environments
that inhibit healthy behaviors.
While genetic factors and sociocultural features cannot be ignored, a growing body of
research suggests that the interactions between physical, social, and biological spheres influence
the course of one’s life and wellbeing (R. G. Evans, Bare, & Marmor, 1994) and shape each
person’s health trajectories since birth and even in utero (Gluckman, Hanson, Cooper, &
Thornburg, 2008). Mounting evidence from multiple disciplines suggests that the built
environment and transportation system play an important role in population health (Dannenberg
et al., 2003; de Nazelle, Rodriguez, & Crawford-Brown, 2009; Jackson, 2003). The lack of daily
physical activity results in higher risk of obesity, diabetes, and cardiovascular diseases (Ladabaum,

4
Mannalithara, Myer, & Singh, 2014). Tailpipe emissions from motorized traffic contribute to
myriad chronic diseases, such as respiratory illnesses, cardiovascular and neurological symptoms
(Health Effects Institute, 2010). Further, poor families living in a racially segregated neighborhood
often lack healthy food options and experience multiple environmental burdens, such as poor
housing condition and exposure to toxic chemicals and air pollutants (G. W. Evans & Kantrowitz,
2002).
Ironically, the modern urban planning practice—responsible for creating the unhealthy
physical and social environment—emerged as a way to respond to public health crisis of the mid-
19th century. Zoning ordinances and subdivision regulations were originally developed to address
chronic urban health issues, such as crowding of tenement housing and sanitation problems
(Corburn, 2004; Jackson, 2003). The Haussman model of Paris was developed to remove
unsanitary conditions and to improve health of urban dwellers through the concept of zoning which
separated urban functions but connected them through circular transportation networks (Corburn,
2004). The purpose of zoning was to create a spatial division between urban dwellers and
unhealthy urban environments, such as coal-fired power plants and industrial complexes. The
Garden City movement heralded by Sir Ebenezer Howard was in the similar vein as the Haussman
model as it applied zoning to create a functional space to improve hygiene and efficiency of urban
living (Frank, Engelke, & Schmid, 2003). However, these planning tools and laws have facilitated
separation of uses and intensified suburban land development, resulting in land use patterns that
encourage motorized trips but discourage physically active trips (Frumkin, Frank, & Jackson,
2004). The contemporary urban planning practices that prioritize functional division of land and
technocratic solutions to environmental externalities fall short of moving us forward in our efforts
toward healthy mobility paradigm.

5
Addressing these problems in the current built environment begins with an accurate
assessment and diagnosis of the root causes of the problem. However, much of the problem lies in
the fact that we have little knowledge about the etiology of various environmental illnesses—how
and why the built environment leads to changes in exposure to environmental stressors and
ultimately influences various health outcomes. As is common with most wicked problems, the
relationship between the built environment and health is complex in nature and operates through
multiple mediators and confounders. Identifying the causal mechanism that links the built
environment and health, therefore, should be tackled with a comprehensive and multidisciplinary
approach. Given this challenge, this dissertation seeks to expand on our understanding of the
relationships between the built environment, travel behavior, and health through three empirical
essays. The goal of this dissertation is not to provide a complete solution to all the problems at
hand, but to offer better insights and evidence of the given problems through integrated conceptual
approach, careful study design, empirical data collection and analysis.
In the following section, I will describe the overall conceptual approach of my dissertation,
and elaborate on each of the pathways between the built environment and health, with particular
focus on travel behavior and air pollution.

Understanding the link between the built environment and health draws on multiple
disciplines and requires cross-disciplinary collaboration. In this study, I focus on how the built
environment creates interactions between travel behavior and air pollution (Figure 1). There are
multiple pathways that link the built environment to certain health outcomes (Frank et al., 2006).
One such pathway is the relationship between the built environment and physical activity. A

6
growing body of research indicates that the current built environments are designed in such a way
that provides little opportunities to engage in regular physical activity (Frumkin et al., 2004; Sallis,
Floyd, Rodríguez, & Saelens, 2012). Planners and public health practitioners now criticize post-
war communities characterized by single family homes, large lot size, and cul-de-sac as being
unhealthy urban design features. Therefore, increasing exposure to healthier built environment
features, such as spending more time in walkable neighborhoods, is deemed important to
maintaining healthy and active lifestyle (Wasfi, Dasgupta, Eluru, & Ross, 2015).

Figure 1. Conceptual approach of this dissertation

Another focus area that I examine relates to how the built environment affects the spatial
distribution of environmental pollution. Environmental pollution, such as toxic chemicals released
to the atmosphere or to the ground, has stationary and non-stationary sources that are associated
with anthropogenic activities. These sources of environmental pollution are often geographically
clustered, creating a spatial patterning of environmental pollution. What is more disturbing is that
this spatial pattern of environment pollution most likely coincides with existing income and racial

7
distribution. For example, one study has shown that low socioeconomic groups are more likely to
be exposed to elevated levels of air pollution due to high traffic densities (Gunier, Hertz, Von
Behren, & Reynolds, 2003) and a concentration of more polluting facilities (Corburn, Osleeb, &
Porter, 2006). Concentration and accumulation of environmental pollution in poor neighborhoods
further aggravate health burdens of such neighborhoods already experiencing significant social
disadvantage (Morello-Frosch, Zuk, Jerrett, Shamasunder, & Kyle, 2011).
Lastly, I examine how travel behavior affects exposure to environmental pollution, or more
specifically, air pollution. Much of the existing literature explore environmental pollution around
residential neighborhoods because time spent at home environment takes up a significant fraction
of one’s daily time use. However, home is not the only microenvironment in which individuals
spend time. Work, school, and commuting environment also take up much of one’s daily pattern,
and the level of exposure from each of the microenvironments may vary by individual lifestyles
and commuting patterns. For example, Dons and her colleagues (2011) found that although time
spent in transport accounts for only 6.2% of daily activities, it was responsible for a quarter of total
exposure to black carbon among 62 subjects in Belgium. Setton and his colleagues (2011) have
also shown that ignoring daily mobility and using residence-only exposure underestimated NO2
by 16% for Metro Vancouver. They also found that for Southern California, bias increased
substantially with increasing time and distance away from home. These studies suggest that
understanding microenvironmental exposure outside of home environment is critical for accurately
assessing individual-level exposure estimates. Given the greater contribution of transport
microenvironment to individual’s daily exposure, failure to include human mobility patterns in
exposure could underestimate health impacts in epidemiological studies and health impact
assessments.

8
With this general conceptual approach in mind, I will explain in detail how the built
environment affects health via three important pathways: 1) the built environment and travel
behavior; 2) travel behavior and air pollution exposure; and 3) the built environment and air
pollution. Figure 2 illustrates the major themes in the literature, and each of these pathways and
the relevant literature are related to the three essays in my dissertation, either directly or indirectly.
As a general point of reference and a building block for the dissertation essays, I provide a broad
overview of the literature related to each of the themes summarized below.

Figure 2. Summary of the major themes in the literature and the linkage to my dissertation essays

One of the most commonly discussed pathways from built environment to health outcomes
concerns travel behavior. The built environment and transportation systems are intricately linked

9
to how one person travels and undertakes physical activity, an important indicator of healthy
lifestyle. Existing theories of the built environment and travel behavior are inherently multi-
disciplinary, originating from a wide range of disciplines in social and behavioral research.

In urban planning and transportation, it is commonly believed that travel behavior is linked
to the built environments, which are typically quantified into measurable terms using the so-called
4D framework – density, diversity, design, and destination accessibility (Cervero & Kockelman,
1997; Handy, Boarnet, Ewing, & Killingsworth, 2002). A number of studies have examined the
assumed link between physical environment and travel behavior through this framework; however,
findings were inconsistent, and little consensus has been reached with regard to the impacts of the
built environment on physical activity (Ewing et al., 2014; Khattak & Rodriguez, 2005; Lund,
2003; Rodríguez, Khattak, & Evenson, 2006).
Some studies have found dense and mixed land use features are associated with increased
walking and shorter vehicle trips (Ewing et al., 2014; Khattak & Rodriguez, 2005; Lund, 2003).
However, other studies have found little difference in the overall trip frequencies between
conventional subdivisions and dense and mixed use neighborhoods, except that the latter was
associated with higher frequency of utilitarian walking (Forsyth, Oakes, Schmitz, & Hearst, 2007;
Rodríguez et al., 2006). Some other researchers using more sophisticated models incorporating
travel costs (i.e. trip lengths and trip distances) have also found mixed results with regard to the
effects of the built environment on travel (Boarnet & Sarmiento, 1998; Crane & Crepeau, 1998;
Greenwald & Boarnet, 2001). These studies suggest that different travel modes may serve different

10
purposes and my operate at different geographic scales (Boarnet & Greenwald, 2000; Handy,
1993).
Other researchers have explored the role of attitudes and preferences and found significant
effects of the attitudinal measures on travel behavior, suggesting that “soft” measures, such as
educational campaigns and programs, can be complementary to conventional land use-based
policies (Joh, Nguyen, & Boarnet, 2011; Van Acker, Van Wee, & Witlox, 2010). Parallel to this
growing interests in the soft measure, researchers have extensively explored the confounding effect
of residential self-selection—the tendency of individuals to sort themselves into certain
neighborhoods based on their travel preferences (Cao, Mokhtarian, & Handy, 2009; Handy, Cao,
& Mokhtarian, 2006; Mokhtarian, 2008). This means that people who are committed to an active
lifestyle would walk or bicycle more regardless of the built environment, and these people will
eventually relocate to more pedestrian and bicycle friendly neighborhoods to match their attitudes
and preferences for an active lifestyle. These studies suggest that, although land use-based polices
are useful for shaping the built environment towards a healthier future in the long term, the built
environment features alone may not be enough to promote active transportation and physical
activity (Badoe & Miller, 2000; Forsyth et al., 2007).

Parallel to this growing interest in attitudes and self-selection in urban planning research,
another useful theoretical development on understanding the link between built environment and
travel behavior emerged from psychology and public health. Researchers have studied a broad
range of factors influencing physical activity, including social and physical environments as well
as individual psychological factors, such as environmental perceptions and cognitive behavioral

11
attributes (Carlson et al., 2012; Handy et al., 2002; Kerr et al., 2010; McNeill, Kreuter, &
Subramanian, 2006; Saelens, Sallis, & Frank, 2003; Timperio et al., 2006). Over the past several
decades, behavioral change models, such as the health belief model (HBM) and the theory of
planned behavior (TPB), have gained popularity in physical activity research and practice (Glanz,
Rimer, & Viswanath, 2008; King, Stokols, Talen, Brassington, & Killingsworth, 2002).
Among the most prominent theory in the context of the built environment and physical
activity is the ecological model which encompasses interpersonal, social, and physical dimensions
of activity promotion (King et al., 2002; Sallis et al., 2006). The ecological model focuses on the
interplay between individual, social, and physical environmental factors to understand individual
behavior, thus aiming to encourage active travel through a multi-faceted strategy (Giles-Corti &
Donovan, 2002). Although this model was developed by psychologists and public health
researchers, it has gained wide support from urban planning and transportation community because
it enables integration of existing land use and transportation policies into promotion of active
travel, contributing towards creating more sustainable and healthier lifestyles (Bauman, Sallis,
Dzewaltowski, & Owen, 2002; Handy et al., 2002; Pickett & Pearl, 2001; Sallis et al., 2006).
Advancement in theoretical understanding of nonmotorized travel behavior has been fueled by this
interdisciplinary research effort, often referred to as active living research which focuses on
synergies between the built environment and nonomotorized travel (Bussel, Leviton, & Orleans,
2009).

Transportation researchers have long been interested in modeling travel behavior, but
active transportation received relatively little attention until recently. Existing travel behavior

12
studies and modeling techniques have been inadequate for describing active travel behavior. For
example, a widely popular four-step model is useful for predicting travel demand at aggregate
levels, but it provides little information with regard to individual behavioral patterns and physical
activity behaviors. Further, much of the discussion on travel behavior modeling have been focused
on some technical details, such as temporal ordering of traveler’s decision on mode choice and
destination (Newman & Bernardin, 2010) or the model efficiency between sequential procedure
and combined model formulation of mode choice and destination (Boyce, 2002; Siegel, Cea,
Fernández, Rodriguez, & Boyce, 2006).
In response to this challenge in understanding active travel behavior, activity-based
approaches, based on discrete choice models, have been proposed. Discrete choice models assume
that individuals make rational travel decisions based on their preferences for particular trips or
travel modes, and the relative cost of making that trip or choosing that travel mode (Domencich &
McFadden, 1975; McFadden, 1974). Ben-Akiva and Lerman (1985) further extended this
framework to develop activity-based approach, with the main goal of predicting individual mode
choice and timing decision of travel (Ben-Akiva & Bierlaire, 1999). Since the early theoretical
development, the activity-based approach has shown some potential for application in physical
activity research. For example, Sener and Bhat (2011) have applied the multiple discrete
continuous extreme value model to examine temporal and spatial context of physical activity
participation. Using the empirical data from the 2000 Bay Area Travel Survey, the authors found
that physical activity participation may be more related to lifestyle choices, rather than quality of
the surrounding built environment. Their findings suggest that different lifestyle choices and life
stages could affect the location and timing of physical activity participation.

13
Unlike the traditional four-step models, activity-based models have the capability to
incorporate active travel and micro-scale behavior patterns. However, few operational activity-
based models to date include a full range of built environment factors into modeling travel behavior
because of substantial data and computational requirements associated with model development
(Liu et al. 2012). Some of the recent activity-based research considers built environment factors,
but most of them use simple measures, such as employment and population density, and distance
to central business district (Bowman & Ben-Akiva, 2000; Pinjari, Pendyala, Bhat, & Waddell,
2011). The activity-based approach is a more promising solution than the conventional four-step
models, but much remains to be done in order to incorporate a full range of built environment
factors and to draw a causal relationship with physical activity.

The relationship between built environment and transportation is a growing area of
research that transcends many disciplines, including urban planning, transportation, and public
health. A brief review of the literature indicates that there are consistent results regarding the role
of attitudes, perceptions, and self-selections for promoting active travel behavior. Further, different
sets of built environment features seem to have different impacts on active travel depending on the
context and the purposes of travel (i.e. utilitarian vs recreational). However, much remains
unknown regarding how much the built environments affect physical activity and what built
environment factors contribute most to increasing physical activity. To advance our knowledge on
the link between built environment and travel behavior, further research will need to employ more
rigorous research design, such as longitudinal data and experimental research design, in order to

14
allow stronger causal inference than is possible in the much more common study designs using
cross-sectional data.

In explaining the travel behavior-air pollution exposure relationship, it is critical to
understand a conceptual framework for linking air pollution sources to health effects. Figure 3
shows this framework illustrating how sources of air pollution is linked to health outcomes through
exposure of human populations and the doses of air pollutants that are inhaled. Moving from left
to right, air pollutants are released into the environment from a source. Many pollutants can be
transformed through a number of processes, including chemical reactions and biological
degradation. Pollutants or their transformation products move through the environment and can be
found in the air. The intensity of exposure depends upon the pollutant concentration in the media,
as well as the duration of contact with humans. Exposure becomes “dose” when the pollutant
moves across the human body barrier, which finally leads to health effects.

Figure 3. A framework for linking air pollution sources to health effects (Addapted from NRC, 1998).

15
The NRC’s framework implies that the exposure pathway from source to human body is
critical to understanding human health impacts. For example, living close to major freeways may
increase the risk of respiratory diseases, but what actually triggers the health effects are the length
and magnitude of exposure to traffic pollutants and the doses of harmful toxins and particulate
matters that enter into the human body. Therefore, linking exposure to health effects has been the
central topic in public health and epidemiological studies.

To understand the link between travel behavior and air pollution exposure, researchers have
used personal monitoring technique to quantify the effect of travel mode choice on air pollution
exposure (Briggs, de Hoogh, Morris, & Gulliver, 2008; Chan, Lau, Lee, & Chan, 2002; Kam,
Cheung, Daher, & Sioutas, 2011; Kaur, Nieuwenhuijsen, & Colvile, 2007; Zuurbier et al., 2010).
Personal monitoring has been increasingly useful to capture exposure levels of mobile subjects
such as pedestrians and bicyclists (Greaves, 2006; J. Gulliver & Briggs, 2004; Hertel, Hvidberg,
Ketzel, Storm, & Stausgaard, 2008; Kaur et al., 2006). Personal monitoring techniques allow
researchers to measure location-specific variations of personal exposure while the subject is in
motion. However, due to its high cost associated with intensive data gathering, the study is
typically done in a short period using a limited number of devices. If sampling coverage and
duration can be expanded, researchers can get a better understanding of exposure levels in
microscale urban environments and can identify specific environmental factors affecting high
exposure levels when combined with personal diary or audio-visual recording device (Kaur,
Nieuwenhuijsen, & Colvile, 2005b; Thai, McKendry, & Brauer, 2008).

16
Previous studies of mode choice and air pollution suggest that personal exposure is
generally higher when using more active travel such as walking and bicycling. Adams and his
colleagues (2001) measured highest concentrations of particulate matter ≤ 2.5 µm in aerodynamic
diameter (PM2.5) in underground trains, followed by buses, cars, and walking. Gulliver and Briggs
(2004) found higher concentration of coarse particles when driving in cars than walking, but no
significant difference was observed for fine particle concentrations. The authors also found that
in-car and walking exposure was similar when routes were comparable, indicating the effect of
traffic conditions on commuter exposure. Zuurbier and her colleagues (2010) found higher
exposure to PM10 and PM2.5 when taking diesel bus and driving a car than riding a bike. However,
ultrafine particle concentrations elevated the levels of exposure for bicyclists on high-traffic routes
compared to low-traffic routes. They also found that modes, fuel types, and travel routes contribute
to the differences in exposure among various commuting modes.

While there is a difference in exposure among various modes, more recent studies have
argued that it is not appropriate to rank modes in order of exposure without detailed consideration
of various factors (Knibbs, Cole-Hunter, & Morawska, 2011). One of the key factors actively
discussed among scholars is the influence of ventilation. For all types of enclosed vehicle,
ventilation determines how much outside toxic air flows into the cabin, and that influences the
particle concentration levels inside cars, trains, buses, and etc. Knibbs and de Dear (2010) found
that exposure to fine particles to be higher when taking a train than driving a car. They noted that
the lowest PM levels in car may reflect the influence of ventilation, and newer buses and trains
may provide protection from particle exposure. Quiros and his colleagues (2013) found

17
significantly lower exposure to PM when driving a car with windows closed and air conditioning
on among all other modes studied (walking, biking, and driving a car with open windows). de
Nazelle and hercolleagues (2012) found consistently higher PM and BC exposure when driving a
car with open-windows than walking, bicycling, or taking transit. They reasoned that different
ventilation settings may have affected their results with consistently higher in-car exposure. These
recent studies suggest that pollutant exposure levels among different commuting modes may be
more complicated by various mediators, among which the key factors include travel routes and
ventilation conditions of different travel modes.

Much of the previous research on the link between transportation and air pollution focus
on emission reduction potential of active transportation, with the assumption that traffic emissions
will be greatly reduced if we make a societal effort to shift to active transportation (Frank et al.,
2006). However, a growing body of research has shown that decline in traffic emissions does not
always lead to reduction in exposure to air pollutants due to variations in air pollution gradients
within urban areas (Marshall, Brauer, & Frank, 2009; Marshall, McKone, Deakin, & Nazaroff,
2005; Schweitzer & Zhou, 2010). Furthermore, the health benefits of active transportation should
be carefully analyzed by making a balanced calculation between benefits accrued from increased
physical activity and risks associated with higher exposure to pollutants (de Nazelle et al., 2011).
Because of the benefits and risks associated with active transportation, there is an intense
debate with regard to future directions of active transportation policies. Some maintained that
although shifting to active transportation can help reduce traffic emissions in general, it may
actually increase exposure to traffic pollutants for those that are making the shift (Adams et al.,

18
2001; Briggs et al., 2008; Kaur, Nieuwenhuijsen, & Colvile, 2005a). In particular, Bicycling in
dense urban environments could disproportionately affect personal exposure to traffic-related air
pollution by placing emission sources closer to travelers and increasing exposure to local ambient
concentrations (Hertel et al., 2008). Further, higher intake fraction and increased dose of air
pollutants among bicyclists can result in increased inhalation of harmful pollutants, potentially
offsetting health benefits of physical activity (Bigazzi & Figliozzi, 2014). For example, Int Panis
and his colleagues (2010) measured exposure and ventilation rates of 55 persons in Belgium and
found that inhaled doses of air pollutants while cycling were four to nine times higher than for car
drivers on the same route. These studies indicate that significantly higher exposure to pollutants
would occur for active transportation users than for car drivers, and the actual amount of pollutants
deposited into the lung could be greater due to higher ventilation rates of active travel users.
Others have maintained that the benefits of active transportation far outweigh the risks
associated with increased exposure to pollutants (de Hartog, Boogaard, Nijland, & Hoek, 2010;
Rabl & de Nazelle, 2012; Rojas-Rueda, de Nazelle, Tainio, & Nieuwenhuijsen, 2011). These
studies typically assumed substantial changes at the population level through counterfactual
scenarios. For example, Rojas-Rueda and his colleagues (2012) made a wide assumption that when
40% of car trips are shifted to cycling, 76 number of deaths would be avoided per year in
Barcelona, due in large part to health benefits accrued from increased physical activity. In addition,
most of the health benefit-cost research have been conducted in developed countries with relatively
good air quality. However, recent studies included low-income countries and found that the break-
event point where the risk from air pollution exposure starts to outweigh the benefits of exercise
is reached only at substantially high levels of pollution or with extended hours of walking (more
than 10 hours) and cycling (more than 1 hours). This result reinforces the findings from earlier

19
studies that the benefits of active transportation generally outweigh the risks associated with air
pollution exposure.

Despite much progress made over the last few years, the current state of knowledge with
regards to the relationship between travel behavior and air pollution exposure is still too early to
make strong policy recommendations. In general, modal difference studies suggest that active
transportation users experience higher levels of exposure than car drivers. However, there are a
number of confounders, e.g. ventilation condition, that modify the effects of mode choice on air
pollution exposure. When it comes to physical activity and air pollution exposure, it is still unclear
whether short-term joint physical activity and air pollution exposure have positive or negative
health outcomes and what the underlying biological mechanisms are. Also, we know little about
whether joint exposure and physical activity make at-risk populations more or less susceptible to
health effects. From a policy perspective, it would be of particular interests to mitigate inequalities
in both air pollution exposure and physical activity opportunities (Marshall et al., 2009). Therefore,
identifying key target areas and population subgroups to address both air pollution exposure and
physical activity will be an important task for future policy and research.

Another important pathway that links built environment and health outcomes is the spatial
distribution of air pollution in the built environment. To quantify the spatial pattern of air pollution,
researchers have used mathematical models (deterministic, stochastic), empirical models

20
(statistical, probabilistic), or a combination of both. The mathematical models are useful for
predicting long-term trends whereas the empirical models perform well for estimating short-term
trends. Another approach is to use natural experiment to empirically assess the impact of certain
interventions on air pollution exposure and health effects.

Researchers have used various methods to establish the link between the built environment
and air pollution. One of the most popular approaches is to use a mathematical model based on
chemical and physical characteristics of air pollutants in the atmosphere. One such model is a
dispersion model which uses Gaussian function to understand how atmospheric processes affect
dispersion of pollutants emitted from either stationary or non-stationary sources. Various models,
such as CALINE, ADMS, and OSPM, have been developed to provide better assessment of
exposure to traffic-related air pollutants at the microscale (Berkowicz, Ketzel, Jensen, Hvidberg,
& Raaschounielsen, 2008; John Gulliver & Briggs, 2011; Vardoulakis, 2003). These dispersion
models, however, have important drawbacks. They require extensive computing power, especially
when the models need to be applied to large areas and at high spatial or temporal resolution. They
also rely on a quality and resolution of input data, including information on source distribution,
activity and emission rates, meteorological conditions and surface terrain, not available in many
local jurisdictions.
Previous studies have shown that air pollution can vary greatly within a city, and the
differences in traffic levels across neighborhoods are largely attributable to intra-urban variations
(Briggs, 2000; M Jerrett et al., 2007). A growing body of research has found that residents living
close to major roadways experience disproportionately higher risk of cardiovascular and

21
respiratory diseases than those living farther away from traffic (Garshick, Laden, Hart, & Caron,
2003; Gauderman et al., 2007; Hoek, Brunekreef, Goldbohm, Fischer, & van den Brandt, 2002;
Schwartz et al., 2005). There is a growing consensus among air quality experts that existing
methods of relying on central air quality monitors may misclassify levels of personal exposure and
health effects to a larger extent than previously thought (Kanaroglou et al., 2005; Wilson,
Kingham, Pearce, & Sturman, 2005). Therefore, air quality modeling based on the central monitors
may not be adequate to explain the extent to which population exposure occurs at a fine-grained
urban environment.

Empirical studies of air pollution exposure constitute another approach in assessing the
impact of the built environment on air pollution. Land use regression is probably the most widely
used technique that incorporates traffic indicators (population density and traffic intensity) and
land use variables as proxy for pollutant concentrations. Several studies have used this model in
long-term exposure assessment of particulate matter and other pollutants (Brauer et al., 2003; Hoek
et al., 2002). This method is suitable for exposure assessment of large population in urban areas
where traffic and land use designation is the main contributor to ambient concentrations. It is
suitable for addressing spatial variations of pollutant concentrations; however, its predictability for
micro-scale exposure assessment is significantly reduced if local traffic or land use data are not
available.
Modeling studies generally suggest that dense urban development patterns can lead to
reducing emissions (Frank, Jr, & Bachman, 2000; Stone, Mednick, Holloway, & Spak, 2007).
However, empirical studies using regression-based approach have found that dense development

22
strategies may actually worsen localized emissions and subpopulation exposure to traffic
emissions (Marshall et al., 2009; Schweitzer & Zhou, 2010). The paradox of urban density is that
there are substantial variations in spatial distribution of air pollution within urban areas (Brauer et
al., 2003; Briggs, 2000), and that these variations generally follow socioeconomic gradients
(Michael Jerrett et al., 2003), further perpetuating the health disparities between the poor and the
affluent communities.

As part of the evaluation of environmental regulatory policies, government entities and
related agencies have conducted assessment of the effectiveness of regulations—termed
accountability research—for improving air quality. Most popular method of the accountability
research is natural experiments to understand how specific interventions in the built environment
influence pollutant formation and distribution. For example, Hedley and his colleagues (2002)
assessed the effects of restriction on sulphur fuel in Hong Kong, and Clancy and his colleagues
(2002) investigated the effects of coal ban in Dublin, Ireland. Both of these studies showed that
targeted interventions were very effective in controlling local air quality, ranging from 50 to 70%
reduction in single pollutant concentrations. Recently, a few attempts have been made to evaluate
the effects of short-term actions designed to reduce a broad range of local air pollutants. Li and his
colleagues (2011) examined the impact of the air pollution control measures on air quality during
the 2008 Beijing Olympic Games, and
In United States, Friedman and his colleagues (2001) have shown that reduced traffic
during the 1996 Atlanta Olympic Games led to a significant reduction in daily peak ozone
concentrations by 28%. Another study investigated the impact of two-day freeway closure in Los

23
Angeles and found that the closure led to 32% reduction in PM2.5 and 16% reduction in ozone
region wide (Hong, Schweitzer, Yang, & Marr, 2015). Whitlow and his colleagues (2011)
examined the impact of “Summer Streets” campaign which eliminated a small section of local
traffic for three consecutive Saturdays. Their results were less impressive than the results of the
long-term intervention, partly because of a confined geographic location and a short term
intervention. While massive reduction in traffic from freeway have shown promising results in
managing traffic pollutants, results of small-scale interventions have been mixed. Summer streets
event in New York City, a small scale traffic exclusion program, resulted in little change in
particulate mass concentration, but substantial reductions in ultrafine particle number
concentrations (Whitlow, Hall, Zhang, & Anguita, 2011). These studies offer some evidence that
the large-scale interventions like city-wide traffic ban may be effective for regional air quality
management, but effectiveness of small scale intervention warrants further investigation.

The relationship between the built environment and air pollution is the subject of ongoing
research that combines the field of geography and environmental science. The land use regression
techniques enhance our understanding of the within-urban variability of air pollution exposure.
However, the current research has been limited to understanding static physical environment,
which raises a question about spatio-temporal aspects of human exposure to air pollution. To
address this important gap, the concept of space-time prism had received growing interests in the
research community, giving rise to the new concept of “activity space” that incorporates all
relevant dynamic geographic contexts in time and space (Chaix et al., 2012; Cummins, Curtis,
Diez-Roux, & Macintyre, 2007; Kwan, 2012, 2013; Steinle, Reis, & Sabel, 2013; Zenk, Schulz, &

24
Matthews, 2011). However, much of the literature on the health effects of the built environment
remains largely on exploring static environmental determinants of health (Steinle et al., 2013).
Previous studies generally do not consider appropriate “exposure” measure in both space and time.
The appropriate measure of exposure is important because it provides causal mechanism in
estimating the extent to which human exposure to air pollution occurs. Therefore, human behaviors
are critical to understanding the temporal and spatial patterns of environmental exposure.

Thus far, I have examined the current state of knowledge and research needs with regard
to the relationships between the built environment, travel behavior, and air pollution. Given these
gaps in the existing literature, especially with regards to urban planning research, I make three
major contributions through this dissertation. First, I use experimental research design and
longitudinal data, which allowed me to make stronger causal inference than is possible in the much
more common study design that relies on cross-sectional data. Although there have been much
discussion about moving beyond the cross-sectional research paradigm, experimental studies are
still rare in urban planning research. Because of the cross-sectional nature of most studies, it is
unclear whether changes in the built environment can have a meaningful impact on active travel
and physical activity (Bauman et al., 2002). My first essay fills this gap in the literature by
presenting results from one of the first longitudinal studies of physical activity change before and
after the construction of a new light rail line. Unlike previous cross-sectional studies which showed
positive impacts of transit investments, the results of my first essay using longitudinal data showed
that new transit interventions may have differential impacts (both positive and negative) depending
on individuals’ strength of past behavior.

25
Second, my dissertation seeks to bring light to the urban planning research by synthesizing
most up-to-date theory in environmental science and public health. Despite much effort in pushing
the boundary, the planning discipline is still in isolation from other disciplines, namely,
environmental science and public health. Moreover, previous studies on the pollutant exposure of
urban commuters have primarily focused on specific modes or general differences among various
commuting modes (Hudda, Kostenidou, Sioutas, Delfino, & Fruin, 2011; Knibbs et al., 2011). To
my knowledge, no studies have conducted exposure levels while commuting on light rail transit in
direct comparison to traveling by car. Only few studies have examined air pollution levels in
ground-level transit systems. Given this gap in the literature, my second essay expands on the
conventional urban planning research by integrating mobile monitoring techniques and stochastic
simulations based on previous study results.
Third, I expand and challenge the conventional urban planning research by focusing on
the impact of traffic on localized emissions exposure, rather than total emissions levels. While
Greenhouse gas (GHG) emission is an important research topic, little attention has been given to
the health effects of human exposure to air pollutants in urban planning research. In my
dissertation, I focus on the impact of the built environment on air pollution exposure as it has direct
implications for human health. Previous studies have shown that urban and traffic density is
positively related to air pollution exposure (Borrego et al., 2006; de Nazelle & Rodriguez, 2009;
Marshall et al., 2009; Sallis et al., 2013). However, it is still unclear how the built environment
can be used as a policy lever to manage chronic problems related to air pollution exposure for local
municipalities. To address this gap, my third essay uses the car-free street events, called CicLAvia,
as natural experiment to understand how traffic restriction at local levels influences exposure of
urban population to air pollution.

26
Combined together, this dissertation seeks to synthesize the current state of knowledge in
urban planning, transportation, environmental science, and public health, and to bring more
experimental research tradition through rigorous research design and original data collection.

This dissertation is organized in five chapters. Chapter 1 (this chapter) is the introduction
which contains an overview of the dissertation and a conceptual approach, followed by a review
of relevant literature. Chapter 2 is the first essay, entitled “New light rail transit and active travel:
A longitudinal study.” In this essay, I employ a quasi-experimental design and longitudinal data
to investigate the before-and-after impact of new light rail transit investment on active travel
behavior and physical activity. Chapter 3 is the second essay, entitled “Commuter’s Paradox: Does
switching from car to light rail transit help us breathe cleaner air?” This essay investigates whether
traffic exposure differs between car and light rail train by using real-time instruments to monitor
five traffic pollutants while driving a car and taking light rail transit simultaneously. Chapter 4
contains the third essay, entitled “Car-free streets and local air quality: A case of CicLAvia in Los
Angeles,” where I use real-time pollutant samples to investigate the effects of a temporary traffic
ban on local air quality. Chapter 5 concludes the dissertation by summarizing the results of the
three empirical essays and provides broader implications for future research and policy.

27

28

e use panel data to investigate the before-and-after impact of a new light rail
transit line on active travel behavior. Participants were divided into a treatment
group and a control group (residing < ½ mile and > ½ mile from a new light
rail transit station, respectively). Self-reported walking (n=204) and accelerometer-measured
physical activity (n=73) were obtained for both groups before and after the new light rail transit
opened. This is the first application of an experimental-control group study design around light
rail in California, and one of the first in the U.S. Our panel design provides an opportunity for
stronger causal inference than is possible in the much more common study designs that use cross-
sectional data. It also provides an opportunity to examine how an individual’s previous activity
W

29

behavior influences the role that new light rail transit access plays in promoting active travel
behavior. The results show that, when not controlling for subject’s before-opening walking or
physical activity, there was no significant relationship between treatment group status and after-
opening walking or physical activity. However, when controlling for an interaction between
baseline walking/physical activity and treatment group membership, we found that living within a
half-mile of a transit station was associated with an increase in walking and physical activity for
participants who previously had low walking and physical activity levels. The results were
opposite for participants with previously high walking and physical activity levels. Future policy
and research should consider the possibility that sedentary populations may be more responsive to
new transit investments, and more targeted “soft” approaches in transit service would be needed
to encourage people to make healthy travel choices.

A sedentary lifestyle is a growing concern in the United States. It is a major risk factor for
obesity and a variety of chronic diseases, such as coronary heart disease, type 2 diabetes, and breast
and colon cancers (Lee, Shiroma, Lobelo, & Puska, 2012; WHO, 2009). However, the share of
adults and children engaged in physically active travel for both work and leisure has declined
sharply over the past decades in the United States (Bassett, Pucher, Buehler, Thompson, & Crouter,
2008; Brownson, Boehmer, & Luke, 2005). Public health and urban planning researchers have
turned to the potential role of the built environment to change behavior and create a pathway to a
physically active lifestyle (Brownson, Hoehner, Day, Forsyth, & Sallis, 2009; Frank et al., 2006).
This new line of research focusing on the potential role of the built environment in promoting
30

active travel behavior has had a profound influence on the current research and practice in urban
planning and transportation (Badland & Schofield, 2005; Bors et al., 2009; Sallis, Frank, Saelens,
& Kraft, 2004).
Public transit may be one means to promote more active and healthy travel choices
(Morency, Trépanier, & Demers, 2011; Stokes, MacDonald, & Ridgeway, 2008). In recent years,
many local and regional governments have built new transit systems, and an increasingly common
secondary justification for those systems is the promotion of active lifestyles (US DOT, 2014;
Zheng, 2008). Previous research has found a positive association between frequent transit use and
moderate physical activity (Besser & Dannenberg, 2005; Lachapelle, Frank, Saelens, Sallis, &
Conway, 2011; Rissel, Curac, Greenaway, & Bauman, 2012). Living close to a transit station was
also found to increase the odds of utilitarian walking (McCormack, Giles-Corti, & Bulsara, 2008).
However, these studies were based on cross-sectional data, showing only a correlation with the
observed relationships. Because of the cross-sectional nature of most studies, it is still unclear
whether changes in the built environment through new transit investments can lead to a meaningful
behavioral change (Bauman et al., 2002).
We fill a gap in the literature by presenting results from one of the first longitudinal studies
of travel behavior change before and after the construction of a new light rail line. Our longitudinal
study design allows stronger causal inference than cross-sectional data. Our research examines the
role of an individual’s previous walking and physical activity levels in influencing the “treatment
effect” of new light rail transit on after-opening walking and physical activity. The use of past
behavior in a longitudinal study of the impact of new light rail is novel in the literature. In addition,
we use more robust measurement of active travel behavior– self-reported walking and
accelerometer-based physical activity, and our findings are similar for both measures.

31

Previous research on the impacts of transit investments suggest that transit is positively
associated with active travel behavior (Besser & Dannenberg, 2005; Lachapelle & Noland, 2012;
Rissel et al., 2012). However, most research in this area consists of cross sectional studies, making
it difficult to assess causal relationships. Longitudinal studies can provide stronger evidence on
the impacts of new transit investments and overcome concerns about the influence of residential
selection on travel behavior (Cao, Handy, & Mokhtarian, 2006). However, longitudinal
evaluations of the travel impacts associated with new light rail transit are still rare. To our
knowledge, only two studies have longitudinally examined the effects of a new light rail transit
line on active travel behavior. Brown and Werner (2007, 2008) used a pre-post study design to
examine the impact of a new light rail line on 51 residents in Salt Lake City, Utah. They found
that using the new transit service was associated with an increase in moderate physical activity,
but no statistically significant association was found between proximity to the transit stations and
physical activity. Using longitudinal samples from Charlotte, North Carolina, MacDonald and his
colleagues (2010) found a strong association between light rail use and body mass index (BMI)
and obesity. However, they found only a marginally significant association between light rail use
and the odds of meeting recommended physical activity. The results from the previous studies
suggest that light rail transit may help overcome some of the barriers to engage in active travel,
but it is still unclear whether exposure to new transit service has any meaningful impacts on
residents’ active travel behavior. We contribute to the literature by extending the scope of
longitudinal, pre-post studies of new light rail with our case in Los Angeles.
32

Drawing from a more general literature on the relationship between the built environment
and physical activity, researchers have studied a broad range of factors influencing physical
activity, including social and physical environments as well as individual psychological factors,
such as environmental perceptions and cognitive behavioral attributes (Carlson et al., 2012; Handy
et al., 2002; Kerr et al., 2010; McNeill et al., 2006; Saelens et al., 2003; Timperio et al., 2006).
Over the past several decades, behavioral change models, such as the health belief model (HBM)
and the theory of planned behavior (TPB), have gained popularity in physical activity research and
practice (Glanz et al., 2008; King et al., 2002). Among the most prominent theory in the context
of the built environment and physical activity is the ecological model which encompasses
interpersonal, social, and physical dimensions of activity promotion (Sallis et al., 2006). This
model has been widely used by urban planners and policy makers because it enables integration
of existing land use and transportation policies into physical activity promotion, contributing
towards creating more sustainable and healthier lifestyles (Bauman et al., 2002; Pickett & Pearl,
2001; Sallis et al., 2006).

Despite much work on theory development, there has been a lack of clarity and consensus
in our understanding of potential mechanisms of physical activity change (King et al., 2002). One
particular area that has received relatively little attention in the active transportation field is the
role of past behavior, although it has been a subject of rigorous research in other arenas. Past
behavior has been actively discussed among researchers studying the theory of reasoned action

33

(TRA), theory of planned behavior (TPB), and habitual travel behavior (I. Ajzen & Fishbein, 1980;
Icek Ajzen, 1991; Gärling & Axhausen, 2003). Previous studies have consistently found an
independent influence of the frequency of past behavior across a range of behaviors, such as drug
use, school attendance, television watching, and recycling behavior (Bentler & Speckart, 1979;
Fredricks & Dossett, 1983; Ouellette & Wood, 1998). In particular, past physical activity behavior
has been found to influence habit formation, and thus influence intentions to engage in later
physical activity (Aarts, Paulussen, & Schaalma, 1997; Hagger, 2001). Although there is an on-
going debate whether past behavior can directly predict later behavior (Icek Ajzen, 2002), studies
have consistently found that past behavior has a significant residual effect beyond cognitive
behavioral constructs (e.g. intentions and perceived behavioral control) while improving model
performance and predictability (Bamberg, Ajzen, & Schmidt, 2003; Hagger, 2001; Norman,
Conner, & Bell, 2000; Bas Verplanken & Melkevik, 2008; Yordy & Lent, 1993). However,
transportation and urban planning research has rarely examined the role of past behavior in
explaining the relationship between the built environment and active travel behavior (Gardner,
2009; Thøgersen, 2006).
In this study, we hypothesize that 1) transit proximity, or exposure to new transit service,
is positively associated with active travel behavior (self-reported walking and accelerometer-based
physical activity); and that 2) the relationship between transit proximity and active travel behavior
will be affected by past behavior, specifically baseline walking trips and physical activity
measured before the new light rail transit opened. This study provides stronger evidence on
causality by using a natural experiment and longitudinal observations to evaluate individual travel
behavior outcomes before and after construction of a new light rail transit system. Compared to
the two previous longitudinal studies which relied on either accelerometry data (Brown & Werner,
34

2007) or survey-derived physical activity measures (MacDonald et al., 2010), we employed both
survey data and accelerometry data. Regarding the second hypothesis, this study takes a cue from
the theory of planned behavior by focusing on the role of baseline walking and physical activity
in influencing the effect of transit proximity on post-opening changes in active travel behavior. To
our knowledge, no studies have examined the potential effect of past behavior in influencing the
impact of new light rail transit on active travel behavior. This study makes an important
contribution to what is known about the relationship between public transit and active travel
behavior by leveraging a longitudinal research design, using both subjective and objective
measures of active travel, and examining the role of baseline walking and physical activity in
influencing the effect of new light rail transit access.

The study neighborhood is located along the Expo light rail line (Expo Line) in south Los
Angeles. This new light rail line extends south and west from downtown Los Angeles, eventually
reaching downtown Santa Monica upon completion of Phase II (expected Phase II completion by
2016). The present study only includes the Phase I construction of the Expo Line connecting
downtown Los Angeles and Culver City, which was opened in stages in April and June 2012.
Figure 1 shows examples of the neighborhoods around the light rail stations before and after the
construction of the Expo Line, obtained from time-lapsed images provided by Google Street View.
Figure 1a shows the changes in the neighborhood around the Western Station (at-grade platform),
and Figure 1b shows the neighborhood around the La Cienega Station (elevated platform). Note

35

that the neighborhoods after the construction of the light rail transit had better landscaping
elements and nonmotorized facilities, such as dedicated bicycle lanes and improved sidewalks.
These added urban design elements around the stations seemed to provide more pleasant and
welcoming environments to pedestrians and bicyclists.

Figure 1. Neighborhoods around Western Station (A); and La Cienega Station (B) before and after the Expo
Line, Los Angeles, CA
Source: Courtesy of Google Maps. The before-opening and the after-opening photos are from archived google street
views dated May 2011 and December 2012, respectively.

Figure 2 illustrates the study area including the approximate home location of participants
in the treatment group (circled areas) and participants in the control group (the area extending in
the broader outline beyond those circles and toward the west and south). The treatment group is
36

defined as residents living within ½ mile of a new Expo Line station; and the control group as
residents living farther away from a station. A half-mile boundary is considered a typical
catchment area within which most transit riders are willing to walk (Ewing, 1999). Therefore, it
was used to delineate between households exposed to new rail service and those less exposed to
the service.

Figure 2. The Expo Line study area: Los Angeles, California, 2011–2013
Note. The participant locations are based on the residential address of the 279 baseline survey sample.

The present analyses use two waves of longitudinal samples from the Expo Line study.
Full details of the recruitment procedures and survey instruments are described elsewhere

37

(Houston, 2014; Spears, Houston, & Boarnet, 2013). In brief, invitation letters were sent to all
households (n = 27,275) in the study area based on addresses purchased from InfoUSA, a
commercial database provider. All households with an interest in participating were selected into
the study. The survey was conducted five to seven months before (Baseline, September 2011 –
January 2012) and after (Follow-up, September 2012 – January 2013) the opening of the Expo
Line. A total of 279 households participated in a demographic survey and a 7-day travel survey
(survey sample), and each of those households was recruited to participate in the same 7-day
tracking after opening of the Expo Line, and 204 households participated in after-opening data
collection. The final size of the survey sample was 204 subjects who participated in both pre- and
post-opening surveys. In approximately half of the before-opening households, 143 primary
respondents (one per household) were recruited for participating in a more detailed survey
involving accelerometer and GPS devices (mobile sample). The mobile sample participants were
instructed to wear the accelerometer device on the right hip on a nylon belt around the waist during
waking hours for seven days. The participants were also instructed to take off the accelerometer
device when engaging in rigorous athletic activities, such as swimming practice or a soccer match.
A project staff member met with each participant to provide training on the use of the equipment,
to ensure compliance with study procedures, and to provide a gift card ($30 for the baseline; $75
for the follow-up) as an incentive for completing the survey. After going through a set of filtering
processes which are detailed in the next section, the total size of the mobile sample came down to
73 subjects who participated in both pre- and post-opening surveys.
Our response rate for the survey sample was 1%. This response rate is comparable to the
response rate of 1.4% in the 2010-2012 California Household Travel Survey (Los Angeles and
Ventura Counties) and that of 0.4% in a county-wide survey that we conducted for an unrelated
38

study of travel behavior (Houston, Boarnet, & Spears, 2015, p. 10). A comparison of the
households in the final sample to the households invited to participate in the study indicated that
response rates did not vary greatly by household and demographic characteristics (Houston et al.,
2015, p. 11). Study methods were approved by the Institutional Review Board at the University of
California, Irvine.

Socio-demographic data. Participants completed baseline and follow-up questionnaires
regarding demographic characteristics, including household income and employment status. For
the regression analysis, the household income variable was converted into a dichotomous measure,
1 denoting household income below $35,000 and 0 otherwise. The education attainment measure
was dropped in the regression analysis because of multicollinearity with the household income
measure. The regression results remained the same when the household income variable was
replaced with the education variable.
Travel survey data. Participants completed a 7-day trip and vehicle odometer log. All
household members over 12 years of age were instructed to carry the travel log, and record their
trip counts for each of the following modes: private vehicle as a driver, private vehicle as a
passenger, motor-cycle/scooter, bus, train, bicycle and walking. This travel survey was used to
identify participants’ transit usage and the frequency of trips by bus and train. A dummy variable
was created to indicate whether participants increased the number of bus/train trips from pre- to
post-opening surveys. The survey was also used to identify participants’ frequency of walking and
bicycling. Due to low frequency of bicycling trips, we used the total number of walking trips
reported by the study participants as a subjective measure of active travel. It should be noted that

39

a separate investigation was performed to validate the self-reported walking trips against more
objective trip estimates derived from the participant’s GPS traces. The results indicate that the self-
reported data were in strong agreement with the GPS-derived measure, with the mean differences
between -0.36 and -0.39 depending on the GPS classification scheme (Houston, Luong, & Boarnet,
2014).
Physical activity data. Participants’ physical activity was measured using GT1M
accelerometers (ActiGraph, Fort Walton Beach FL, 2005). The accelerometer provides objective
estimates of moderate-intensity physical activities such as light or brisk walking (Sirard,
Melanson, Li, & Freedson, 2000). The accelerometer collected minute-by-minute activity counts
that were translated into minutes spent on different physical activity levels of light, moderate, hard,
and very hard activity based on cutpoints derived from previous research (Freedson, Melanson, &
Sirard, 1998).
A common data reduction procedure was used to clean the accelerometer data (Matthews,
Hagströmer, Pober, & Bowles, 2012; Troiano et al., 2008). First, accelerometer “nonwear” time
was defined as hours when the device was not worn, and was determined as having more than 60
minutes of consecutive zeroes for vertical acceleration, with allowance of up to 2 minutes of
nonzero readings within the 60 minutes. Accelerometer “wear” time was defined as hours when
participants were awake and wore the device. The “wear” time was determined by subtracting the
“nonwear” time from 24 h. A valid day was then determined as having 10 or more hours of
accelerometer “wear” time. Participants with at least three valid days were selected into the final
sample, and average moderate-and-vigorous physical activity (MVPA) minutes were calculated
by combining minutes spent on moderate, hard, and very hard activities. We used the
accelerometer-based MVPA minutes as an objective measure of active travel.
40

Demographics and active travel outcomes (self-reported walking and accelerometer-based
physical activity) were compared between treatment and control groups using the t-test, and fisher
test was used for categorical data. The longitudinal analysis was performed using an ANCOVA
(analysis of covariance) framework, wherein the dependent variable was regressed on past
behavior, the treatment condition (proximity to transit), and an interaction between the past
behavior and the treatment condition. The form of the regression is shown in Equation (1).

𝑌 2,𝑖 = 𝛽 0
+ 𝛽 1
𝑌 1,𝑖 + 𝛽 2
𝑇 𝑖 + 𝛽 3
(𝑌 1,𝑖 × 𝑇 𝑖 ) + 𝑋 𝑖 ′
𝛾 + 𝜀 𝑖 ,𝑡 (1)

Y2,i is an active travel outcome for individual i at Time 2 (follow-up or “after opening”
survey); Y1,i is an active travel outcome for individual i at Time 1 (baseline or “before opening”
survey); Ti is the treatment condition dummy variable, equal to 1 for individuals living within ½
mile of an Expo Line light rail and 0 for the control group living farther than ½ mile from a station;
𝑋 𝑖 ′ represents the set of covariates, including transit usage and demographics; εi,t is the error term.
The interaction term 𝑌 1𝑡 × 𝑇 𝑖 was included to examine the effect of past behavior on the
relationship between new light rail transit and active travel in Time 2 (after opening). Other
covariates included in the model were transit usage variables and general demographic variables
which were typically hypothesized to influence active travel, such as age, gender, income, and
employment status. Accelerometer data from the mobile sample were fitted using a linear
regression model, and the walk trip data from the survey sample were estimated using Poisson and
negative binomial regression models as those models are better suited to characterize count data.

41

All analyses were performed using R version 3.1 and Stata version 13 (R Development Core
Team, 2014; StataCorp, 2013).

Table 1 shows the descriptive statistics of the mobile and survey samples. Roughly two
thirds of the sample is female (66% in the mobile sample; 74% in the survey sample), and average
age is 52 (SD=14) for the mobile sample and 50 (SD=14) for the survey sample. About half of the
sample is African-American (55% in the mobile sample; 49% in the survey sample), reflecting the
predominantly non-white population in the south Los Angeles neighborhood. Compared to the
control subjects’ characteristics, treatment subjects are slightly younger (51 vs 53) in the mobile
sample, but there is virtually no difference in average age among the survey sample. Treatment
subjects have greater proportion of graduate degree holders than control subjects (26% vs 12%) in
the mobile sample, but the proportion is the same across experimental and control groups in the
survey sample. The proportion of households with income above $75,000 is almost the same in
the treatment group and control group. The treatment group has a slightly more employed primary
respondents compared to the control group (59% vs 56% in the mobile sample; 63% vs 61% in the
survey sample). However, none of these differences between treatment group and control group
are statistically significant, implying that the two groups are similar in terms of demographic
characteristics.
The treatment subjects had lower train/bus usage at baseline compared to the control
subjects, but treatment subjects’ transit usage increased at follow-up. The treatment subjects spent
42

more minutes on daily MVPA than the control subjects at baseline (23 minutes vs. 20 minutes),
but the difference was not statistically significant. The treatment group significantly increased
daily train trips from baseline to follow-up (+1.14 in the survey sample, p < 0.01; +0.23 in the
mobile sample, p = 0.02). The uptake of train trips at follow-up is expected because the new Expo
Line is the only train service that is reasonably close to the study participants. The daily mean
physical activity levels decreased but daily mean walk trips increased from baseline to follow-up,
but these changes were not statistically significant, partly due to the small magnitude of the
changes. It should be noted that the survey participants who indicated car driving as their main
commute mode exhibited significantly lower levels of walking than those who indicated otherwise
(0.77 daily average walk trips vs. 1.21 daily average walk trips, p < 0.01).

43

Table 1. Descriptive statistics of study participants

Survey Sample (N=204) Mobile Sample (N=73)
Time 1 (Baseline) Time 2 (Follow-up) Time 1 (Baseline) Time 2 (Follow-up)
Treatment
(N=101)
Control
(N=103)
Treatment
(N=101)
Control
(N=103)

Treatment
(N=32)
Control
(N=41)
Treatment
(N=32)
Control
(N=41)
Mean (SD)
or %
Mean (SD)
or %
Mean (SD)
or %
Mean (SD)
or %

Mean (SD)
or %
Mean (SD)
or %
Mean (SD)
or %
Mean (SD)
or %
Age
49.68
(14.49)
49.67
(14.37)
49.68
(14.49)
49.67
(14.37)

51.38
(14.39)
53.20
(13.44)
51.38
(14.39)
53.20
(13.44)
Sex (%)

Female 79 70 79 70 72 61 72 61
Male 21 30 21 30 28 39 28 39
Race (%)
White 27 28 27 28 28 29 28 29
Black 45 53 45 53 53 56 53 56
Asian 13 11 13 11 9 2 9 2
Hispanic 9 5 9 5 6 7 6 7
Other 6 3 6 3 3 5 3 5
Education (%)

<12th grade 7 4 7 4 6 5 6 5
High school 0 6 0 6 0 7 0 7
Some college 27 23 27 23 26 29 26 29
Associate 16 10 16 10 16 12 16 12
Bachelor 26 34 26 34 26 34 26 34
Post graduate 24 24 24 24 26 12 26 12
Household Income (%)

<15k 17 15 17 15 16 15 16 15
15-35K 26 25 26 25 31 29 31 29
35-55K 20 19 20 19 9 10 9 10
55-75K 12 18 12 18 9 15 9 15
75-100K 12 11 12 11 19 15 19 15
>100K 12 14 12 14 16 17 16 17
Employment status (%)

Not employed 37 39 37 39 41 44 41 44
Employed 63 61 63 61 59 56 59 56
Average daily train trips 0.36
(1.76)
0.20
(0.81)
1.50
(3.39)
0.23
(1.24)

0.01
(0.07)
0.05
(0.15)
0.24
(0.58)
0.01
(0.07)
Average daily bus trips 3.13
(6.34)
2.23
(5.76)
2.77
(6.40)
1.83
(4.79)

0.32
(0.62)
0.43
(0.95)
0.46
(0.98)
0.24
(0.54)
Average daily MVPA
minutes

23.09
(17.49)
19.81
(18.01)
21.52
(16.24)
18.56
(17.02)
Average daily walk trips 1.00
(1.12)
0.79
(0.97)
1.29
(1.63)
0.79
(1.07)

Note. Numbers may not sum to 100% due to rounding. Significance of difference was determined by t-tests for continuous variables and Fisher
tests for categorical variables.
* p < 0.05, ** p < 0.01, *** p < 0.001

Table 2 shows the results of Poisson and negative binomial models using total walk trip counts
obtained from the survey sample (n=204). In the Poisson model, the treatment variable was
44

positively associated with the total walk trip counts (β=0.27, p=0.001). Baseline walk trips were
also positively associated with walk trip counts at follow-up (β=0.05, p<0.001). The interaction
term between treatment and baseline walk trips was negatively associated with total walk trips (β=-
0.02, p=0.008). This suggests that the positive relationship between transit and walk trips is
significantly affected by the subjects’ past walking conditions. The coefficients on increased train
and bus usage (1 if the study subject increased self-reported train or bus trips from the pre- to the
post-opening survey) were positive and highly statistically significant (β=0.54, p<0.001 and
β=0.51, p<0.001 for train and bus, respectively), indicating a high correlation between transit usage
and walking trips. Note that the treatment effect of transit proximity on walking trips was still
evident and strong after controlling for transit usage. The negative binomial regression model
results were more or less the same as that of the Poisson model, and none of the demographic
controls were statistically significant. However, the negative binomial model appears to give more
reliable results as it has substantially better goodness-of-fit measures (AIC/BIC) than the Poisson
model. On net, the walk trip models confirm our hypothesis that the new rail transit had a positive
impact on walking behavior of residents in the treatment group, but that impact appears be
influenced by the previous (before-opening) walking behavior of study participants. Persons in
the treatment group who walked less prior to the Expo Line opening had the largest post-opening
increase in walking, controlling for other factors.

45

Table 2. Poisson and negative binomial models of total walk trip counts
Variable
Poisson Model Negative Binomial Model
β p β p
Treatment (within ½ mile = 1) 0.27 0.001
***
0.52 0.021
**

Baseline walk trips 0.05 <0.001
***
0.09 <0.001
***

Treatment × Baseline walk trips -0.02 0.008
***
-0.05 0.019
**

Train usage (increased trips = 1) 0.54 <0.001
***
0.51 0.026
**

Bus usage (increased trips = 1) 0.51 <0.001
***
0.70 0.001
***

Age (20-84 y) 0.00 0.189

0.00 0.914

Gender (male = 1) 0.12 0.049
**
0.07 0.724

Household income (below 35k = 1) 0.35 <0.001
***
0.28 0.135

Employment (employed = 1) 0.27 <0.001
***
0.08 0.669

N 200 200
Log-likelihood -875.6 -551.2
AIC 9.2 5.6
BIC 814.3 101.1
Note. The dependent variable is a total walk trip count in Time 2. Four subjects with partial data were excluded from all models
* p < 0.10, ** p < 0.05, *** p < 0.01

Table 3 shows the results for the accelerometer-based physical activity data. The physical activity
data were obtained from a smaller sample but more objective accelerometer data, allowing us to
corroborate our result with a more precise measure of active travel behavior. In Model 1, the
baseline moderate-and-vigorous physical activity (MVPA) minutes were positively associated
with MVPA minutes at follow-up (β=0.47, p < 0.001). The coefficient on the treatment variable
was positive but not statistically significant (β=1.89, p=0.547), meaning that the new light rail had
no direct effect on physical activity outcomes in Model 1. Similar to the walk trip model results
presented in Table 2, Model 2 adds the interaction term 𝑌 1𝑡 × 𝑇 𝑖 . Compared to Model 1, both the
effect size and the statistical significance of the treatment effect increase in Model 2. Being in the
treatment group was associated with more minutes of daily MVPA at follow-up (β=9.29, p=0.066)
in Model 2. However, this positive treatment effect was attenuated by baseline physical activity
46

levels, as indicated by the negative sign of the interaction term between treatment and baseline
MVPA (β=-0.34, p=0.063). Model 3 shows the effect of increasing train or bus usage, by adding
dummy variables = 1 if the study subject increased self-reported train or bus trips from the pre- to
the post-opening survey. The coefficient on increased train usage was positive but not statistically
significant (β=3.14, p=0.516). However, the coefficient on increased bus usage was positive and
statistically significant (β=14.63, p<0.001). The coefficients (magnitude and significance) on the
treatment dummy variable and the interaction term 𝑌 1𝑡 × 𝑇 𝑖 are similar in Model 3 and Model 2.
Other covariates followed the expected signs and patterns as previous studies (Troiano et al.,
2008). Overall, the physical activity model results suggest that there is a positive relationship
between treatment (living close to the new transit station) and physical activity, but the interaction
variable, 𝑌 1𝑖 × 𝑇 𝑖 , indicates that treatment subjects with less baseline physical activity had larger
post-opening increases in physical activity.

Table 3. Linear regression models of moderate-and-vigorous physical activity
Variable
Model 1 Model 2 Model 3
β p β p β p
Treatment (within ½ mile = 1) 1.89 0.547

9.29 0.066
*
7.90 0.088
*

Baseline MVPA 0.47 <0.001
***
0.60 <0.001
***
0.60 <0.001
***

Treatment × Baseline MVPA -0.34 0.063
*
-0.37 0.026
**

Train usage (increased trips = 1) 3.14 0.516

Bus usage (increased trips = 1) 14.63 <0.001
***

Age (20-84 y) -0.19 0.126

-0.20 0.102

-0.24 0.038
**

Gender (male = 1) 8.91 0.010
***
10.06 0.004
***
8.51 0.009
***

Household income (below 35k = 1) 1.52 0.636

2.42 0.450

-0.42 0.889

Employment (employed = 1) 3.75 0.267

3.83 0.249

3.28 0.278

N 73 73 72
R
2
0.44 0.47 0.59
Adjusted R
2
0.38

0.41

0.53

Note. The dependent variable is daily average MVPA minutes in Time 2. One subject with partial data was excluded in Model 3.
* p < 0.10, ** p < 0.05, *** p < 0.01

47

To further illustrate the differential effects of baseline physical activity, we plotted a
smooth regression line of predicted MVPA at follow-up using Model 3 while holding age at the
mean and other covariates (binary variables) at zero. Figure 3 shows the predicted MVPA at
follow-up as a function of the light rail treatment effect and baseline MVPA while controlling for
other covariates included in Model 3. The regression lines illustrate that the previous MVPA has
a non-linear relationship with the later MVPA, and that this non-linear pattern is more pronounced
for the treatment group (solid line) than the control group (dashed line). The crossing of the two
regression lines highlights that, all else equal, the light rail treatment effect changes with the
subjects’ previous physical activity levels. Compared to the control subjects, the treatment subjects
below the 48
th
sample percentile of baseline MVPA had a positive light rail treatment effect
whereas those above the 48
th
percentile had a negative treatment effect. This graph demonstrates
that the relationship between the previous physical activity and the later physical activity is non-
linear, and the new light rail transit had a differential impact on participant’s later physical activity
depending on their previous activity levels.
48

Figure 3. Differential impact of past behavior on light rail treatment
Note. The solid line and dashed line each represent the regression line for control group and treatment group based on
Model 3 in Table 2, holding age at the mean and other covariates at zero. The regression line was fitted with a loess
function in R statistical package. Some negative values on the y-axis are based on extrapolation and are shown for
illustrative purpose. The shaded area around each regression line represents the 95% confidence interval.

Because of the small sample size for the physical activity model (Table 3), we further examined
the possibility that the results are driven by extreme values in the data set. Four additional models
were developed with different outlier removal criteria (Figure 4). The outlier models 1 and 2 define
outliers as the absolute and relative changes in MVPA between the pre- and post-surveys based on

49

a commonly used Tukey method (Tukey, 1977). Using this method, we calculated an interquartile
range for the absolute change in MVPA (outlier model 1) and the changes in MVPA relative to
pre-opening MVPA or post- opening MVPA (outlier model 2). Then, outliers were selected based
on these change scores being greater than 1.5 times interquartile range. The outlier models 3 and
4 were developed based on a comprehensive set of outlier indicators using residuals, leverage,
Cook’s D, DFITS, and DFBETA (Hamilton, 2012). The outlier model 3 excludes the most
common outliers, and the model 4 excludes all outliers identified by the outlier indicators.
Comparison of the outlier models indicates that the effect sizes and the significance values
for the key variables do not vary greatly across the different outlier criteria (Figure 4). The
coefficients on the treatment variable and the interaction term are stable and significant at the 5 to
10% level across the models. The coefficient on baseline MVPA is also stable and highly
significant at the 0.1% level across all the models. A comparison between the original model and
the outlier models indicated that the results are robust to different outlier removal criteria, and that
our model results are not driven by the potential outliers in the data set.
50

Figure 4. Comparison between the original model and the outlier models
Note. The thick line indicates 90% CI, and the thin line indicates 95% CI.

To further ensure the robustness of our models, we also checked the possibility of a
‘regression toward the mean’ effect. Regression to the mean can occur when a baseline measure
of the outcome variable is used to predict changes over time (Bland & Altman, 1994). It is possible
that in our sample subjects’ physical activity levels move toward the mean over time for reasons
having little to do with the treatment effect, and we want to confirm that the negative coefficient
on the interaction terms in Tables 2 and 3 is not simple regression to the mean. We developed a
Monte Carlo Simulation of random treatment models to test whether the negative interaction terms
reflect real interactions with the light rail treatment effect. Rather than using the treatment dummy
variable (= 1 for subjects living within ½ mile of an Expo Line station), we randomly assigned the
treatment dummy variable to subjects in the mobile sample. We then reran the regression in

51

Equation (1). The treatment effect variable, being randomly assigned, should be statistically
insignificant, as should the interaction term 𝑌 1𝑡 × 𝑇 𝑖 . If the interaction term is negative and
statistically significant in this random treatment effect model, that would be evidence that the
interaction term reflects regression to the mean. On the other hand, if the interaction term is not
significantly different from zero, that increases our confidence that the negative interaction terms
in Tables 2 and 3 indicate real interactions between baseline physical activity and the light rail
treatment, not regression to the mean. We ran a set of regression models with the randomly
assigned treatment variable using the same functional form and specifications as the original
models (the Negative Binomial Model from Table 2 and Model 3 from Table 3). These models
were replicated 1,000 times with different random assignment of the treatment variable.
Figure 5 shows the results of the Monte Carlo simulation of the random treatment models.
As expected, the mean parameter estimates for the baseline MVPA and baseline walk trips are
similar to what was observed in the original models. However, the parameter estimates for the
random treatment variable and the interaction term, 𝑌 1𝑡 × 𝑇 𝑖 where Ti is the randomly assigned
treatment variable, were both approximately symmetrically distributed with zero mean. This
suggests that there is no treatment effect with the random treatment models, and we conclude that
the interaction term in our original models does not reflect regression to the mean.

52

Figure 5. Monte Carlo simulation of the random treatment models
Note. The red vertical lines indicate the mean values. The means and SDs (in parenthesis) are 0.45 (0.19), 0.16 (6.39), and 0.01
(0.37) from left to right in the first row; and 0.07 (0.01), -0.01 (0.28), and 0 (0.03) from left to right in the second row.

This study is one of the few quasi-experimental, longitudinal studies on the relationship
between transit and active travel behavior. The unique research design provided a rare opportunity
to compare the same households before and after a major improvement in transportation
infrastructure, and we were able to employ both objective (accelerometer) and subjective (travel
surveys) measures of active travel behavior. One of the major findings from this study is that there
was no significant light rail treatment effect in a simple model that does not control for the effect
Baseline MVPA Random Treatment Interaction Term
Density
Baseline Walk trips Random Treatment Interaction Term
Density
(A) Random physical activity model results
(B) Random walk trip model results
Range of parameter estimates
Range of parameter estimates

53

of baseline walking and physical activity. This is consistent with the two previous longitudinal
studies in that exposure to the new light rail transit did not directly predict physical activity and
walking, and even resulted in a declining trend of total physical activity (Brown & Werner, 2007;
MacDonald et al., 2010). In our study, the non-linear relationship between the previous and later
MVPA also confirms that this declining pattern is present even for a relatively short period (Figure
3). In fact, declining physical activity is common in other longitudinal studies of youth and older
adults. Evenson and her colleagues (2010) reported a declining physical activity trend over a 2-
year period among a cohort of adolescent girls, and transport and neighborhood measures were not
associated with changes in physical activity. This contradicts their own findings from an earlier
cross-sectional study which showed positive associations between the transport and neighborhood
measures and physical activity (K. R. Evenson et al., 2010; K. Evenson, 2006). For older adults,
Bijnen and his colleagues (1998) found age-related reductions in total physical activity in a cohort
sample of elderly Dutch men over a 10-year period. Visser and his colleagues (2002) also found
that total physical activity of older adults (both men and women) in Amsterdam declined over a 3-
year study period.
Although we found little association between transit proximity and active travel in the
simple model, the subsequent ANCOVA models supported our second hypothesis about the role
of past behavior on the light rail treatment effect. Both walking and physical activity levels at
baseline influenced the relationship between transit and walking and physical activity at follow-
up, and after controlling for the baseline effect, the new light rail transit line was associated with
an increase in walking and physical activity for the least active study subjects. This result generally
confirms the expectations from the theories of planned behavior and habitual behavior. According
to the theories, regular and routine behaviors as TV watching and recycling would allow formation
54

of strong habits; therefore, it would be hard to break those habits and illicit behavioral change (Icek
Ajzen, 2002; B. Verplanken, Aarts, Van Knippenberg, & Moonen, 1998). However, low-
opportunity behaviors such as exercise or going to a movie would be less likely to form strong
habits. In our study, the subjects who exhibited lower levels of walking and physical activity, i.e.
those who are less likely to form a strong habit of active lifestyle, were the ones who were most
impacted by the new environment – introduction of the new light rail transit service in the
neighborhood. On the contrary, those who already maintained a more active lifestyle were less
affected by the changes in the environment. Our results suggest that environmental interventions
intended to change active travel behavior would likely to be more effective for people with
previously low walking and physical activity levels or with less tendency for habitual walking and
physical activity.
Growing empirical evidence also agrees with our findings that previous behavioral patterns
determine differential individual responses to certain interventions. In a study of a recess-based
intervention program targeted to school children (Saint-Maurice, Welk, Russell, & Huberty, 2014),
children had differential effects when exposed to various interventions. They observed that low
active students were more responsive to changes in the environment of school recess
programming, and the nature of the intervention influenced the effectiveness of the intervention.
Thøgersen (2006) also found that past behavior was a strong predictor of current behavior, over
and above the psychological and socioeconomic variables, including attitude, perceived
control/opportunities to use public transit, and car ownership. He cautiously interprets this result
as providing evidence that when behavior is performed frequently in a stable manner, habitual
processes are partly responsible for the effect of past on current behavior. Gardner (2009) observed
that habits, which correlate strongly with past behavior, influenced the intention to use a car or a

55

bicycle. He noted that among habitual travelers, their behaviors were mainly influenced by habitual
tendencies, not intentions. However, intention had a strong effect on behavior among less habitual
travelers, implying that policy interventions that are designed to affect intentions and behavior
change may have differential impacts depending on individuals’ strength of past behavior and
behavioral motivations. These findings resonate with our results in that the treatment subjects who
were the least physically active (i.e. people with less tendency for habitual exercise) were more
responsive to exposure to the new light rail transit service.
Our findings reveal an important relationship between new transit investment and active
travel behavior; however, there are several limitations to our study. First, although all households
in the study area were invited to participate, and there were no differential response rates between
the study households and the households recruited to participate in the study, participating
households could have been more highly motivated to participate in multiple phases of data
collection and to carry an accelerometer in each phase than non-respondents. Second, the small
sample size and geographic focus of this study may limit the generalizability of our findings. To
substantiate the findings of this study, it would be desirable to replicate our study using an
experimental research design, preferably in different contexts and populations.

The findings from this study provide important contributions and implications for future
research and policy. Our results indicate that environmental interventions are not straightforward,
and a health impact assessment of transit investments should use an experimental approach that
can account for potential confounders or moderators. The results of this study suggest that a new
transit service may not necessarily promote active travel for everyone in the community, calling
56

for caution in making a general assumption about the effects of transit investments on active travel
behavior. Another key contribution of this study lies in finding the differential effects of past
behavior on the relationship between new light rail transit and active travel. Our results suggest
that while transit proximity or exposure to new transit influences active travel, the impact of a new
transit service would depend on past behavior. In particular, for the treatment subjects who
previously had low walking and physical activity levels, living within a half-mile of a transit station
was associated with an increase in walking and physical activity after the opening of the new light
rail transit line.
Some policy implications can be drawn from this study. Strategies that focus on expanding
transit system and service provision should be considered as a necessary, but not sufficient,
condition for encouraging healthy travel choices. While previous studies on transit and active
travel suggest a positive association (Besser & Dannenberg, 2005; Freeland, Banerjee,
Dannenberg, & Wendel, 2013), our longitudinal analysis provides more nuanced and deeper
insights into the relationship between transit investment and changes in travel behavior. The
findings of our study suggest that the health effect of transit investment may be context-dependent
and the differences in the effect of such investment will most likely occur at the neighborhood
level. To offer one example of policy application, transit agencies may focus on addressing the
“first and last mile” problem which concerns connectivity issues to and from transit stations
(Advocacy Advance, 2014). Solving this problem requires understanding individual travel patterns
at the neighborhood level, and calls for improving neighborhood urban design elements, such as
pedestrian and bicycling access to transit stations. It also requires serious financial commitment
and integration of land use planning and transportation planning at the institutional level. However,
such strategies aimed at solving the first/last mile problem will effectively increase individual’s

57

opportunities to exert physical activity at the neighborhood, and potentially bring about positive
changes especially in a deprived and an underdeveloped community.
Another important implication of this study is that, in a place like Los Angeles where
different lifestyles and values coexist, “soft” approaches should be adopted to promote healthy
transportation choices. As opposed to “hard” measures which focus on investment and operations,
soft measures place more emphasis on management and marketing strategies (Cairns et al., 2008).
For example, transit agencies may incorporate a variety of soft measures, such as educational
campaigns, financial incentives or penalties, and local coordination for work and school travel
plans. As evident in our study, a new light rail transit may serve as an effective intervention
strategy to break the initial barrier for people who are less likely to make a habit of active lifestyle.
However, for those who already maintain an active lifestyle, promoting public transit as a healthy
transportation option may not be enough because public transit may only have a marginal impact
on their activity levels. Therefore, a desirable strategy for transit agencies would be to consider
constraints and motivations of different groups of individuals and develop more targeted
approaches depending on different market segments. Most transit agencies may not have the
capacity to adopt soft measures, and it would be necessary to create cross-departmental
partnerships to facilitate innovative marketing, operations, and management practices.
Understanding different market segments and adopting a variety of soft measures may allow
people to make healthy travel choices easy choices by removing social and psychological barriers
that inhibit positive behavioral change. The takeaway of this study is that transit agencies will need
to move beyond just providing transit service to incorporate more “soft” policy instruments in
tandem with hard policy measures to create a healthy mobility paradigm.
58

7.1. Summary of work flow
Accelerometer data processing typically involves two phases: Data Reduction and Data
Calibration. In the first phase (Data Reduction), valid hours/days are identified. Different criteria
can be applied to achieve this task, but valid minutes usually consist of periods of activity counts
excluding non-wear time (60 consecutive 0s) and valid days typically consist of days with 10 valid
hours/day (total of 600 minutes). In the second phase (Data Calibration), cleaned data are
translated into physiologically meaningful information such as METs, kcal, bouts, etc.

7.2. Data cleaning and reduction phase
Of the total 131 samples, one sample was removed because it had no data (B0EF). Also, ten
subjects were missing critical personal information, such as age and weight, so they were excluded
during the data cleaning phase (2003, 47F9, 4D4B, A101, A62D, B3F6, B5ED, B8D7, C4BA,
D99D). So the total number of samples was down to 120 subjects.

Next is the data reduction phase which follows the steps described below:
1. Identify non-valid minutes: remove 20-60 minutes of consecutive zeros (wear/non-wear time)
2. Identify outliers: typically counts > 16,000
3. Determine valid hours: typically 8-10 hours of valid measurements per day
4. Determine valid days: typically four valid days per week
5. Identify bouts: typically 10 minutes of session with 2 min interruption

59

To test the sensitivity of the different criteria, in-house R programs
1
were developed to apply five
frequently used settings described below (These criteria were adapted from Masse, 2005).
Table 1. Description of five data reduction criteria
Criteria A B C D E
Non-valid minutes
(exclude
continuous 0s)
20 min 20 min 60 min 60 min 60 min
Outliers >16,000 > 16,000 >16,000 > 16,000 > 16,000
Valid Hours 8 h 10 h 8 h 10 h 9 h
Valid Days 4 days 4days 4 days 4 days 4 days

7.3. Data calibration and summarization phase
Data calibration produces a set of dependent variables that can be used for regression analysis. The
variables are typically categorized into four groups:
1. Movement-based variable (ex. counts per day)
2. Time-based variable (ex. average MVPA minutes)
3. Energy expenditure-based variable
 Total physical expenditure (Kcal per day)
 Physical activity energy expenditure (MET-minutes, MET-hours per day)
4. Activity-based variable (daily/weekly time spent walking)

MeterPlus can read a large batch of cleaned data to perform the calibration process and to produce
a set of dependent variables. For each of the data produced by the five settings above, MeterPlus
was used to produce the total of 15 variables. Comparison of the results is presented in the next
chapter.

1
Two R codes perform batch data reduction process: Meterplus_mpd_processor_v6.R; Meterplus_mpd_processor_functions_v2.R
60

7.4. Description of the outcome measures
Total of five sets of outcome measures were created, each with different data reduction criteria.
Looking at Table 2, when four valid days is considered as the cut point, the total number of samples
and counts/days/person vary by each criterion used. It should be noted that the longer the non-
valid minutes and the shorter the valid hours, the more samples but the less counts/day/person
were obtain. For example, Criterion C (most loose criterion) produced the largest sample of 99,
but the average counts/day/person is the lowest. In contrast, Criterion B (most strict criterion)
produced the smallest sample size, but the average counts/day/person is the largest.

Table 2. Results of five data reduction criteria
Criteria A B C D E
Non-valid minutes 20 min 20 min 60 min 60 min 60 min
Outliers >16,000 > 16,000 >16,000 > 16,000 > 16,000
Valid Hours 8 h 10 h 8 h 10 h 9 h
Valid Days 4 days 4 days 4 days 4 days 4 days
Samples with 4 valid
days
93 80 99 87 96
Counts/valid
day/person
218,466 221,629 210,295 217,270 213,992

Table 3 explains each variable being created in the final data sets (The data sets re included in the
zipped file). The most important measures are VldDays, TotVdsedentary - TotVdvery_hard,
TotVdcou, KCal_sedentary - KCal_very_hard. Note that each row in the data set represents each
subject, so the measure in each row represents the total values per person. To get daily mean values,
total measures must be divided by the number of valid days. For example, under the Criterion A,
the subject 440E has 182 total minutes of moderate activity. To get the daily minutes of moderate

61

activity for this subject, 182 should be divided by the number of valid days, which is 6 in this case.
As the result, the subject 440E spent about 30 minutes of moderate activity on average per day.

Table 3. Explanation of the variables in the dataset
Variable Unit Description
HID - Household ID
Age - Age of the subject
Weight_kg Kilogram Weight of the subject in Kg
Date Date The first date in the MPD file
TotDays Day The total number of days saved in the MPD files
VldDays Day The total number of valid days saved in the MPD files
Vldhrvd Day Total valid hours for valid days only
VldHours Hour Total valid hours across all days
TotVdnot_wearing Minute Total minutes of non-valid minutes
TotVdsedentary Minute Total minutes of sedentary activity across all valid days (<100 counts/min)
TotVdlight Minute Total minutes of light activity for all valid days (101-1951 counts/min)
TotVdmoderate Minute Total minutes of moderate activity for all valid days (1952-5724 counts/min)
TotVdhard Minute Total minutes of hard activity for all valid days (5725-9498 counts/min)
TotVdvery_hard Minute Total minutes of very hard activity for all valid days (>9499 counts/min)
TotVdcou Day Total counts for valid days only
Bout_min Minute Minimum number of consecutive minutes to define a bout
Bout_cut_thres_low Count Lower activity count threshold to define a bout
Bout_cut_thres_high Count Upper activity count threshold to define a bout
Bout_tol Minute Tolerance (number of minutes that can fall outside the specified count)
Tot_bout_num Bout Total number of bouts
Tot_bout_length Minute Total length of all bouts in minutes
Tot_bout_avg Minute Average length of all bouts in minutes
Tot_kcal Kilocalorie Total calories expended in all activity categories across all valid days
KCal_mean Kilocalorie Mean daily calories expended across all valid days
KCal_peak Kilocalorie Peak daily calories expended across all valid days
KCal_not_wearing Kilocalorie Total calories expended in non-valid minutes across all valid days
KCal_sedentary Kilocalorie Total calories expended in sedentary activity across all valid days
KCal_light Kilocalorie Total calories expended in light activity across all valid days
KCal_moderate Kilocalorie Total calories expended in moderate activity across all valid days
KCal_hard Kilocalorie Total calories expended in hard activity across all valid days
KCal_very_hard Kilocalorie Total calories expended in very hard activity across all valid days

62

7.5. Comparison of data reduction criteria
To compare the results of the different data reduction criteria, I first chose one representative
measure (valid counts per day) from each criterion, and put them in a combination of ten pairs: (A,
B), (A, C), (A, D), (A, E), (B, C), (B, D), (B, E), (C, D), (C, E), (D, E). Then I created a scatter
plot and ran a correlation test to compare the relationship between the results in each pair.

Correlation between A and B (0.9351) Correlation between A and C (0.9940)

Correlation between A and D (0.9587) Correlation between A and E (0.9858)

63

Correlation between B and C (0.9255) Correlation between B and D (0.9726)

Correlation between B and E (0.9430) Correlation between C and D (0.9543)

Correlation between C and E (0.9865) Correlation between D and E (0.9658)
Figure 1. Scatterplot and correlation test for each pair of criteria (total counts for valid day)

64

As can be seen from Figure 1, there is a systematic underestimation or overestimation of the count
measures depending on which criteria I used. However, this difference is quite small. The
correlation coefficients range from 93 to 99, indicating that the outcome measures are highly
correlated with one another. In addition to the count measures, I also ran a correlation test for
different measures, including minutes in moderate activity, total kcals, and kcals in sedentary
activity. Like the count measures, all the correlation coefficients are greater than 0.9, indicating
that they are also highly correlated with one another regardless of which criterion is used.

Table 4. Correlation between different criteria in a pair
Correlation pair Counts per day Minutes in moderate activity Total KCal Kcal sedentary
A-B 0.9351 0.9205 0.9316 0.9586
A-C 0.9940 0.9949 0.9942 0.9932
A-D 0.9587 0.9583 0.9590 0.9769
A-E 0.9858 0.9872 0.9884 0.9888
B-C 0.9255 0.9072 0.9199 0.9467
B-D 0.9726 0.9657 0.9691 0.9746
B-E 0.9430 0.9349 0.9348 0.9516
C-D 0.9543 0.9541 0.9552 0.9734
C-E 0.9865 0.9872 0.9894 0.9918
D-E 0.9658 0.9680 0.9644 0.9789

7.6. Recommendation
Different data reduction criteria produced different results for all 120 samples. However, as seen
from the scatterplots and the correlation table, it can be argued that this difference is not significant.
The correlation coefficients for all combinations of count measures are above 0.9, indicating high

65

correlation among different criteria. Other outcome measures also follow the similar results, with
the correlation coefficient above 0.9. Therefore, it would be desirable to use the criterion that
produces the largest sample size, which is the Criterion C.

66

ight rail transit (LRT) is often regarded as “green” transportation. Taking LRT is
encouraged over car driving as it helps reduce harmful traffic emissions in the
environment. But, does switching from car to LRT reduce exposure to traffic-related
air pollution? Few studies have systematically examined under what circumstances traffic
exposure changes when switching from car to LRT. This study investigates whether traffic
exposure differs between car and LRT by conducting a real-time air pollution measurement while
driving a car and taking an LRT at the same time for 20 weekdays (a total of 80 trips). Simultaneous
measurement allowed us to control for daily variations in meteorological condition and other
L

67
unknown confounders. We further conducted a robustness check to test the effects of other vehicle-
specific factors, including fan strength, vehicle speed, and vehicle age. The results indicate that
the car generally exhibits higher exposure than the LRT; however, traffic exposure was
significantly altered by ventilation status and, to a lesser extent, traffic microenvironment and other
vehicle-specific factors, such as fan strength, vehicle speed, and vehicle age. The results from this
study suggest that mode shift from car to LRT will be particularly beneficial to those owning an
older vehicle with a sub-par ventilation system. This implies that free transit pass or transit subsidy
programs targeted to low income households will help poor families avoid exposure to higher
traffic emissions. Furthermore, the results provide justification for vehicle rebate programs, such
as Car Allowance Rebate System of 2011 (also known as “Cash for Clunkers”) in order to retire
older vehicles from the current vehicle stocks.

Light rail transit (LRT) has recently gained momentum as a prominent solution to urban
transportation problems. Compared to private motor vehicles, LRT can carry more passengers and
has the potential to generate less emissions per passenger mile (TRB, 2011). If a critical mass of
population can make a modal shift from automobile to LRT, it may be possible to alleviate urban
traffic congestions while achieving a significant reduction in traffic emissions. While modal shift
from car to LRT has clear potential for reducing mobile source emissions, recent studies suggest
that modal difference in pollutant exposure may be more complicated, calling for more systematic
approach in understanding traffic exposure between car and LRT.
68
Extensive studies have been conducted on modal differences in pollutant exposure;
however, the findings have been inconsistent (For example, see Kaur et al., 2007; Knibbs et al.,
2011). While there is a difference in exposure among various modes, more recent studies have
found that it is not appropriate to rank modes in order of exposure without detailed consideration
of various factors (Knibbs et al., 2011). One of the key factors actively discussed among the experts
is the influence of ventilation and transport microenvironments. For all types of enclosed vehicle,
ventilation determines how much outside toxic air flows into the cabin, and that influences the
particle concentration levels inside cars, trains, buses, and etc. Under good ventilation, in-cabin
exposure is highly affected by on-road concentrations or ambient concentration levels, which are
typically determined by transport microenvironments.
While modal difference studies dominate the commuter exposure literature, mode-specific
studies provide more comprehensive information about how exposure patterns may differ between
car users and transit users. For example, vehicle exposure studies suggest that pollutant
concentrations inside vehicles can be highly variable. Even for the same type of vehicle, in-vehicle
exposure may differ significantly depending on ventilation settings, roadway types and traffic
condition, and vehicle and driving characteristics (Fruin, Hudda, Sioutas, & Delfino, 2011; Hudda
et al., 2012; Knibbs, de Dear, & Atkinson, 2009; Ott, Klepeis, & Switzer, 2008). Previous studies
of vehicle exposure indicate that understanding factors that influence both on-road concentrations
and penetration of outside pollutants into vehicles would be critical. Therefore, an accurate
assessment of in-vehicle exposure would require consideration of key factors related to driver and
vehicle characteristics as well as transport microenvironments.
Similar to the vehicle exposure studies, transit passenger exposure studies found that
factors related to ventilation status and transport microenvironment can alter in-train PM

69
concentrations. However, for above-ground transit systems such as buses and light rail transit,
passenger exposure well correlates with ambient particle concentrations (Kam et al., 2011). Most
train systems are air-conditioned, and inside air is recirculated through ventilation systems with
very little mixing occurs with outside air except when doors are opened for passenger boarding
and alighting (Cartenì, Cascetta, & Campana, 2015; Chan, Lau, Lee, et al., 2002; Chan, Lau, Zou,
Cao, & Lai, 2002). Therefore, to accurately assess in-train exposure, other factors regarding
ambient pollutant concentration levels and penetration of outside pollutants into the transit vehicles
need to be adequately accounted for. These factors include air conditioning and ventilation status,
transit system types, and transit microenvironments.
Given high variability in commuter’s air pollution exposure, estimating the effects of mode
choice on pollutant exposure calls for understanding the effects of confounders other than mode
difference. To our knowledge, no studies have conducted exposure measurements while driving a
vehicle and taking a train at the same time. Moreover, few studies have examined the confounding
effects of ventilation condition and transport microenvironment on modal differences. To fill this
important research gap, this study examines whether traffic exposure differs between car and train
commuters by conducting a simultaneous measurement of in-transit exposure through a controlled
experiment with varying ventilation conditions and traffic microenvironments.

Figure 1 shows the two sampling routes selected for this study. The vehicle and train routes
were carefully selected to represent different transport microenvironments for urban commuters.
70
The vehicle route consists of three highways (I-10, I-110, and I-210) or local streets connecting
Culver City and Pasadena. Highway I-10 is the major east-west corridor that connects Los Angeles
to Santa Monica, with daily traffic volume reaching 280,000 vehicles. I-110 is one of the busiest
highways in the US with traffic volume reaching over 328,000 vehicles per day. I-210 is also
heavily trafficked highways with traffic volume reaching 298,000 vehicles per day. I-210 is the
major east-west corridor that connects Los Angeles and San Bernardino. The vehicle route also
includes local streets that run almost the same as the highway route, except that they are typically
four-lane local streets.
The train route was similar to the vehicle route, consisting of three lines of the Los Angeles
Metro system (Expo, Red/Purple, and Gold lines). The Expo line is a ground-level light rail transit
system connecting downtown Los Angeles to Culver City, eventually reaching downtown Santa
Monica upon completion of Phase II (expected Phase II completion by 2016). Average weekday
ridership for the Expo line is estimated to be 30,000 as of June 2015. The Red/Purple line is an
underground subway system that connects downtown Los Angeles to North Hollywood. The
Red/Purple line carries over 140,000 passengers per weekday, accounting for almost 42% of the
systemwide rail ridership. The Gold line is also a ground-level light rail system connecting
downtown Los Angeles to Pasadena, and to East Los Angeles. The Gold line is a major commuter
transit with weekday ridership estimated to be about 40,000 passengers per day. The Gold line
consists of two routes going northeast to Pasadena and southwest to East Los Angeles. Only the
northeast route that goes to Pasadena was sampled in this study.

71

Figure 1. Map of sampling routes, the vehicle route and the train route

The main objective of this study is to assess the effects of changing mode from car to LRT
on air pollution exposure under various travel conditions. We address this objective by designing
the study around two key assumptions: (1) LRT commuters have relatively consistent exposure
levels due to traveling in a temperature-controlled transit cabin on a fixed route; and (2) the effect
of switching from car to LRT on air pollution exposure is mainly driven by differences in driving
characteristics, in particular, ventilation status and traffic microenvironment. We conducted a
robustness check of the results by considering the effects of three vehicle-specific factors,
including fan strength, vehicle speed, and vehicle age. We hypothesize that different travel
conditions will significantly modify the effects of modal shift on air pollution exposure. In
72
particular, ventilation condition and traffic microenvironment of vehicle commuters will play the
key role in determining the modal difference in traffic exposure between car and LRT.
We tested our hypothesis using four different experimental conditions, each of which is
associated with distinct exposure scenarios (Figure 2). The first experimental condition involves
driving a vehicle is in an air recirculating (RC) mode with all windows closed, and the vehicle was
driven exclusively on local streets, defined as typical neighborhood streets avoiding state and
federal highways. This condition represents in-vehicle exposure scenario of low ventilation and
low traffic condition. The second experimental condition involves driving a vehicle in an RC mode
with windows closed, but the whole trip is made on highways. This condition is intended to
simulate in-vehicle exposure condition of low ventilation and high traffic condition. The third
condition involves driving a vehicle on local streets with the windows on the two front seats opened
in half. This condition is intended to measure on-road pollutant concentrations in low traffic
condition. The last experimental condition constitutes driving a vehicle on highways with the two
front seat windows opened. Similar to the third experimental condition, this condition quantifies
on-road pollutant concentrations in high traffic condition.

Traffic microenvironment

Low High
Ventilation status
Low Condition 1 Condition 2
High Condition 3 Condition 4

Figure 2. Experimental condition matrix

73
The four experimental conditions were randomly assigned for each measurement session
(see Table S1 for the actual sampling schedule, supporting material). The number of sample run
was the same across the experimental conditions—10 daily samples per mode for each condition,
yielding a total of 80 daily samples. During the field measurement, the technicians were instructed
to use a field note to record any deviation from the specified experiment condition for that sampling
session. Based on the coding from the field note, we extracted only the relevant samples that match
the intended experimental condition, minimizing any measurement errors arising from
misclassification. For example, for the second experimental condition (low ventilation + high
traffic), the technician could not avoid using local streets to get on the highways. In this case, any
instance of driving on non-highways (e.g. highway ramp, extra trip segment) was excluded from
the sample to match the corresponding experimental condition. For other experimental conditions,
the same approach was applied to extract only the relevant samples corresponding to each
experimental condition. We also manually checked the coding scheme against each experimental
condition using the video footages and the GPS trajectories simultaneously obtained during the
sampling campaign.

For the field measurement, two trained technicians simultaneously collected pollutant
samples on weekdays from October 13 through November 14, 2014, resulting in over 803 hours
of samples (a total of 80 one-way trips). For the daily sampling campaign, two sampling sessions
(two trips) were carried out, consisting of a forward trip from the Culver City Station to the Sierra
Madre Villa Station in Pasadena and a backward trip from the Sierra Madre Villa Station to the
Culver City Station (Figure 1). Along the sampling routes, one technician drove the car and the
74
other technician took the LRT at the same time from the beginning to the end and vice versa,
typically from 7:00 am through 12:00 pm. The driving route consists of various local streets and
three highways (I-10, I-110, and I-210). LRT route consists of two Metro lines: the Expo line and
the Gold line. The Red/Purple lines were excluded from the study due to a small sample size and
a different exposure condition as a subway system. As is typical with Metro at this time of the
year, the transit system was always air conditioned, except for when the train reached destination
stations.
To ensure the validity of our study protocol, we conducted a visual survey to determine
whether open-window/close-window status reflects real driving environment in the Los Angeles
area. Based on a spot visual survey of vehicles via video recordings (Supporting material, Figure
S1), it was confirmed that more than one third of the vehicles observed had their windows opened
at least in half, despite a high ambient temperature (about 86F) during the spot survey.
Furthermore, we conducted a pilot sampling for two weeks using five portable Aethalometers
(AE51, Magee Scientific
TM
, Aethlab
TM
). The pilot sampling indicated that sampling in the morning
proved to be more effective because afternoon pollutant concentrations showed little or no
differences between car and LRT (Supporting material, Figure S2). The pilot study confirmed that
the ventilation setting offered two contrasting experimental conditions – one that gives the
maximum air exchange rate (AER) close to 80 – 120 h
-1
(windows opened), and another that gives
the minimum air exchange rate close to 3.5 – 9.5 h
-1
(windows closed). Previous studies have
shown that three important parameters, such as ventilation condition, vehicle model and age, and
vehicle speed, largely influence in-vehicle exposure (Fruin et al., 2011; Hudda et al., 2012, 2011).
Because of our limitation in using only one vehicle, 2006 Ford Focus ST, we conducted a Monte

75
Carlo-type simulation to test the effects of other vehicle-specific parameters on in-vehicle
exposure and compared them against the measured concentrations inside light rail transit.

Table 1 provides a summary of the instruments used in this study. Two Condensation
Particle Counters (CPCs) were employed to measure ultrafine particle (UFP) inside the vehicle
and the rail transit cabin (Supporting material, Figure S3). The TSI’s CPC 3007 is a handheld
device that measures the number of particle size 0.01 – 1m. This device uses isopropanol as a
condensing liquid and can be operated up to 6 hours with one fill-up. For each sampling session,
we tested the flow rates of CPCs and performed a zero-check with a high efficiency particulate air
filter (HEPA) on a daily basis.

Table 1. Summary of instruments used in the study
Device Manufacturer Measures Time resolution
CPC 3007 TSI Inc., MN, USA Particle count, 10 nm - 1 m 10 secs
SidePak AM510 TSI Inc., MN, USA PM 2.5 mass concentration 30 secs
Q-trak TSI Inc., MN, USA CO, CO2, temp, humidity 5 secs
AE-51 Aethalometer AethLab CA, USA BC mass concentration 1 min
BT-Q1000XT Qstarz, Taipei, Taiwan Location (latitude, longitude) 1 sec
SJ4000 SJCAM, Shenzhen, China Video footage continuous
Smartphone Neukadye Timestamped Filed Notes Unusual events n/a

Two SidePak AM510 units measured concentrations of particulate matter ≤ 2.5 µm in
aerodynamic diameter (PM2.5) inside the vehicle and the rail transit cabin. The TSI’s SidePak is a
real-time photometric aerosol monitor which uses light scattering method to quantify the airborne
76
concentration of particulate matter size 1.0, 2.5, and 10 µm. The SidePaks were fitted with a 2.5
µm impactor to control the cut-off size of particles entering the device. The impactor was cleaned
and applied new grease after each sampling session. Prior to each sampling, the SidePak was zero
calibrated with the included HEPA filter, and the flow rate was always set to 1.7 L/min. Because
the SidePak devices were factory-calibrated using A1 test dust (Arizona road dust), the actual
particles in the air may differ in size, shape, and reflective index. Therefore, we used a calibration
factor of 0.29 to reflect the actual PM2.5 particle concentrations. Two portable Aethalometers
(AE51, Magee Scientific
TM
, Aethlab
TM
) were used to measure black carbon concentrations.
Aethalometer detects changes in the optical absorption of light transmitted through accumulated
black carbon particles captured on a quartz-fiber filter. The air is continuously pumped into the
devices, and the devices record concentration levels of black carbon content present in the outside
the vehicle and transit car. Flow rate was always set to 150 ml/min, ATN < 50. All the devices
were updated with the latest Firmware version 706. The AE51 is susceptible to shocks and
vibrations, and the measurement can be biased for short sample lengths. Thus, we selected 1-
minute sample interval to minimize biases from noisy data.
Other devices used in this study include TSI’s Q-trak, GPS, and portable video camera. Q-
trak devices were used to measure CO2, temperature, and humidity. CO2 was measured as a tracer
gas to determine whether the research staff was inside or outside the car and the transit cabin. The
GPS devices (Qstarz BT-Q1000XT) were used to determine the locational information of the
researchers while driving the car and taking the transit. The GPS data were lost during the time
when the researcher was taking the subway and inside the underground stations. The lost data
account for less than 5% of the GPS data recorded for each sampling session. To record any
unusual events, two portable video cameras (SJCAM SJ4000) were mounted inside the car and

77
clipped on the backpack of the researcher. Smartphone-based field note application (Neukadye
Time-stamped Filed Notes) was also used to record starting/ending time and any unusual events.
For each sampling campaign, all the devices including the monitoring instruments were
synchronized according to the researcher’s wristwatch to match the timestamp.

All the raw measurements were carefully post-processed based on a set of procedures (See
Section S4 for more detail, supporting material). In brief, PM2.5 measurements were corrected for
biases affected by ambient relative humidity. Light-scattering based nephalometers generally
overestimate PM 2.5 mass concentrations at higher relative humanity (McMurry, Zhang, & Lee, 1996).
Thus, a correction method used in previous research (Chakrabarti, Fine, Delfino, & Sioutas, 2004;
Ramachandran, Adgate, Pratt, & Sexton, 2003) was applied to adjust for the effect of relative humidity
on the PM 2.5 measurements (Supporting material, Figure S5). BC measurements were post-processed
using the Optical Noise-reduction Averaging algorithm developed by Hagler (2011) to
dynamically reduce any erroneous readings due to presence of optical and electronic noises
(Supporting material, Figure S6). MicroAeth device also tends to underestimate measurement with
increased filter loading (Jimenez et al., 2007; Kirchstetter & Novakov, 2007). Thus, the loading
effect of the BC measurement was corrected using the empirical function developed by
Kirchstetter and Novakov (2007) to reduce the bias introduced when sampling highly light-
absorbing particles (Supporting material, Figure S7). Lastly, UFP number concentrations were
post-processed to reduce any biases arising from particle coincidence effects (more than one particle
in the optical scattering volume at a time) when the number concentrations exceed 100,000 #/cm
3
(Supporting material, Figure S8) (Westerdahl, Fruin, Sax, Fine, & Sioutas, 2005).
78
To ensure comparability between the measurements for each instrument pair, all the
monitoring instruments were collocated for about 30 minutes before and after each sampling
campaign. The instruments were placed on a passenger seat of the technician’s vehicle side by side
while driving in a normal condition with open-windows. The measurement was taken while the
vehicle was driven to and from the daily starting position (a parking lot of the light rail transit
station at Culver City), providing a wide range of instrument readings for robust inter-comparison.
Figure 3 shows results of the inter-comparison for each instrument pair. The collocated
measurements were compared using a correlation function, and showed generally high correlations
between the two units for all the instrument pairs – the correlations were stronger for the BC and
UFP measurements (R
2
= 0.96 and 0.97, respectively) than the correlations for the PM2.5 and CO2
measurements (R
2
= 0.93 and 0.94, respectively). Note that instrument bias have been observed
between portable monitors in the past (Matson, Ekberg, & Afshari, 2004), and according to the
manufacturer, the difference between the two identical units can be within 20%. Even though the
differences between the instruments were within this margin of error, it was necessary to reduce
this instrument bias and to ensure accurate comparison of the measurements conducted while
driving and taking transit. A linear correlation equation was applied to correct the readings of the
instrument labeled “2” against the instrument labeled “1” for each pair. The base instrument
(whether it be the instrument 1 or instrument 2) for this correction was not important due to our
primary interest in the differences (i.e.  measurements) between two microenvironments (e.g. in-
vehicle vs. in-transit).

79

Figure 3. Comparison of collocated measurements for (a) PM 2.5, (b) BC, (c) UFP #, (d) CO2

The data were analyzed using boxplots, probability density, and a descriptive summary.
Wilcoxon’s signed ranks test was performed to compare the differences in measurements between
the car and the train. A nonparametric test was used instead of parametric test (e.g. t-test) because
80
of the non-normal distribution of the pollutant samples. In addition, we used an ANOVA (Analysis
of Variance) to compare the effects of the different experimental conditions on the  values. For
the ANOVA analysis, we used difference ( ) in measurements between the paired instruments,
instead of the raw measurements taken from each individual device. The differencing technique
was employed in order to minimize daily variations caused by changing meteorological conditions.
For example, PM2.5 was calculated by subtracting the in-transit PM2.5 from the in-vehicle PM2.5
for each trip. Therefore, the sign of the  values represent either the increase (+) or the decrease (-
) in air pollution exposure when switching from car to LRT. The strength of the  values represents
the magnitude of the changes in air pollution exposure when switching mode from car to LRT.

Table 2 shows the descriptive summary of the measured concentrations for PM2.5, BC,
UFP, and CO2. The general pattern is that in-transit concentration is higher than in-vehicle
concentration under the experimental conditions 1 and 2; however, the opposite pattern is observed
u the experimental conditions 3 and 4. The mean difference in PM2.5 under the experimental
condition 1 is 5.14 (p = 0.11), indicating that PM2.5 levels inside LRT are higher than the levels
experienced among car drivers when windows are closed. Note that the mean difference in PM 2.5
between LRT and Car is not statistically significant for the condition 2 (p = 0.3). In contrast, the
mean difference in PM2.5 under the condition 3 is -15.75 (p < 0.001) and -11.39 (p < 0.001) under
the condition 4.

81
Other pollutant measurements share the same pattern as the PM2.5 measurement. The mean
difference in BC ranges is 1.08 μg/m
3
(p < 0.01) under the condition 1 and 0.55 μg/m
3
(p = 0.13)
under the condition 2. However, the mean difference ranges between -0.85 μg/m
3
(p < 0.01) and -
1.22 μg/m
3
(p < 0.01) under the conditions 3 and 4. The mean difference in UFP number
concentration is 6223 #/cm
3
(p < 0.05) for the experimental condition 1 and 1017 #/cm
3
(p = 0.73)
for the condition 2. The mean difference in UFP number concentration changes to -14623 #/cm
3

(p < 0.01) under the condition 3 and -30597 #/cm
3
(p < 0.01) under the condition 4, suggesting
that in-vehicle UFP concentrations in open-windows setting is much higher than what is normally
experienced by LRT users. Note that the mean difference in UFP under the condition 4 is twice
the value of the condition 3, suggesting that on-road UFP concentrations along freeway are much
higher than the UFP concentrations along arterial roads.
As expected, the mean difference in CO2 is negative for the conditions 1 and 2, but positive
for the conditions 3 and 4. This suggests in-vehicle and in-transit CO2 concentrations exhibit the
opposite pattern as the other three pollutants. The increase in CO2 means tighter cabin
environment, resulting in lower I/O ratios and thus less exposure to on-road pollutant
concentrations. The decrease in CO2 suggests that there is more mixing between inside and outside
air, which raises the I/O ratios close to 1 and subsequently increases exposure to on-road
concentrations.

82
Table 2. Descriptive summary of the pollutant measurements
Variable
Experimental
condition
LRT (A) Car (B)
Mean
Diff.
(A - B)
p
Mean
(SD)
Median Min Max
Mean
(SD)
Median Min Max
PM2.5
(μg/m
3
)
Condition 1
28.75
(17.38)
22.76 10.22 157.27
23.61
(9.93)
20.01 10.18 76.71 5.14 –
Condition 2
31.09
(17.92)
27.32 6.78 96.26
27.1
(8.88)
26.36 10.05 74.47 3.99 –
Condition 3
39.13
(15.93)
37.36 13.98 100.81
54.88
(20.14)
56.87 15.12 124.66 -15.75 ***
Condition 4
29.81
(21.81)
24.49 6.78 270.32
41.2
(18.25)
40.51 6.94 101.35 -11.39 ***
BC
(μg/m
3
)
Condition 1
2.69
(1.63)
2.58 0.34 9.74 1.61 (0.9) 1.84 0.16 4.4 1.08 ***
Condition 2
2.57
(1.67)
2.22 0.22 12.69
2.02
(1.16)
1.65 0.35 5.99 0.55 –
Condition 3
1.98
(1.47)
1.77 0.27 11.21
2.83
(2.22)
2.04 0.58 31.95 -0.85 ***
Condition 4
1.8
(1.14)
1.68 0.18 12.24
3.02
(2.07)
2.81 0.55 22.66 -1.22 ***
UFP
(#/cm
3
)
Condition 1
23223
(9302)
20286 9299 86590
17001
(13399)
14390 994 145900 6223 **
Condition 2
22808
(10252)
20923 8396 85183
21791
(11407)
19943 1860 104621 1017 –
Condition 3
18833
(8726)
17194 8854 119978
33456
(29009)
24028 5892 487258 -14623 ***
Condition 4
21674
(8313)
20257 9219 88685
52271
(41299)
41399 7539 571277 -30597 ***
CO2
(ppb)
Condition 1
880
(271)
866 458 1773
1949
(797)
1813 701 4033 -1069 ***
Condition 2
828
(227)
776 428 1558
1814
(776)
1506 759 3652 -987 ***
Condition 3
788
(170)
721 451 1551 530 (83) 523 422 1117 258 ***
Condition 4
762
(176)
839 428 1315 566 (80) 573 431 1548 196 ***
Note: [Condition 1: low ventilation + low traffic]; [Condition 2: low ventilation + high traffic]; [Condition 3: high ventilation + low
traffic]; [Condition 4: high ventilation + high traffic]. P-values were calculated using Wilcoxon’s signed ranks test.
* p < 0.05, ** p < 0.01, *** p < 0.001

Figure 4 through Figure 7 exhibit boxplots and associated probability density of the
samples for each of the pollutant measurements. In Figure 4, the in-transit PM2.5 concentrations
are relatively consistent variation as opposed to more variations observed for the in-vehicle PM2.5
concentrations. Under the conditions 1 and 2 which represent closed-window microenvironment,

83
in-vehicle PM2.5 concentrations showed less variations than in-transit concentrations, with similar
means between LRT and car. However, the mean difference between LRT and car becomes larger
under the conditions 3 and 4 which represent open-window microenvironment. In-vehicle PM2.5
concentrations under the conditions 3 and 4 show wider sample distribution, reflecting more or
less the typical range of on-road pollutant concentrations.

Figure 4. In-transit and in-vehicle PM 2.5 concentration by experimental condition
Note: [Condition 1: low ventilation + low traffic]; [Condition 2: low ventilation + high traffic]; [Condition 3: high ventilation + low
traffic]; [Condition 4: high ventilation + high traffic]. The probability density was plotted using kernel density function.
84
The black carbon mass concentrations show similar patterns as the PM2.5 concentrations,
but there are less variability in sample distribution (Figure 5). Distribution patterns between in-
transit and in-vehicle are similar, but for the conditions 3 and 4, the in-vehicle PM2.5 concentration
distribution tends to be wider than that of in-transit concentration.

Figure 5. In-transit and in-vehicle BC concentration by experimental condition
Note: [Condition 1: low ventilation + low traffic]; [Condition 2: low ventilation + high traffic]; [Condition 3: high ventilation + low
traffic]; [Condition 4: high ventilation + high traffic]. The probability density was plotted using kernel density function.

85
UFP number concentration also follows the similar pattern as the previous two pollutants
– narrower distribution under the conditions 1 and 2, and wider distribution under the conditions
3 and 4 (Figure 6). Especially, there is a stark difference in the results for the condition 1 and
condition 4. Under the condition 1 which represent closed-window environment on arterial roads,
UFP number concentration for in-transit microenvironment has higher mean than for in-vehicle
microenvironment, but the variance looks much the same for both microenvironments. However,
under the condition 4 which is open-window environment on freeways, in-vehicle concentration
has a substantially higher mean and wider sample distribution than in-transit concentration. This
suggests that UFP is more sensitive to changes in the microenvironment than other pollutants, and
the measurement results sufficiently captured the effects of microenvironmental changes through
the experimental condition.
86

Figure 6. In-transit and in-vehicle ultrafine concentration by experimental condition
Note: [Condition 1: low ventilation + low traffic]; [Condition 2: low ventilation + high traffic]; [Condition 3: high ventilation + low
traffic]; [Condition 4: high ventilation + high traffic]. The probability density was plotted using kernel density function.

Figure 7 shows the boxplots and the probability density plots for the CO2 measurements.
Compared to the relatively consistent patterns of the in-transit concentration, in-vehicle
concentration shows wide variations in terms of sample mean and distribution. Under the condition
1 which represents the closed microenvironment on arterial roads, the mean CO2 concentrations
for the in-vehicle environment is substantially higher than that for the in-transit environment, and

87
are widely distributed from 0 to 4,000 ppb. Contrasting this result with the condition 4 which
represent open-windows microenvironment, the mean values for the in-vehicle environment are
much lower than that for the in-transit environment with much narrower probability distribution.

Figure 7. In-transit and in-vehicle CO 2 concentration by experimental condition
Note: [Condition 1: low ventilation + low traffic]; [Condition 2: low ventilation + high traffic]; [Condition 3: high ventilation + low
traffic]; [Condition 4: high ventilation + high traffic]. The probability density was plotted using kernel density function.

88
A one-way between subjects ANOVA was conducted to compare the effect of the
experimental condition on the mean difference in exposure between LRT and car (Figure 8). There
was a significant effect of travel condition on the mean difference at the 1% significance level for
all the pollutant measurements (PM2,5: F(3, 3137)=312.8, p < 0.001; BC: F(3, 3086)=327.9, p < 0.001;
UFP#: F(3, 9294)=973.2, p < 0.001; CO2: F(3, 981)=393.6, p < 0.001). These results suggest that travel
condition has a significant effect on modal difference in pollutant exposure. Specifically, our
results suggest that when commuters drive a car in a high ventilation mode (OA with open
windows), switching from car to LRT will lead to a significant reduction in exposure levels,
indicated by the negative signs of the mean difference in pollutant exposure in the experimental
conditions 3 and 4. For PM2.5, the experimental condition 4 does not appear to significantly differ
from the condition 3. However, for BC and UFP number concentrations, the effects of mode switch
is more pronounced in condition 4 than in condition 3 (Condition 3: M = -0.75, p < 0.001;
Condition 4: M = -0.75, p < 0.001), suggesting that there is a compounding effect of high traffic
microenvironment and high ventilation condition. Interestingly, mode switch from car to LRT
results in positive mean  in the experimental conditions 1 and 2 across all pollutants, except
exposure to CO2. This suggests that commuters who drive vehicles in a low ventilation condition
would experience an increase in exposure when switching from car to LRT. Unlike the results
from the conditions 3 and 4, there is little difference in the effects of traffic on mean  when
operating in a low ventilation condition (RC with closed windows). This result provides evidence
that vehicle cabin and ventilation system provide some protection from outside pollutants
regardless of traffic condition.

89

Figure 8. ANOVA results comparing different travel conditions
Note: Condition 1: low ventilation + low traffic; Condition 2: low ventilation + high traffic; Condition 3: high ventilation + low
traffic; Condition 4: high ventilation + high traffic

The results for CO2 is the opposite from the results for PM2.5, BC, and UFP. The mode
switch from car to LRT leads to a decrease in exposure when operating in a low ventilation
condition (Conditions 1 and 2) but an increase in exposure when operating in a high ventilation
condition (Conditions 3 and 4). This result makes sense because low ventilation condition inside
a vehicle creates an environment where CO2 gets trapped inside a car. A cabin environment inside
LRT is generally more ventilated than a tightly sealed passenger vehicle, therefore, commuters
90
driving a tightly sealed vehicle would experience a decrease in mean CO2 levels when making a
mode shift to LRT. Because CO2 is substantially less toxic than the other three pollutants, the
increase in CO2 in a low ventilation condition is less of a concern for most people, although long-
term exposure to CO2 may have a potential adverse effect and raise a concern for some sensitive
population.

We checked the robustness of the results against other vehicle-specific factors. Previous
studies suggest that in-vehicle I/O ratios significantly differ between RC (air recirculate) and OA
(outside air) setting, and developed predictive functions to estimate in-vehicle I/O ratios for
various travel conditions (Fruin et al., 2011; Hudda & Fruin, 2013; Hudda et al., 2012; Ott et al.,
2008). In a separate analysis, the I/O ratios computed from the predictive functions (adapted from
Hudda et al (2012)) were evaluated against the I/O ratios computed from our own sample, and the
predictive functions provided reliable estimates of I/O ratios (Supporting material, Table S5).
Using the predictive functions, we first calculated I/O ratios for different fan settings under
RC and OA condition while holding other parameters at their median values (speed 40 mph and
vehicle age 7). Likewise, the I/O ratios for different vehicle speed and vehicle age were calculated
while holding other parameters at the mean (Table 3). The calculated I/O ratios were multiplied
by the sample measurement obtained from the experimental conditions 3 and 4. We assumed that
the sample measurement from the experimental conditions 3 and 4 is representative of the on-road
concentrations for both arterial roads and freeways.

91
Table 3. I/O ratios for RC and OA under various conditions of fan speed, vehicle speed, and vehicle age
Parameters Conditions RC OA
Fan setting Low (20%) 0.18 0.72
Med (50%) 0.20 0.75
High (70%) 0.25 0.80
Vehicle speed Low (20 mph) 0.13 0.65
Med (40 mph) 0.20 0.75
High (60 mph) 0.30 0.83
Vehicle age Low (2 year) 0.14 0.73
Med (7 year) 0.20 0.75
High (11 year) 0.27 0.76

Because our pollutant samples followed a log-normal distribution, we used a log-normal
distribution to estimate an empirical distribution function from bootstrap sampling of 10,000
iterations under RC (air recirculate) and OA (outside air) condition. We then plotted a cumulative
distribution function with log-scale to compare the distributional patterns of in-vehicle against in-
transit exposure for three vehicle-specific factors, including fan strength, vehicle speed, and
vehicle age. Bootstrap mean, standard error, and 95% confidence interval were also calculated to
provide a statistical property of the empirical distribution.

Three fan settings, low (20%), medium (50%), and high (70%), were chosen to test the
effect of fan strength on in-cabin concentration. Using the calculated estimates of in-vehicle and
in-transit UFP concentrations, a cumulative distribution function was plotted with log-scale for in-
vehicle concentration under each of the parameter settings. Representative probability
distributions of UFP concentrations are shown in Figure 9. The predicted in-cabin concentrations
shows that under RC condition, we would expect to see in-transit concentrations two or three-fold
92
higher in exposure than in-vehicle concentration. Under OA condition, however, we expect that
in-vehicle concentrations would be 1.5 to 2 times higher than that of in-transit concentrations. In
both conditions, the increase in fan strength from 20% to 70% increased UFP concentration by
28%. The differences in RC and OA were larger than the uncertainly associated with the fan
strength.

Figure 9. Effect of fan strength on in-vehicle UFP concentrations compared to in-transit UFP
concentrations

93
As shown in Figure 10, three vehicle speeds, low (20 mph), medium (40 mph), and high
(60 mph), were arbitrarily chosen to test the effects of vehicle speed on in-cabin UFP
concentrations. Under RC condition, in-transit UFP concentrations were expected to be two to
four-times higher than in-vehicle UFP concentrations. As evident from the probability plot, the
effect of vehicle speed on in-vehicle UFP concentrations is more pronounced under RC condition.
The increase in vehicle speed from 20 mph to 60 mph increased the in-vehicle UFP concentration
by 50%. However, in-vehicle concentrations were still lower than in-transit concentrations across
all vehicle speed settings for RC condition. Comparing this result with the OA condition, we would
expect only a 1.2 to 1.7-fold increase in in-vehicle concentrations compared to in-transit
concentrations. This means that driving speed would have little impact on the difference between
in-transit and in-vehicle concentrations if a person typically drives under OA condition.

94

Figure 10. Effect of vehicle speed on in-vehicle UFP concentrations compared to in-transit UFP
concentrations

To examine the effect of vehicle age on UFP concentrations, three parameters, low (2
years), medium (7 years), and high (11 years), were arbitrary chosen. As can be seen from Figure
11, the effect of age is much more pronounced under RC condition than OA condition. Under RC
condition, the increase in vehicle age from 2 to 11 increases in-vehicle UFP concentration by 46%,
whereas the changes in vehicle age under OA condition have little or no impact on in-vehicle UFP
concentrations. Under RC condition, we would expect a two to three-fold increase in in-vehicle

95
UFP exposure compared to in-transit exposure. Under OA condition, in-vehicle UFP concentration
is expected to be 1.5 times higher than in-transit UFP concentration across the three vehicle ages.

Figure 11. Effect of vehicle age on in-vehicle UFP concentrations compared to in-transit UFP
concentrations

The controlled experiment provided evidence that both internal and external
microenvironments matter when quantifying the modal difference in traffic exposure. The internal
microenvironment is relatively stable for LRT commuting, whereas substantial variations exist
96
when driving an automobile. Especially, changing ventilation condition of a vehicle from closed-
windows to open-windows increased in-vehicle exposure by more than two-fold across all
pollutant measurements. This finding is in agreement with previous studies of in-vehicle UFP
exposure. Hudda & Fruin (2013) found that I/O ratios for in-vehicle PM2.5 and UFP were three
times higher for OA (outside air intake) condition than for RC (air recirculate) condition. Knibbs
and his colleagues (2010) found that median I/O ratios for in-vehicle UFP ranged between 2 and
4 depending on vehicle models based on a tunnel study. Quiros and his colleagues (2013) also
found that UFP concentration was 40%-75% higher when driving with open windows as opposed
to driving with closed windows and recirculation on.
Combined with the previous in-vehicle exposure studies, the results of this study suggest
that a simple comparison of in-transit exposure and in-vehicle exposure would fail to capture the
variability in air pollution exposure of typical urban commuters. Except for one study (Quiros et
al. 2013), most previous studies looking at the modal differences in pollutant exposure do not
control for ventilation condition or make incorrect assumptions about ventilation condition. For
example, Wang and Gao (2011) did not explicitly control for ventilation conditions when
measuring PM2.5 and fine particle concentrations for different travel modes, yielding unusually
lower PM2.5 mass concentration for automobile (400% to 1400% lower) compared to all other
modes. Briggs and his colleagues (2008), which is frequently cited as the most comprehensive
study of commuter exposure, conducted in-vehicle measurement with closed windows while
acknowledging that the ventilation status is the most important factor of in-vehicle exposure. Due
to substantial variability associated with in-vehicle microenvironment, it is critical to consider
ventilation condition of a vehicle when conducting commuter exposure studies that involve
automobiles.

97
Another important factor to consider when quantifying the effect of mode shift would be
roadway types. Driving on a freeway was two to three times more polluted (PM 2.5 and UFP) than
driving on arterial roads when windows were opened (Figure 8). This is perhaps due to the
differences in on-road concentrations between arterial roads and freeways (Fruin et al. 2008;
Weijers et al. 2004), but there seems to be a combined effect of roadway type and ventilation
condition. For example, the effect of switching mode from car to LRT on BC exposure was two to
three time larger when roadway type changed from arterial road to freeway for either closed-
window condition or open-window condition (Figure 8). Although the effect of roadway type
seems to be less influential than ventilation condition, the compounding effect of ventilation status
and roadway types is a topic that we suggest is ripe for further research.
Lastly, some other factors that were considered include vehicle fan strength, vehicle speed,
and vehicle age. Results of the robustness check demonstrated that fan strength had similar effects
on in-vehicle exposure across different settings (Figure 9), while the differences due to vehicle
speed and vehicle age were larger (two to three-fold) under RC condition than under OA condition
(Figure 10 and Figure 11). Especially, the effect of vehicle age on UFP exposure was almost
negligible when driving under OA condition (Figure 11), suggesting that certain vehicle
characteristics are more influenced by internal ventilation condition.

The results of this study suggest that modal difference in commuter exposure between car
and LRT is, in large part, driven by ventilation status. Other factors, such as roadway type, vehicle
fan strength, vehicle speed, and vehicle age, are likely to influence the modal difference in a more
subtle way. The effect of ventilation status (RC or OA) on in-vehicle exposure was several orders
98
of magnitude larger than the effects due to factors related to fan strength, vehicle speed, and vehicle
age. No other factors resulted in differences large enough to change the effect of mode shift from
car to LRT. Therefore, a common belief that switching from car to LRT would lead to reduced
exposure needs to be re-examined. Failure to consider ventilation status would lead to incorrect
assessment of commuter exposure to traffic-related air pollution.
This study offers important policy implications. People who drive their cars with windows
opened will almost always experience higher exposure regardless of all the other factors involved.
Although ventilation status has much to do with individual’s driving habit, a subset of population
with sub-par (or malfunctioned) air conditioning system has no choice but to open their windows
while driving, especially during a hot summer season. In California, it was reported that roughly
5 percent of the households are driving a vehicle older than 20 years old (2010-2012 California
Household Travel Survey). At the national level, about the same proportion (5%) of the population
was reported to drive a vehicle aged 20 or older (2009 National Household Travel Survey). If we
assume that a vehicle older than 20 years old is close to its retirement, it is highly likely that the
5% of the population owns a vehicle fleet with a sub-par ventilation system.
Given that old vehicle stocks also happen to be the highest polluters on the street, it would
be an important policy question to examine how much benefits would occur if we were to provide
financial subsidies to retire these old vehicle stocks. In the U.S, similar programs (e.g. Car
Allowance Rebate System, also known as Cash for Clunkers) were briefly introduced in 2011 as
economic stimulus while improving fuel efficiency of vehicles. If we consider the health effects
of such program, especially with regard to exposure reduction, the health benefits would be much
larger, providing further justification to continue the rebate program. Moreover, the results of the
present study suggest that mode shift from car to light rail will be particularly beneficial to those

99
owning an older vehicle because they are more likely to have an order of magnitude larger impact
on mitigating traffic exposure than those owning a newer vehicle. One idea would be to give the
option of free transit pass to the families that own old vehicle fleets and qualify as “low-income”
under the federal government’s definitions—the household income below 80 percent of an area’s
median income (AMI) after adjustment for family size. Combined together, targeting these policies
(either vehicle subsidy or free transit pass) to low income population will be justified as a way to
reduce or eliminate the disproportionately high burden of environmental pollution that is causing
much harms to the families already experiencing significant social and environmental
disadvantage.

100
S1. Random assignment of experimental condition
Table S1. Sampling schedule of field measurements (a total of 80 trips)
Sampling
Days
Date Day
Sampling
Trips
Origin-Destination*
Condition 1
(Close-
Arterial)
Condition 2
(Close-
Highway)
Condition 3
(Open-
Arterial)
Condition 4
(Open-
Highway)
Day 1 10/13/2014 Mon
Trip 1 A to B 2
Trip 2 B to A 2
Day 2 10/14/2014 Tue
Trip 3 A to B 2
Trip 4 B to A 2
Day 3 10/15/2014 Wed
Trip 5 A to B 2
Trip 6 B to A 2
Day 4 10/16/2014 Thu
Trip 7 A to B 2
Trip 8 B to A 2
Day 5 10/17/2014 Fri
Trip 9 A to B 2
Trip 10 B to A 2
Day 6 10/20/2014 Mon
Trip 11 A to B 2
Trip 12 B to A 2
Day 7 10/21/2014 Tue
Trip 13 A to B 2
Trip 14 B to A 2
Day 8 10/22/2014 Wed
Trip 15 A to B 2
Trip 16 B to A 2
Day 9 10/23/2014 Thu
Trip 17 A to B 2
Trip 18 B to A 2
Day 10 10/24/2014 Fri
Trip 19 A to B 2
Trip 20 B to A 2
Day 11 10/27/2014 Mon
Trip 21 A to B 2
Trip 22 B to A 2
Day 12 10/28/2014 Tue
Trip 23 A to B 2
Trip 24 B to A 2
Day 13 10/29/2014 Wed
Trip 25 A to B 2
Trip 26 B to A 2
Day 14 11/5/2014 Wed
Trip 27 A to B 2
Trip 28 B to A 2
Day 15 11/6/2014 Thu
Trip 29 A to B 2
Trip 30 B to A 2
Day 16 11/7/2014 Fri
Trip 31 A to B 2
Trip 32 B to A 2
Day 17 11/10/2014 Mon
Trip 33 A to B 2
Trip 34 B to A 2
Day 18 11/11/2014 Tue
Trip 35 A to B 2
Trip 36 B to A 2
Day 19 11/12/2014 Wed
Trip 37 A to B 2
Trip 38 B to A 2
Day 20 11/13/2014 Thu
Trip 39 A to B 2
Trip 40 B to A 2
* A = Culver City Station; B = Sierra Madre Villa Station 20 20 20 20

101
S2. Pilot study prior to the field measurement

Visual survey of vehicle ventilation
We conducted a visual survey to determine the proportion of vehicles with open or closed
windows. This information was necessary for determining sampling strategy for driving exposure
because in-cabin exposure levels can be significantly affected by open/close status of vehicle
windows (Hudda et al., 2012). On August 27, one hour of video was recorded in the morning
(10:27-11:27) and in the afternoon (18:19-19:19) at the intersection of La Brea Ave. and Jefferson
Blvd (Figure S1).

Figure S1. (a) Location of visual survey and (b) the camera recoding setup

Because the temperature was high (87 F in the morning; 84 F in the afternoon), we did not
expect to see many vehicles with open windows. However, the proportion of vehicles with open
windows to all vehicles was 32% in the morning, and 41% in the afternoon (Table S2). On average,
37% of vehicles were open-windows mode whereas 63% were close-windows mode. Given the
high proportion of vehicles with open windows, future sampling strategy would need to
incorporate two different in-cabin conditions: open-windows and closed-windows conditions.
102
Table S2. Summary of vehicle count
Date Time Location Count Open % Open
8/27/2014 10:27 - 11:27 La Brea & Jefferson 1197 387 32%
8/27/2014 18:19 - 19:19 La Brea & Jefferson 1645 674 41%

Sampling protocol validation
A total of five portable Aethalometers (AE51, Magee Scientific
TM
, Aethlab
TM
) were used in this
pilot study. All devices were updated with the latest Firmware version 706. Flow rate was set to
150 ml/m, ATN < 50, Minute-by-minute measurement interval was selected. Seven days of pilot
sampling was conducted from August 26 to September 3, 2014. All the sampling was conducted
during rush hour periods, typically 7:00-9:00 and 17:00-19:00. Three days were allocated for
measuring exposure while commuting by train and the other three days while commuting by car.
The flow rate was set to 150 ml/m to ensure accurate monitoring in situations with low
concentrations. One fixed route was chosen for the field measurement. A typical train commute
included a walking trip from a work location near USC to the Expo USC station, a train trip from
the USC station to Culver City station, and another walking trip from the Culver City station to a
home location near the intersection of Venice Blvd and Bagley Ave. A typical car commute
included a single trip from USC to a home location near Culver City.

Table S3 and Figure S2 present a descriptive summary and boxplots of BC exposure
between car and transit commuters. The table also includes modal differences in BC exposure
during the morning commute and afternoon commute. For both morning and afternoon commute,
the mean BC levels were higher during a transit trip (1.08 μg/m3) than a car trip (1.43 μg/m3).
Transit trip exhibited higher BC exposure than car trip regardless different times of day. The
difference in mean BC levels between car and train was statistically significant at the 0.1% level

103
across all time groups. Interestingly, the modal difference in the mean BC levels was more
pronounced during the morning peak hour. This difference can be explained by a commonly
known diurnal effect in ground-level pollutant concentrations (Oke, 1988). Pollutant
concentrations are generally higher in the morning than in the afternoon for two reasons: 1)
temperature inversion caused by cold lower layer near the surface and warmer upper layer; and 2)
intensified anthropogenic activities causing particle resuspension in the lower layer. Small modal
difference in BC in the afternoon suggests that the difference in BC in the morning contributes
most to the modal difference in BC between a car trip and a train trip.

Table S3. Descriptive summary by mode and time of day
All Morning Afternoon
car transit car transit car transit
N 809 1295 446 773 363 522
Mean 1.08 1.43 1.51 1.97 0.55 0.64
SD 1.33 1.21 1.48 1.17 0.84 0.74
Median 0.58 1.22 1.08 1.69 0.37 0.44
Min 0 0 0 0 0 0
Max 13 12 13 12 10 10
sig. diff. *** *** ***
Significant difference between the modes was tested using Wilcoxon Rank-Sum test
* p < 0.05, ** p < 0.01, *** p < 0.001

Figure S2. Modal difference in BC in (a) all day; (b) morning; and (c) afternoon
104
S3. Preparation of sampling equipment

Monitoring platform construction
Two sets of identical instruments were employed in this study. One set of instruments was
placed on a co-pilot passenger’s seat in a car (Figure S3a). The instruments were put inside a small
bag which was attached to a passenger seat next to the driver’s seat. The air tubes connected to the
instruments were slightly elevated to proximate exposure levels of the driver’s breathing zone. A
GPS (global position system) was also attached to the bag to log locational information of the car,
and a portable camera was fixed on the front windshield to record any events on the road. We also
employed a note-taking app to log any events on the field campaign. All these information were
later synchronized with the pollutant samples using a timestamp.
Another set of instruments were fitted into a backpack carried by the field technician
(Figure S3b). All the devices were strapped and put inside the backpack, and the tubes connecting
these devices were fixated on the top portion of the backpack to provide good contact with the
outside air. The air is continuously pumped into the devices, and the devices record concentration
levels of various pollutants. GPS was also attached to the backpack to record geographic location
of the field staff. A portable action cam was attached on the chest of the research staff to record
any unusual events during the field campaign. We also took field notes using a smartphone app.
The geographic information, the video clips, and the field notes were later synced with the air
pollutant samples.

105

Figure S3. Sampling instruments setup. (1) MicroAeth AE51: Black carbon, SidePak AM510: PM2.5; (2) CPC 3007:
Particle number concentrations (P-trak is shown in this picture as an example); (3) Q-Trak: CO2, CO, temperature,
humidity; (4) GPS

Instrumentation
Each set of instruments consists of four air pollution monitors, one GPS and one camera,
which were paired up for field measurements. Table S4 provides a summary of the instruments
used in this study. Two Condensation Particle Counters (CPCs) were employed to measure
ultrafine particle (UFP). The TSI’s CPC 3007 is a handheld device that measures the number of
partible size 0.01 – 1um. This device uses isopropanol as a condensing liquid and can be operated
up to 6 hours with one fill-up. For each sampling session, we tested the flow rates of CPCs and
performed a zero-check with a high efficiency particulate air filter (HEPA) on a daily basis.

106
Table S4. Summary of instruments used in the study
Device Manufacturer Measures Time resolution
CPC 3007 TSI Inc., MN, USA Particle count, 10 nm - 1 um 10 secs
SidePak AM510 TSI Inc., MN, USA PM2.5 mass concentration 30 secs
Q-trak TSI Inc., MN, USA CO, CO2, temp, humidity 5 secs
AE-51 Aethalometer AethLab CA, USA BC mass concentration 30 secs
BT-Q1000XT Qstarz, Taipei, Taiwan Location (latitude, longitude) 1 sec
SJ4000 SJCAM, Shenzhen, China Video footage 1 sec
Note-taking app Neukadye Timestamped Filed Notes Unusual events –

Two SidePak AM510 units measured concentrations of particulate matter ≤ 2.5 µm in
aerodynamic diameter (PM2.5) inside the vehicle and the rail transit cabin. The TSI’s SidePak is
a real-time photometric aerosol monitor which uses light scattering method to quantify the airborne
concentration of particulate matter size 1.0, 2.5, and 10 µm. The SidePaks were fitted with a 2.5
µm impactor to control the cut-off size of particles entering the device. The impactor was cleaned
and applied new grease after each sampling session. Prior to each sampling, the SidePak was zero
calibrated with the included HEPA filter, and the flow rate was set to 1.7L/min. Because the
SidePak devices were factory-calibrated using A1 test dust (Arizona road dust), the actual particles
in the air may differ in size, shape, and reflective index. Therefore, we used a calibration factor of
0.29 to reflect the actual PM2.5 particle concentrations.
Two portable Aethalometers (AE51, Magee Scientific
TM
, Aethlab
TM
) were used to measure
black carbon concentrations. Aethalometer detects changes in the optical absorption of light
transmitted through accumulated black carbon particles captured on a quartz-fiber filter. The air is
continuously pumped into the devices, and the devices record concentration levels of black carbon
content present in the outside the vehicle and transit car. Flow rate was set to 150 ml/m, ATN <
50. All the devices were updated with the latest Firmware version 706. The AE51 is susceptible to
shocks and vibrations, and the measurement can be biased for short sample lengths. Thus, we
selected 1-minute sample interval to minimize biases from noisy data.

107
Other devices used in this study include TSI’s Q-trak, GPS, and portable video camera. Q-
trak devices were used to measure CO2, temperature, and humidity. CO2 was measured as a tracer
gas to determine whether the research staff was inside or outside the car and the transit cabin. The
GPS devices (Qstarz BT-Q1000XT) were used to determine the locational information of the
researchers while driving the car and taking the transit. The GPS data were lost during the time
when the researcher was taking the subway and inside the underground stations. The lost data
account for less than 5% of the GPS data recorded for each sampling session. To record any
unusual events, two portable video cameras (SJCAM SJ4000) were mounted inside the car and
clipped on the backpack of the researcher. Smartphone-based field note application (Neukadye
Timestamped Filed Notes) was also used to record starting/ending time and any unusual events.
For each sampling campaign, all the devices including the monitoring instruments were
synchronized according to the researcher’s wristwatch to match the timestamp.

108
S4. Data cleaning and post-processing

Post-processing of SidePak (AM510) measurements
PM2.5 sampling were conducted at 30 s intervals and a flow rate of 1.7L/min. Fine PM
measurements from Nephalometer, such as SidePak and DustTrak, have been known to be affected
by relative humidity. Light-scattering based nephalometers generally overestimate PM2.5 mass
concentrations at higher relative humanity (McMurry et al., 1996). Relative humidity over 50%
may affect aerosol size due to water accretion on the surfaces, causing the overestimation in the
PM2.5 mass concentration. Thus, a correction method used in previous research (Chakrabarti et al.,
2004; Ramachandran et al., 2003) was applied to adjust for the effect of relative humidity on the
PM2.5 measurements.

𝐶𝐹 = 1 + 0.25
𝑅𝐻
2
(1 − 𝑅𝐻 )

𝑃𝑀
2.5
(𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑 ) =
𝑃𝑀
2.5
(𝑜𝑟𝑖𝑔𝑖𝑛𝑎𝑙 )
𝐶𝐹

CF is a correction factor, and RH denotes a relative humidity measure, obtained from an
instantaneous relative humidity reading from the Q-Trak device with the resolution of 5 s. Figure
S4 shows a one-day sample of the relative humidity measurement and the corresponding correction
factor for in-vehicle measurements on October 14, 2014.

109

Figure S4. Relative humidity (RH) and nephelometer correction factor (CF) by time of day. The data are
for in-vehicle measurement on October 14, 2014

The correction factor was taken into account for calculating the corrected PM2.5 measurement.
The corrected measurement was obtained by dividing the original PM2.5 measurement by the
correction factor for the corresponding time interval. Figure S5a compares the original PM2.5
measurement (gray dashed line) and the corrected PM2.5 measurement (black line). Figure S5b
shows the comparison between the original measurement and the measurements adjusted for the
overestimation effect of relative humidity.

Figure S5. (a) Original PM 2.5 data (dashed gray) and corrected data using the RH correction technique. The
data are for in-vehicle measurement on October 14, 2014. (b) Regression plot between uncorrected data
and corrected data for all in-vehicle measurements.

110
Post-processing of MicroAethalometer (AE51) measurements
Black carbon measurements were conducted at 30 s intervals and a flow rate of
150mL/min. MicroAethalometer measures back carbon mass concentrations using an attenuation
strength of a specific wavelength of light through a quartz fiber filter. Athalometer-based devices
were known to have sensitivity issues when sampling at short intervals at very low BC
concentrations (Hagler, 2011). The device is susceptible to mechanical shocks and vibrations due
to presence of optical and electronic noises which may affect the filter loading and create erroneous
low values or negative values. The erroneous readings do not necessarily reflect BC concentrations
at a given time.
To correct for this erroneous data, Hagler and her colleagues (2011) developed the Optical
Noise-reduction Averaging (ONA) function to dynamically reduce noise by taking into account
the light attenuation values (ATN) related to the internal loading rate of the filter. This algorithm
searches for the instances of high relative deviance in concentrations due to irregular filter loading
events and average them out. The algorithm first applies detects filter changes based on a cutoff
ATN value (the default is 5), and then calculates a cumulative sum of ATN differences. The
cumulative sums are grouped by typical incremental values expected for normal filter loading
process (the default incremental value is 0.05). Then, the BC data are averaged for each grouping
to create noise-reduced average BC measurements. The original algorithm was written in
MATLAB, and I have simplified and refined the code in R for public use in the following (Box
1).

111

Box 1. R Code for Optical Noise-reduction Averaging (ONA) algorithm

ONAprocessor = function(my.df, cutoffATN=5, defATN=0.05){
#' --------------------------------------------------
#' Optical Noise-Reduction Averaging (ONA)
#' \code{ONAprocessor} is an R verision of Gayle Hagler's ONA post-processing function
#' http://www.aaqr.org/Doi.php?id=8_AAQR-11-05-OA-0055&v=11&i=5&m=10&y=2011
#'
#' @description The function detects filter change based on cutoff ATN
#' Then, calculate every cumulative sum of ATN difference within default ATN increment value
#' And then, BC is averaged for each grouping to create bc.corr values
#' Finally, correction algorithm for Kirchstetter and Novakov (2007) is applied
#' @param my.df, pollutant data.frame
#' @param cutoffATN, cutoff value for ATN difference that detects filter change. defaul is 5
#' @param defATN, default ATN incremental value, default is 0.05
#' small value means smooth function, large value means coarse function
#' @author Andy Hong
#' --------------------------------------------------

## Calculate ATN differences
my.df$ATN.diff = c(0, abs(diff(my.df$ATN, lag=1, differences=1)))

## Determine filter change time based on ATN differences > 5
my.df$ATN.flag = ifelse(my.df$ATN.diff>cutoffATN | is.na(my.df$ATN.diff), 1, 0)

## Create a counter for each filter change
my.df$ATN.count = cumsum(my.df$ATN.flag==1)

## Calculate cummulative sum of ATN differences for each filter change phase
my.df = ddply(my.df, "ATN.count", transform, ATN.cum = cumsum(ATN.diff))

## Create a grouping index of cummulative sum < defATN
## %/% is same as division (/) but only takes the first decimal value
my.df$ATN.group = (my.df$ATN.cum %/% defATN) + 1

## Calculate cum average for each grouping
my.df = ddply(my.df, .(ATN.count, ATN.group), transform, bc.ona = mean(bc, na.rm=T))

## ---------------------------------------------------------------
## Data correction using Kirchstetter and Novakov (2007)
my.df$TR = exp(-my.df$ATN / 100)
my.df$bc.corr = my.df$bc.ona / (0.88 * my.df$TR + 0.12)

## Remove variables other than the core variables
my.df = subset(my.df, select=-c(TR, ATN.diff, ATN.flag, ATN.count, ATN.cum, ATN.group))

return (my.df)
}

112
Figure S6 illustrates one sample of the original BC data (gray line) plotted against the ONA-
processed data, and the corresponding ATN values are presented below. The ATN values
continuously increases over time, showing an incremental filter loading process. As the smoothed
line shows, the ONA approach is capable of detecting spurious spikes in the data and averaging
them to obtain smoothed data.

Figure S6. (a) Original BC data (gray) and post-processed data using the optimized noise averaging (ONA)
approach; the data are for in-vehicle measurement on October 28, 2014. (b) The corresponding filter
attenuation raw signal for the MicroAethalometer (model AE51)

Another issue with the MicroAeth device concerns underestimation of measurement with
increased filter loading (Jimenez et al., 2007; Kirchstetter & Novakov, 2007). Attenuation
coefficient (ATN) tends to diminish as filter transmission decreases with increased filter loading.
The loading effect of filter-based BC measurement was corrected using the empirical function

113
developed by Kirchstetter and Novakov (2007) to reduce the bias introduced when sampling highly
light-absorbing particles. The empirical function is as follows:

𝑇𝑟 = 𝑒 −
𝐴𝑇𝑁 100

𝐵𝐶 (𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑 ) =
𝐵𝐶 (𝑜𝑟𝑖𝑔𝑖𝑛𝑎𝑙 )
(0.88𝑇𝑟 + 0.12)

𝑇𝑟 is the MicroAeth filter transmission calculated by an exponentiated negative attenuation
coefficient (ATN). BC measurement was corrected using the 𝑇𝑟 values with a corresponding time-
base for each BC measurement interval. Figure S7a shows an example of the original BC plotted
against the corrected BC using this empirical function for the in-vehicle measurement collected on
October 28, 2014. This empirical function adjusts the original data which tend to be
underestimated when sampling at a high filter loading rate (Figure S7b).

Figure S7. (a) Original BC data (dashed gray) and corrected BC data; The data are for in-vehicle
measurement on October 28, 2014. (b) Regression plot between original data and corrected data for all
in-vehicle measurements.

114
Post-processing of CPC 3007 measurements
Ambient ultrafine particle (UFP) number concentrations were sampled at 1s intervals at a flow
rate of 700 mL/min. Westerdahl and his colleagues (2005) observed that significant particle
coincidence effects (more than one particle in the optical scattering volume at a time) occur with
the CPC 3007 device when the number concentrations exceed 100,000 #/cm
3
, resulting in
undercounting of UFP number concentrations. The following correction factor from Westerdahl
(2005) was applied to adjust the UFP measurements for the coincidence effects:

𝑈 𝐹𝑃 (𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑 ) = 38,456 × 𝑒 𝑈𝐹𝑃 (𝑜𝑟𝑖𝑔𝑖𝑛𝑎𝑙 ) × 0.0001
(𝑓𝑜𝑟 𝑈𝐹𝑃 > 100,000 𝑐𝑚 −3
)

UFP (original) is the measured UFP number concentration from the CPC 3007 Figure S8
shows a comparison of the original UFP measurement and the corrected data for a subset of the
sample (in-vehicle measurements on November 13, 2014). Figure S8a illustrates how the
exponential function adjust for the undercounting of UFP number concentrations > 100,000 cm
-3
.
For the entire sample, only 1.1% of the UFP data exceeded number concentrations > 100,000 cm
-
3
; therefore, only a small fraction of the data were corrected for the coincidence.

Figure S8. (a) Original UFP data (dashed red) and corrected UFP data; the data are for in-vehicle
measurement on November 13, 2014. (b) Regression plot between original data and corrected data for
all in-vehicle measurements.

115
S5. Inter-comparison and instrument validation
The instruments were collocated for about 30 minutes before and after each sampling
campaign in order to ensure comparability between the measurements for each instrument pair
(PM2.5, BC, UFP, and CO2). The instruments were placed on a passenger seat of the technician’s
vehicle side by side while driving in a normal condition with open-windows. The measurement
was taken while the vehicle was driven to and from the daily starting location (a parking lot of the
light rail transit station at Culver City), providing a wide range of instrument readings for robust
inter-comparison.
Figure S9 shows results of the inter-comparison for each instrument pair. The collocated
measurements were compared using a correlation function, and showed generally high correlations
between the two units for all the instrument pairs – the correlations were stronger for the BC and
UFP measurements (R
2
= 0.96 and 0.97, respectively) than the correlations for the PM2.5 and CO2
measurements (R
2
= 0.93 and 0.94, respectively). Note that instrument bias have been observed
between portable monitors in the past (Matson et al., 2004), and according to the manufacturer,
the difference between the two identical units can be within 20%. Even though the differences
between the instruments were within this margin of error, it was necessary to reduce this instrument
bias. A linear correlation equation was applied to correct the readings of the instrument labeled
“2” against the instrument labeled “1” for each pair. The base instrument (whether it be the
instrument 1 or instrument 2) for this correction was not important due to our primary interest in
the differences (i.e.  measurements) between two microenvironments (e.g. in-vehicle vs. in-
transit).
116

Figure S9. Comparison of collocated measurements for (a) PM2.5, (b) BC, (c) UFP #, (d) CO2

117
S6. Validation of the equation-based I/O ratios against the empirical measurement
The robustness check relies on the equations developed from previous studies of in-vehicle
concentrations studies (Fruin et al., 2011; Hudda & Fruin, 2013; Hudda et al., 2012). Two
equations derived from the previous studies are as follows:

Equation 1: Logit (I/O) under RC condition
𝐿𝑜𝑔𝑖𝑡 (
𝐼 𝑂 ) = −3.24 + 0.54 × 𝑓𝑎𝑛 𝑠𝑡𝑟𝑒𝑛𝑔𝑡 ℎ + 0.025 × 𝑠𝑝𝑒𝑒𝑑 + 0.103 × 𝑎𝑔𝑒 (1)

Equation 2: Logit (I/O) under OA condition
𝐿𝑜𝑔𝑖𝑡 (
𝐼 𝑂 ) = −0.29 + 0.54 × 𝑓𝑎𝑛 𝑠𝑡𝑟𝑒𝑛𝑔𝑡 ℎ + 0.025 × 𝑠𝑝𝑒𝑒𝑑 + 0.017 × 𝑎𝑔𝑒 (2)

To evaluate the validity of the above equations, we compared the I/O ratios computed from
the equation against the I/O ratios computed from our own sample. The calculated I/O ratio was
based on samples when the windows were closed and the fan was set to low settings (20%), and
this environment is similar to the RC (air recirculate) condition from the previous. In the beginning
and at the end of closed-window measurement, windows were opened for collocated measurement,
providing UFP measurement for both on-road concentration and in-vehicle concentration. We used
CO2 measurement to determine whether the measurement is for in-vehicle concentration or for on-
road concentration. To reduce noise in the data, we calculated a 20-min moving average of CO2
measurement, and visually inspected the data to identify the cut-off point that differentiate between
inside and outside condition.
As shown in Figure S10, CO2 is a good proxy to determine inside/outside concentrations.
The CO2 concentration changes abruptly around 9:10 am, and that point matches the time when
the windows were closed from our field note. Based on the visual inspection, it was determined
that 1,000 ppm was a good cut-off point that distinguishes between in-vehicle and on-road
118
concentration. We assumed that the on-road concentration is consistent throughout the sample for
that day. Thus, the I/O ratio was calculated by dividing the measured in-vehicle concentration by
the on-road concentration measured in the beginning and the end of the sampling.

Figure S10. Identification of in-vehicle and on-road concentration by moving average CO 2

In addition to the inside/outside concentration, two roadway types were used in this study to
contrast the effect of vehicle speed on I/O ratio. As shown in Figure S11, the average vehicle speed
is 38 mph for the arterial road and 48 mph for the freeway. For the arterial road type distribution
of the vehicle speed generally takes the form of unimodal distribution whereas for the freeway
road type the vehicle speed follows a bimodal distribution. The difference in distribution pattern
between arterial road and freeway is somewhat expected because of different speed limits on each
road type.

119

Figure S11. Average vehicle speed by roadway type

Using the CO2 changes and the 1,000 cut-off point as the indication of in-vehicle/ on-road
concentration, we obtained I/O ratio for two roadway types: arterial road and freeway. Table S5
shows the I/O ratios calculated for different dates. The average I/O ratio is 0.20 for the arterial
road and 0.25 for the freeway. This difference in the I/O ratio for road type is expected because of
vehicle speed difference. Also, higher traffic density on freeways result in higher on-road
concentration. Combined together, I/O ratio on freeway is expected to be higher than that on
arterial road. The calculated I/O ratios represent travel microenviroment for low ventilation
condition (20% fan setting) under RC condition.

120
Table S5. I/O ratios for arterial road and freeway under closed-window condition
Arterial road Freeway
Date I/O ratio Date I/O ratio
14-Oct 0.22 13-Oct 0.19
16-Oct 0.34 15-Oct 0.22
22-Oct 0.11 17-Oct 0.22
24-Oct 0.11 21-Oct 0.28
10-Nov 0.33 27-Oct 0.44
12-Nov 0.11 29-Oct 0.18
6-Nov 0.22
13-Nov 0.23
Average 0.20 Average 0.25

The I/O ratios from the measurement sample were compared against the I/O ratios
computed from the equation. The computed I/O ratio for the arterial road (38

mph) was 0.2, and
the ratio for the freeway (48 mph) was 0.25 (Table S7). Comparing these values against the values
obtained from the equations, the estimates were essentially the same as the sample mean for arterial
road (estimated value = 0.198; sample mean = 0.2) and freeway (estimated value = 0.241; sample
mean = 0.25). Therefore, it would be appropriate to apply the equations to estimate I/O ratios and
obtain reliable estimates for UFP number concentrations for various scenarios.

121

122

obile source emissions constitute the major source of air pollution in most urban
areas. To bring urban air pollution problems under control, regulatory agencies
have implemented top-down strategies to restrict motor vehicles on particular
days or within defined areas. In Los Angeles, a car-free street event, called CicLAvia, launched in
2010 to ban cars from streets and open them to bicyclists and pedestrians on specific Sundays. Few
studies have investigated the impacts of car-free street events on local air quality. Using an
instrumented mobile monitoring platform, we measured concentrations of particulate matter less
than 10 and 2.5 microns in size, ultrafine particle number, black carbon, and particle-bound
polycyclic aromatic compounds in three different urban locations in Los Angeles on Sundays
before, during and after CicLAvia Sunday. Meteorological conditions were similar across the
M

123

Sundays measured and most pollutants did not show statistically significant changes despite an
increase in traffic congestion generated by the event. The results of this study suggest that
CicLAvia events should be regarded as traffic-attracting events, which in turn, could offset its
positive effects in providing more opportunities for walking and biking. Furthermore, the results
suggest that continuous technological improvements in gasoline-powered vehicles now allow
minimal air quality impacts from changes in traffic congestion. While it is hopeful that car-free
street events like CicLAvia would be expected to improve air quality, more careful traffic
management plans and air pollution mitigation efforts will be needed in order to fully yield more
positive health outcomes from such events.

Air pollution is one of the most pervasive environmental problems in major cities around
the world. The World Health Organization (WHO) reports that, on average, more than half of the
cities that monitor ambient air pollution fail to meet WHO guidelines for safe levels of coarse
(PM10) and fine (PM2.5) particulate matter (World Health Organization, 2014b). The estimated
mortality related to urban air pollution amounts to 3.7 million premature deaths in 2012, and this
rate has been increasing over the last decades (World Health Organization, 2014a). Developing
countries, in particular, have experienced worsening of urban air quality due to rapid
industrialization and increased use of fossil fuels. Developed countries are not immune to air
pollution problems either. Los Angeles, for example, has been struggling to meet federal and state
standards for ozone and fine particulate matters despite several decades of massive and costly
efforts to bring the air pollution problems under control (Künzli et al., 2003). This prevailing

124

problem related to air pollution presents significant challenges to many cities as population growth
is expected to concentrate in urban areas which will likely worsen urban air quality (United
Nations, 2014).
In both developed and developing countries, tailpipe emissions from motorized vehicles
have been identified as one of the largest contributors to urban air pollution and subsequent health
impacts (Colvile, Hutchinson, Mindell, & Warren, 2001). Traffic-related air pollution can vary
greatly within a city, and the intra-urban variation in ambient pollutant concentration is largely
attributable to the differences in traffic, population density, and land use across the city(Briggs,
2000; M Jerrett et al., 2007). A growing body of research has found that residents living close to
major roadways experience disproportionately higher risk of cardiovascular and respiratory
diseases than those living farther away from traffic (Garshick et al., 2003; Gauderman et al., 2007;
Hoek et al., 2002; Schwartz et al., 2005). There is a consensus among air quality experts that
existing methods of relying on central air quality monitors misclassify levels of personal exposure
and weaken the associations with health effects (Kanaroglou et al., 2005; Wilson et al., 2005).
Another factor that compounds assessing exposure is the difficulty in managing and controlling
mobile-source emissions because of temporal and spatial variations in pollutant formation and
distribution within an urban area.
To manage urban air pollution levels, regulatory agencies have implemented demand-side
strategies, such as tolling and congestion pricing, with some limited success (Albalate & Bel,
2009). Supply-side strategies have also been adopted in some cities focusing on reducing traffic
flow and speed through traffic calming and road diet. One possible strategy to radically curtail
urban air pollution levels is to reduce the total number of cars through traffic bans on particular
streets or within defined areas. In 2015, Paris implemented a stringent traffic ban to combat severe

125

episodes of smog and achieved about 40% drop in nitrogen dioxide levels (Willsher, 2015). In
U.S., reduced traffic during the 1996 Atlanta Olympic Games led to a significant reduction in daily
peak ozone concentrations by 28% (Friedman et al., 2001). Two-day freeway closure in Los
Angeles also translated into 32% reduction in PM2.5 and 16% reduction in ozone region wide
(Hong et al., 2015). While massive reduction in traffic from freeways have shown promising
results in managing traffic pollutants, results of small-scale interventions have been mixed. A pre-
post study of the congestion charging scheme introduced in central London in 2003 indicated that
no clear changes in air quality was observed despite 18% reduction in traffic volume entering the
charging zone. Summer streets events in New York City, a small scale traffic exclusion program
similar to CicLAvia, resulted in little change in particulate mass concentrations, but substantial
reductions in ultrafine particle number concentrations (Whitlow et al., 2011). These studies offer
some evidence that the large-scale interventions may be effective for regional air quality
management, but effectiveness of small scale interventions warrants further investigation.
In Los Angeles, car-free street closure events, referred to as CicLAvia, launched in 2010
to restrict street traffic to bicyclists and pedestrians on pre-selected Sundays and to allow
participants to imagine how bicycling in Los Angeles might be without motorized traffic
(CicLAvia, n.d.). The events drew lots of media attention and likely build support for active,
emissions-free transportation, but the effect of these events on local air quality are not known.
Based on participant surveys conducted by the CicLAvia hosting organization, approximately 40-
50% of the survey respondents indicated that they drove their cars to the event (C. Batteate,
personal communication, December 17, 2016). More vehicle trips and traffic congestion in and
around the event site might generate more air pollution, potentially offsetting the positive impacts
of the car-free events. A recent study examined the impact of one CicLAvia event in Downtown
126

Los Angeles, and found a 49% reduction in PM2.5 concentrations and 21% reduction in UFP
number concentrations (Shu, Batteate, Cole, Froines, & Zhu, 2015). However, this study
encountered substantially different meteorological condition on CicLAvia Sunday compared to
comparison control Sunday conditions (which also have later traffic volume peaks and
dramatically reduced diesel vehicle traffic). In particular, the CicLAvia Sunday studied was much
warmer and had different wind patterns than typical Sunday for the morning measurement, making
it difficult to accurately assess the effects of CicLAvia.
In an attempt to obtain better conditions to compare CicLAvia events against typical
Sundays (i.e., Sundays with similar meteorology), we conducted a comprehensive study of air
quality and traffic impacts associated with three subsequent CicLAvia events in the Los Angeles
area. We measured neighborhood air quality by employing a real-time mobile monitoring platform
during CicLAvia Sundays, and compared the measurements against levels experienced during
typical Sundays within one to three weeks before or after the event. A backpack outfitted with
portable monitors was also used to measure the direct effect of street closure by walking along and
around the CicLAvia routes. In addition, we monitored traffic volume along CicLAvia corridors
and their adjacent streets using pneumatic tubes installed on major arterial roads within the study
areas. We expected that a roadway closure would reduce air pollution levels within an immediate
vicinity of the closed arterials, but surrounding traffic congestion might get worse during the
closure and negatively affect neighborhood air quality. We also observed that many event
participants chose to drive their cars to the event, potentially increasing traffic volumes compared
to other Sundays. Overall, the Ciclavia events can be considered as examples of traffic attracting
and traffic disrupting events such as sporting or other large-scale entertainment events in an urban
setting with traffic volumes already near roadway capacity (Latoski, Dunn Jr, Wagenblast,

127

Randall, & Walker, 2003). Therefore, these events are of relevance in evaluating larger questions
about traffic impacts on urban air quality, that is, what is the air quality impact of event attracting
large numbers of gasoline-powered passenger cars to a localized area of otherwise lower traffic
volumes?

Figure 1a shows the study area and the location of the three CicLAvia events. Compared
to typical Sundays, no traffic was allowed on several arterials from 9 am to 4 pm on CicLAvia
Sundays (Figure 1b and Figure 1c). The CicLAvia events took place three times in 2015, May 31
st
(Pasadena), August 9
th
(Culver City), and October 18
th
(Downtown Los Angeles, [DTLA]). The
length of the closed road segments ranged from 5.4 km (Pasadena) to 9.3 km (DTLA) and 9.9 km
(Culver City). The City of Pasadena has a population density of 3,124/km
2
and the City of Culver
City has a slightly lower population density (2,937/km
2
) than Pasadena. DTLA has a comparable
population density (2,985/km
2
) as Pasadena. Culver City and Pasadena are mostly residential
whereas DTLA is mostly commercial and mixed-used development. One of the major arterials
closed during the Pasadena event was Colorado Blvd. that has daily average traffic of 22,400 (City
of Pasadena, 2015). In Culver City, the main roadway segment that was closed during the event
was Venice Blvd., which is one of the busiest arterial roads in Los Angeles with daily average
traffic of 36,900 (LADOT, 2013). In DTLA, 7
th
Street and S. Spring Street were the two major
streets closed during the event, with average daily traffic of 16,600 and 14,000, respectively
(LADOT, 2013). Thus, the closures of these busy arterial streets in the three CicLAvia events
128

present a unique opportunity to assess the impacts of traffic regulation on local air pollution levels
in various urban contexts.

Figure 1. (a) Location of three CicLAvia events; (b) A photo taken during a typical Sunday in downtown Los
Angeles; and (c) A photo taken during the Downtown Los Angeles CicLAvia in October 18, 2015

129

We used a pre-post research design to quantify the effects of CicLAvia on local air
pollution. On CicLAvia Sunday of each event location, we measured air pollution levels along the
CicLAvia route and around the neighborhood. To get a baseline data for comparison, we measured
air pollution levels on “Non-CicLAvia” Sundays, consisting of two Sundays with similar
meteorology as the “CicLAvia” Sunday. For the Pasadena location (May 31
st
), a baseline
measurement was conducted a week before (May 24
th
) and a week after (June 7
th
) the event. For
the Culver City location (August 9
th
), we conducted a baseline measurement one week before
(August 2
nd
) and two weeks after (August 23
rd
) the event. For the DTLA location (October 18
th
),
we conducted a baseline measurement on two consecutive Sundays after the event (November 1
st

and November 8
th
) because the ambient temperature on Sundays prior to the event was
substantially different from that on the CicLAvia Sunday.
Figure 2 to Figure 4 show the detailed maps of the sampling routes for each CicLAvia
event. We selected two sampling routes per location: a driving route for mobile sampling and a
walking route for backpack sampling. Both the driving route and the walking route began at the
same time at the electricity charging point (as indicated on the map). The driving sampling route
was chosen to best represent the air pollution levels of the neighborhood around each CicLAvia
site. The driving route typically circled around either the entire CicLAvia route or a major portion
of the CicLAvia route in a clock-wise direction to allow more right turns and quicker completion
of each loop. We designed one driving loop to take about 30-40 minutes to complete in usual
traffic, but traffic conditions affected each loop completion time, generally increasing as the day
went on. The entire sampling campaign usually consisted of five to six loops in the morning (9 am
130

– 1 pm) and another five to six loops in the afternoon (2 pm – 5 pm) with roughly one-hour break
in between for charging the main battery powering the monitoring instruments.
We concurrently conducted sampling on foot to monitor air pollution levels on the streets
closed to vehicle traffic. Each walking route also included an additional north loop and a south
loop of the closed street to cover upwind and downwind of the closed street. For example, the
DTLA walking route consisted of two zones to monitor pollutant levels on different CicLAvia
routes perpendicular to each other (Figure 4). This sampling route design is similar to the cloverleaf
pattern described in the previous mobile sampling study (Larson, Henderson, & Brauer, 2009).
Prior to each measurement, we did a test run for both the driving and the walking route, and
checked for any presence of unknown issues, such as a roadway construction or a sidewalk
maintenance. We designed each walking loop to take about 40-50 minutes. Under normal walking
speed (5 kmh
-1
), a technician was able to complete four to five loops in the morning session (9 am
– 1 pm) and another four to five loops in the afternoon session (2 pm – 5 pm).

131

Figure 2. Sampling routes of Pasadena CicLAvia
132

Figure 3. Sampling routes of Culver City CicLAvia

133

Figure 4. Sampling routes of DTLA CicLAvia

134

To characterize air pollution differences on CicLAvia and Non-CicLAvia Sundays, we
used a combination of vehicle-based mobile platform and a backpack outfitted with personal
exposure monitors (Figure 5). A hybrid vehicle (2010 Honda InSight) served as the mobile
platform. To eliminate the possibility of self-pollution, the vehicle was operated in “green mode”,
which shuts off the engine when stopped. Table 1 shows a complete list of monitoring instruments.
A fan-driven sampling duct was installed across the rear windows to pull outside air into the
manifold connected to the inlet of each monitoring device. This mobile platform was capable of
measuring ultrafine particle number (UFP), black carbon (BC), and particle-bound polycyclic
aromatic compounds (PM-PAH). The platform also included TSI’s DustTrak (model DRX) which
simultaneously measured total suspended particles (TSP) and particulate matter less than 1, 10,
and 2.5 microns in size (PM1, PM10, PM2.5). Other devices outfitted in the platform include TSI’s
Q-trak, global positioning system (GPS), and portable video camera. A Q-trak device was used to
measure carbon dioxide (CO2), temperature, and humidity. The GPS device (Garmin eTrex)
recorded speed and location of the vehicle during sampling. A Sony Mini DV camera was attached
on a windshield of the vehicle to provide continuous surveillance of the road and traffic conditions.
For each sampling campaign, all the instruments were synchronized according to the GPS time for
primary timekeeping.

135

Figure 5. Instrument setup (a) mobile monitoring platform; (b) backpack-based portable instrument

Table 1. Monitoring instruments employed in the field measurement
Instrument Parameters measured Time resolution
Mobile platform
TSI Portable CPC, model 3007 Ultrafine particle count, 10 nm–1 mm 10 s
Magee Scientific portable aethalometer, model 42 Black carbon (μgm
-3
) 60 s
EcoChem PAH Analyzer, model PAS 2000 PM-PAH (ngm
-3
) 2 s
TSI DustTrak, model DRX TSP, PM1, PM2.5, PM10 (μgm-3) 10 s
TSI Q-Trak Plus Monitor, model 8554 CO, CO2, temp, humidity 10 s

Backpack instrument
TSI Portable CPC, model 3007 Particle count, 10 nm–1 mm 10 s
TSI DustTrak, model 8520 PM10 (μgm
-3
) 10 s
Brüel & Kjær mediator 2238 Type I LAeq (dBA) 5 s

In addition to the mobile platform, a small backpack was outfitted with a TSI DustTrak
(model 8520) to measure PM10 mass concentrations. PM10 was chosen as a potential marker of
resuspended road dust. The inlet of the instrument was connected to a tube which was fixed to the
top of the backpack, creating an exposure environment that mimics a normal breathing zone. A
136

portable CPC (TSI 3007) which measures UFP number concentrations was hand carried by a
trained technician during the walking sample. GPS device (Qstarz BT-Q1000XT) was also
attached to the backpack to record geographic location of the field technician. A portable action
cam (SJCAM SJ4000) was clipped on the technician’s backpack to record any unusual events
during the field measurement. QA/QC was conducted regularly by an experienced technician (See
supporting material S1 for a more detailed QA/QC procedure).

A traffic monitoring was performed at the Culver City and DTLA locations to characterize
local traffic patterns. Pneumatic tubes were placed on CicLAvia routes and several intersections
of the adjacent streets parallel to the CicLAvia route. For each monitoring location, two pneumatic
tubes were installed to measure traffic volume on both directions. To avoid possible data loss due
to broken or malfunctioned pneumatic tubes, backup counters were placed approximately one
block away from the original location. The locations of the pneumatic tubes for the Culver City
event and the DTLA event are shown in Figure 3 and Figure 4, respectively. Each traffic counter
was assigned a unique number which was later associated with pollutant measurements. Figure 6
illustrates how each traffic counter was spatially matched with pollutant measurements. Circular
buffers of diameters in 100 m multiples various sizes (100-600) were compared to measurements
(Supporting material, Figure S7 and Figure S8), and 400 m was selected as the buffer appeared to
fully contain concentration gradients while minimizing overlap with adjacent buffers.

137

Figure 6. Spatial matching between traffic data and pollutant measurements based on circular buffers; (a)
Culver City, (b) Downtown Los Angeles

In addition to the traffic counter data, we obtained traffic speed data from the portable GPS
device instrumented in the mobile platform. The GPS-derived speed data presented a snapshot of
138

the instrumented vehicle’s velocity, and therefore, provided a rough estimate of how much
congestion there was at certain point in time. For the disaggregate analysis, the speed data was
also spatially matched with each traffic counter using the 400 m circular buffer. Because the
sampling was repeated multiple times on the same route (an average of ten loops), the average
speed recorded for each traffic counter location can be thought of as a daily average of traffic speed
(congestion) experienced at that location.

Hourly meteorological data (ambient temperature in °C; relative humidity in %; wind speed
in m/s; and wind direction in degrees) were obtained from the nearest weather monitoring station
operated by South Coast Air Quality Management District (SCAQMD). These meteorological
stations are located within 5 km of the event site (Figure 2 to Figure 4). It should be noted that
there was no precipitation, except for the early morning during the DTLA CicLAvia event
(10/18/2015). Slight rain (0.5 mm) was observed between 7:27 AM and 7:47 AM before the actual
field measurement. Weather data were combined and analyzed for the duration of each sampling
campaign, typically occurred between 9 am to 5 pm. The effects of wind speed and wind direction
on air pollution dispersion was examined using time series plots with vector averaging method
(EPA, 2000). Wind rose diagrams were also generated to summarize differences in wind direction
and wind speed between CicLAvia and Non-CicLAvia days.

139

Because of daily variations in regional ambient concentration levels, we subtracted hourly
background concentrations before comparisons were made between event and non-event days. For
pollutants not measured at the regulatory stations (BC, PB-PAHs, UFP) we estimated ambient
background concentrations based on the sampling data using three algorithms: rolling minimum,
rolling 25
th
percentile, and spline of minimums (See Brantley et al. 2014 for a comprehensive
review of these algorithms). After comparing the performance of these algorithms (Supporting
material, Figure S2 - Figure S4), the spline of minimum showed the most promising result for
background estimation. The spline of minimums were then subtracted from the raw measurements
to obtain background-adjusted measurements. Probability plots of the adjusted measurements
indicated that the data generally followed a log-normal distribution; hence, the pollutant
measurements were log-transformed, and subsequent statistical tests were performed using the log-
transformed time series data.

Hourly time series plots were created for each meteorological variable as well as traffic
and pollutant measurement to examine temporal trends. The starting and ending times slightly
differed between sampling campaigns, but generally fell between 9 am and 5 pm, corresponding
to the official CicLAvia event time (9 am to 4 pm). To compare the differences in the
meteorological and traffic volume data between CicLAvia and Non-CicLAvia, we used
nonparametric Wilcoxon rank-sum test as the data were not normally distributed. For the pollutant
samples and traffic speed data, the Wilcoxon test was not directly applicable due to temporal
140

correlation violating statistical assumptions of independent samples. With high-resolution time
series such as our pollutant sample (typically 10 s time resolution), neighboring values (in time)
were more similar than distant values, and estimates of standard error underestimated sampling
variability due to inflation of the number of independent samples and producing greater chances
of statistically significant p values. (Peng & Dominici, 2008; Woodruff et al., 2009; Zwiers & Von
Storch, 1995).
To check the presence of autocorrelation in the pollutant samples, we used Durbin-Watson
(D-W) test and autocorrelation function (ACF) to visualize autocorrelation processes (Supporting
material, Figure S5). In case significant autocorrelation was detected, two approaches were
adopted to minimize the bias in the analysis. First, the data were aggregated with sample intervals
longer than 30 s. If autocorrelation was still significant, we then constructed a first-order
autoregressive model, AR(1), and performed the D-W test to determine if the residuals of the
AR(1) model still exhibited autocorrelation. We repeated these steps by applying longer sample
intervals incrementally until autocorrelation became non-significant. To analyze time series data
with significant autocorrelation, we performed a modified t-test using a novel method proposed by
Zwiers and von Storch (1995). A detailed description of the modified t-test is provided in the
sections S4 and S5 of the supporting material. In brief, this method is an extension of a traditional
t-test to correct for the underestimated standard error in autocorrelated time series by reducing the
sample size through a first-order autoregressive process. A similar method has been applied
previously to compare the means of time series samples of particulate matter with a 10-sec
sampling interval (O’Shaughnessy & Cavanaugh, 2015).
For more detailed analysis, we matched the traffic data with selected pollutant
measurements for each traffic count location. For each traffic count location, Wilcoxon rank-sum

141

test was used to compare the differences in the means of traffic volume between CicLAvia and
Non-CicLAvia, and the aforementioned modified t-test was applied to compare the means of the
traffic speed and pollutant measurements. In addition, a spatial pattern of selected pollutant
concentrations was visualized using CartoDB (“Carto DB,” n.d.) with a custom function that
calculates the average values of clustered points. This method allowed us to create a weighted
heatmap surface taking into account a given pollutant measurement along the sampling route. The
color bands for the heatmap were derived from the four percentile values (25
th
, 50
th
, 75
th
, and 95
th
)
of a given pollutant measure. All the data analyses were performed using R statistical software
(version 3.2.3), and ArcGIS 10.3 was used to create rest of the maps (ESRI, 2011; R Development
Core Team, 2014).

Figure 7 shows the hourly changes in temperature, relative humidity, wind direction, and
wind speed for all three locations. Differences in daily average values between CicLAvia and Non-
CicLAvia are presented in Table S3. Temperature did not vary significantly at the Pasadena
location, while consistently lower temperature was observed during the Culver City CicLAvia
event (p < 0.033). In downtown Los Angeles, diurnal pattern was much more pronounced during
Non-CicLAvia. However, no significant differences were observed in the mean temperature (p =
0.348) in the downtown when limiting the analysis only for the time of the measurement (9 am –
5 pm).
142

Comparison of wind speed and direction showed no statistically significant differences
between CicLAvia and Non-CicLAvia across all three locations during the hours from 9:00 am to
5:00 pm (Figure 7c and Figure 7d). Wind rose diagrams (Supporting material, Figure S1) showed
that, for the Pasadena location, the vector-averaged wind direction was northwesterly at 316˚
during Non-CicLAvia and 272˚ during CicLAvia. For the Culver City location, westerly wind
prevailed with mean vector-averaged wind directions at 258˚ during Non-CicLAvia and 243˚
during CicLAvia. The DTLA location was primarily influenced by southwesterly wind with mean
resultant wind directions at 212˚ during Non-CicLAvia and 228˚ during CicLAvia. Wind speeds
in the downtown were the lowest among all the locations, averaging 0.94 m/s during Non-
CicLAvia and 1.14 m/s during CicLAvia (p = 0.309). Across all three locations, wind speed and
wind direction were consistent between CicLAvia and Non-CicLAvia during the time of actual
field measurement from 9am to 5pm.
Relative humidity during the DTLA event was significantly lower during CicLAvia (64%)
compared to what was observed during Non-CicLAvia (74%). Slight rain during the early morning
of CicLAvia Sunday may have caused this substantial difference in relative humidity. However,
humidity is not a strong factor compared to wind and temperature; thus, we assumed that the
difference in humidity had a negligible impact on pollutant concentrations. Overall, the
meteorological patterns were similar between CicLAcia and Non-CicLAvia.

143

Figure 7. Hourly time series of selected weather parameters. (a) Hourly average temperature, (b) Hourly
average relative humidity, (c) Hourly average wind speed, (d) Hourly average wind direction.
Note: The shaded area indicates the actual sampling time corresponding to the official event duration. The hourly
average wind direction was calculated using vector-averaging of orthogonal components (u and v) of wind direction.
144

Figure 8 presents the hourly time series in traffic and selected pollutants. Generally, traffic
sharply increased from 8:00 am, peaked around 12:00 pm, and decreased after 5:00 pm. In
Pasadena, overall traffic speed decreased by 24% (p = 0.027) during CicLAvia (Supporting
material, Table S8). However, overall traffic volume and speed for the Culver City event remained
relatively the same between CicLAvia and Non-CicLAvia Sundays. The DTLA location
experienced a significant increase in traffic volume (32%, p < 0.001) and a marginal decrease in
traffic speed (-12%, p = 0.064) on CicLAvia Sunday compared to Non-CicLAvia Sundays
(Supporting material, Table S8). For the Culver City and DTLA events, slower traffic speed was
observed as the days move toward afternoon and evening, suggesting that traffic congestion
generally got worse as the event progressed (Figure 8b).
In terms of pollutant concentrations from the mobile samples, all three locations showed
different temporal patterns and percentage changes. The patterns were similar for both unadjusted
and background-adjusted measurements, so we only report the adjusted measurements. In
Pasadena, a consistent increase in all mobile pollutant samples was observed during CicLAvia
(adjusted PM10 = 21%, p < 0.001; adjusted PM2.5 = 145%, p < 0.001; adjusted BC = 124%, p <
0.001, adjusted PAH = 48%, p < 0.001, adjusted O3 = 10%, p = 0.003). In Culver City, consistently
lower levels in adjusted PM10 (-24%, p < 0.001), PM2.5 (-18%, p < 0.001), and BC concentrations
(-23%, p < 0.001) were observed during CicLAvia. However, adjusted PAH concentrations were
significantly higher (%75, p < 0.001) during the event. The DTLA location showed similar results
as the Culver City event. Adjusted PM10 decreased 18% (p < 0.001) during CicLAvia, but adjusted
BC and PAH concentrations increased 19% (p < 0.001) and 56% (p = 0.004), respectively.

145

Overall, levels of larger-size particles such as PM10 concentrations significantly increased
during the Pasadena event, but Culver City and DTLA locations experienced a significant decrease
in levels of coarse particles. In contrast with the coarse particles, UFP number concentrations were
higher during CicLAvia across all the locations (Pasadena: 60%, p < 0.001; Culver City: 15%, p
= 0.002; DTLA: 3%, p < 0.079). Interestingly, for the walking samples, a downward trend in both
adjusted PM10 and UFP levels were experienced during CicLAvia in all the locations, but only the
decrease in PM10 levels was statistically significant.
146

Figure 8. Hourly time series of traffic and selected pollutant measurement from the mobile platform. (a)
Hourly average traffic volume, (b) Hourly average traffic speed, (c) Hourly average adjusted PM 10 mass
concentration, (d) Hourly average adjusted PM 2.5 mass concentration, (e) Hourly average adjusted UFP
number concentration, (f) Hourly average adjusted BC mass concentration

Note: All the pollutant measurements are based on background-adjusted values. The shaded area indicates the actual
sampling time corresponding to the official event duration.

147

We conducted a more detailed analysis using spatially resolved traffic data and pollutant
samples. In Culver City (Figure 9a), except for the traffic counter location 2 where traffic volume
decreased (-19%, p = 0.02), traffic volume increased at the traffic counters 1, 4, and 5 (7%, 12%,
and 65%, p < 0.05). Note that the counter location 6 is on a CicLAvia route where no traffic was
allowed during the event, hence 100% reduction in traffic during CicLAvia. Traffic speed
decreased across all the locations, but only the locations 2 and 3 were statistically significant
(location 2 = -33%, p = 0.04; location 3 = -33%, p = 0.06, Figure 9b). Across all the locations in
Culver City, adjusted PM10 levels displayed a consistent decline, ranging from -8% to -36%, and
these reductions were statistically significant at least at the 10% level, except for the one at the
traffic counter location 5. A statistically significant reduction in adjusted PM 2.5 levels was
observed at the traffic counter locations 2 and 3, but the results in other locations were not
statistically significant. Adjusted BC mass concentrations also showed a general decline across the
locations; however, none of the differences were statistically significant. The UFP number
concentrations showed more or less mixed results, showing a decrease at some locations (locations
3, 4, and 6) and an increase at some other locations (locations 1, 2, and 5). However, except for
the location 6 that has a marginally significant reduction (-26%, p = 0.096), none of the differences
were statistically significant.
In downtown Los Angeles, traffic volume increased across the board during CicLAvia,
with percentage changes in the mean values ranging from 21% to 86%. These changes were
statistically significant at least at the 10% level, except for the locations 4 and 5 (Figure 10a). We
observed a statistically significant decline in traffic speed across the locations 1 through 4 (-19%
to -40%, Figure 10b). Based on the traffic data and field observation, it is apparent that the
148

CicLAvia event attracted more traffic and caused severe traffic bottleneck around the event site.
During the CicLAvia event, traffic congestion generally worsened in the downtown area, possibly
due to traffic detour and limited number of road crossings. In terms of pollutant measurements,
the adjusted PM10 levels decreased across the traffic count locations (-4% to -70%), but these
changes were only statistically significant at the traffic counter locations 2, 5, 7, and 8. PM2.5 levels
during CicLAvia showed a significant increase (36%, p = 0.052) at the location 6 but marginally
significant decrease (28%, p = 0.06) at the location 2. Similarly, the adjusted BC mass
concentrations showed mixed results, but none of the changes were statistically significant.
Changes in the adjusted UFP number concentrations were also mixed – a decrease at the locations
1, 2, 3, and 7, and an increase at the locations 4, 5, 6, and 8. But, only the changes at the location
2 (-26%, p = 0.063) and location 8 (22%, p = 0.018) were statistically significant at least at the
10% level.

149

Figure 9. Boxplots of selected pollutants by traffic counter location in Culver City: (a) Traffic volume, (b)
Traffic speed, (c) Adjusted PM 10 mass concentration, (d) Adjusted UFP number concentration, (e) Adjusted
PM 2.5 mass concentration, (f) Adjusted BC mass concentration
Note: All the pollutant measurements are based on background-adjusted values. Red dot indicates the mean values.
Wilcoxon rank-sum test was used for traffic volume, and modified t-test was used for traffic speed and pollutant
concentrations. Statistical significance. Statistical significance: *p <0.10; **p <0.05; ***p <0.01
150

Figure 10. Boxplots of selected pollutants by traffic counter location in DTLA: (a) Traffic volume, (b) Traffic
speed, (c) Adjusted PM 10 mass concentration, (d) Adjusted UFP number concentration, (e) Adjusted PM 2.5
mass concentration, (f) Adjusted BC mass concentration
Note: Red dot indicates the mean values. Wilcoxon rank-sum test was used for traffic volume, and modified t-test was
used for traffic speed and pollutant concentrations. Statistical significance: *p <0.10; **p <0.05; ***p <0.01

151

Figure 11 displays the results of heatmaps showing the spatial patterns of adjusted PM 10
and UFP during CicLAvia and Non-CicLAvia in Culver City. The results are generally consistent
with the overall trends summarized in Figure 9. Adjusted PM10 concentrations declined across the
sampling locations, whereas adjusted UFP number concentrations generally increased during
CicLAvia, broadly consistent with changes in traffic speed (e.g., less resuspended dust at lower
speeds). For example, adjusted PM10 levels reached over 16 μg/m
3
around the segment B during
Non-CicLAvia, but the PM10 levels at that segment reduced to about half during CicLAvia. The
UFP levels seemed to be elevated during CicLAvia across all the sampling locations. Interestingly,
the segment D (Lincoln Blvd.), one of the heavily trafficked corridors, experienced a decline in
adjusted PM10 concentrations but showed an increase in UFP levels during CicLAvia.

152

Figure 11. Heatmaps for the Culver City location: Adjusted PM 10 levels during (a) Non-CicLAvia and (b)
CicLAvia; Adjusted UFP levels during (c) Non-CicLAvia and (d) CicLAvia.

Figure 12 shows the spatial patterns of adjusted PM10 and UFP in downtown Los Angeles.
Adjusted PM10 concentrations generally decreased during CicLAvia on almost all the sampling
locations. The results of the UFP levels are less apparent than those of PM 10. The UFP levels
during CicLAvia appear to be slightly decreased across the sampling locations, but the levels
consistently peaked around 17,103 #/cm
3
throughout the sampling route on either Non-CicLAvia
Sundays or CicLAvia Sunday. Overall, the heatmaps of the Culver City and the DTLA events

153

indicated that a clear downward trend in PM10 was observed throughout the sampling locations,
but less clear spatial patterns were found in UFP number concentrations.

Figure 12. Heatmaps for the DTLA location: Adjusted PM 10 levels during (a) Non-CicLAvia and (b) CicLAvia;
Adjusted UFP levels during (c) Non-CicLAvia and (d) CicLAvia

The results of the three CicLAvia events indicate that a CicLAvia event had a varying
effect on local pollutant concentrations. Compared to the baseline measurement collected during
154

Non-CicLAvia Sundays, results from the Pasadena location showed that a two-fold increase in
larger-size particles such as PM10 were observed during the CicLAvia event. Because no traffic
volume data were available for the Pasadena location, it is hard to tell what exactly caused these
extreme changes in particulate matter and black carbon concentrations during the event. Ambient
PM10 levels recorded at the nearest South Coast AQMD monitoring stations (North Main St. in
Downtown Los Angeles) indicated that the ambient PM 10 concentrations were about 17% higher
(p = 0.054) on CicLAvia Sunday compared to Non-CicLAvia Sundays (Supporting material,
Figure S9 and Table S5). Ambient PM2.5 levels during CicLAia were twice (p < 0.001) that of
Non-CicLAvia levels (Supporting material, Table S5). Because regional ambient concentrations
largely affect PM2.5 concentrations, it is likely that the regional differences in ambient
concentrations may have influenced the mobile measurements. However, even after removing the
influence of the background concentrations, this upward trend in pollutant concentrations still
remained, implying that there may be an independent effect of CicLAvia. One possible reason for
the elevated particle and BC levels could be due to other local pollutant sources. During the
Pasadena event, there were many diesel-powered vehicles operating around the event site during
the Pasadena event, and the presence of these support vehicles may have affected the levels of BC
and PM-PAH.
For the Culver City event, we observed a significant decline in PM10, a marginal reduction
in PM2.5, and a mixed results in UFP number concentrations. We offer two explanations for this
result. First, there is a substantial difference in how road traffic affects particle concentrations.
PM10 is affected by resuspended road dust and tire abrasion (Lenschow, 2001), whereas UFP is
largely affected by motor vehicle emissions via dynamics of traffic congestion (including
concentration spikes from hard accelerations) and vehicle fleet compositions (Fruin, Westerdahl,

155

Sax, Sioutas, & Fine, 2008). Our measurements indicate that more congestion on event day would
lower traffic speeds on arterial streets, produce more hard accelerations and long queues at stop
lights, which in turn, would lead to lower PM10 but higher UFP concentrations. Second, a larger
magnitude of reduction in the adjusted mobile measurement compared to the ambient levels
suggests that other factors are also at play (Supporting material, Table S6). Despite some traffic
congestion occurred during the event day, we note that the total traffic volume substantially
reduced (36%, p < 0.001) due to the closure of highly trafficked arterial roads (e.g., Venice Blvd.
and Washington Blvd.). A back-of-the-envelope calculation (Supporting material, Table S8)
indicated that magnitude of the reduction was substantial enough to offset some traffic congestion
experienced during the event. The stronger decline in both unadjusted and adjusted PM 10 levels
during CicLAvia suggest that despite traffic congestion, larger magnitude of traffic removal
possibly resulted in lowering the PM10 levels.
Similar to the results from the two previous events, a significant decline in PM10 and mixed
PM2.5 and UFP levels were observed during the DTLA event. The background concentrations
around downtown Los Angeles indicate a strong downward drop in PM10 and a strong upward
increase in PM2.5 during CicLAvia (Supporting material, Table S7). After removing the influence
of the background concentrations, the difference between changes in PM 10 and PM2.5 still
remained. One possible explanation of this difference among the PM measurements is the
interaction between emissions of precursors (mainly VOCs and NOx) and local meteorology. It is
well known that mobile sources not only contribute to primary particles but also contribute to
secondary PM2.5 through oxidation processes of ozone and photochemical reaction of other
gaseous precursors from tailpipe emissions and evaporative losses, such as SO2, NO2, ammonia,
and VOCs. These gaseous precursors from mobile sources account for significant fractions of
156

PM2.5 in the Los Angeles area (Gillies, Gertler, Sagebiel, & Dippel, 2001; Kim, Teffera, & Zeldin,
2000). In DTLA, the ambient PM2.5 measurements showed frequent peaks throughout the
CicLAvia event (Supporting material, Figure S9b), suggesting that the PM2.5 levels north of the
downtown could be under the influence secondary PM sources. Furthermore, winds were more
southwesterly (mean wind direction, 228˚), possibly transporting the traffic pollutants accumulated
in the downtown core to the north. Severe traffic congestion occurred in the downtown core
combined with the prevailing southwest winds likely complicated the urban mix of primary and
secondary PM2.5 species, and possibly led to the difference between the PM10 and PM2.5
measurements.

This study investigated the impact of local street closures on traffic and air pollution
through three case studies of CicLAvia, a car-free street event in Los Angeles. To quantify the
localized effects of such small-scale intervention, we employed an instrumented mobile
monitoring platform and a backpack outfitted with portable monitoring equipment. The mobile
monitoring results indicated that the effects of the roadway closure varies across the city. Levels
of PM10, PM2.5, and UFP number concentrations significantly increased during the Pasadena event.
However, we observed a consistent and a statistically significant decline in PM10 levels but mixed
results in UFP levels during the Culver City and DTLA events. These varying results in particle
concentrations are somewhat unexpected as roadway closure is likely to have some localized
effects in removing pollutant concentrations.
The results of this study highlight the variable and somewhat unpredictable localized
effects of varying traffic congestion on concentration and dispersion of traffic pollutants. During

157

all three CicLAvia events, roadway closure and detours created some confusion and delays among
drivers passing by the event site. Furthermore, many CicLAvia participants drove their cars to the
event and circled around the neighborhood in search for limited parking spots. Combined together,
the CicLAvia events looked very similar to what traffic engineers would call a planned special
event—the term that is used to describe large-scale events such as sports game, concerts, and
festivals (Latoski et al., 2003). The severe traffic congestion experienced during all three events
counterbalanced the effect of the roadway closure, and likely led to the mixed results in traffic-
related pollutants, such as UFP, BC, and PM-PAH.
Another implication of this study is that recent and continuing technological improvements
in gasoline-powered vehicles may have resulted in the current fleet being very low emitting,
collectively, once cold starts are excluded. Despite the continuous growth in the number of motor
vehicles in California, stricter emission standards and advanced emission control technologies
likely increased fuel consumption while reducing the amount of traffic pollutants being emitted
from both gasoline- and diesel-powered vehicles (Lurmann, Avol, & Gilliland, 2015; McDonald,
Dallmann, Martin, & Harley, 2012; McDonald, Gentner, Goldstein, & Harley, 2013). The results
of the current study is consistent with this declining trend in mobile-source emissions in California.
The CicLAia events seemed to attract more traffic and disrupt normal traffic flow, and more traffic
generated during such events, in the past, would have led to a substantial increase in traffic-related
pollution in and around the event location. However, the impacts of such events on the localized
pollutant concentrations in Los Angeles now appear to be minimal, due in part to the continuous
improvements in emissions control technologies spurred by federal regulations (Pokharel, Bishop,
Stedman, & Slott, 2003).
158

In conclusion, while it is hopeful that car-free street events, such as CicLAvia, would be
expected to improve air quality, our results indicate that the localized effects of street closure are
mixed and are likely to be complicated by local meteorology and possible disruptions of normal
traffic patterns, especially in crowded urban areas. We also noted that severe traffic congestion
occurred during the three CicLAvia events did not lead to the corresponding increase in traffic
pollutants, thanks to the continuous improvements in vehicle technologies. One important lessons-
learned from this study, however, is that more careful traffic management plans and air pollution
mitigation efforts will be needed in order to fully yield all the positive environmental health
outcomes from car-free street events.

159

S1. QA and QC for mobile sampling
For quality assurance (QA), we performed zero checks for CPC, DustTrak, and DRX on a
regular basis, and none of the devices failed the test. All the instruments calculate flow rate
internally, and we used a flow check using a TSI’s flowmeter (model 4140) to conduct a flow
check when needed. For the alcohol-based CPC, we generally soaked the cartridges in fresh
alcohol that was changed at least once a week when we were actively sampling. The wicks for the
CPC were always soaked in alcohol overnight after each sampling day, and we carried extra wicks
while sampling in case of replacement during sampling.
The mobile platform required more specific QA procedure. Before starting a sampling run,
we made sure the battery voltage for the main battery that powers all the instruments is between
12.5-13V. The proper functioning of the air-intake and plenum fan are verified before sampling.
All instruments were started and warmed up for a minimum of 45 minutes and ambient
concentrations recorded at the end of this warm up period. Visually check to see if GPS and video
recorder are on and recording while the mobile platform is in motion. All instruments were
allowed to run for 45 minutes at the end of sampling and ambient concentrations recorded again.
Quality control (QC) during the sampling campaign consists of visually inspecting all instruments
for proper performance and the data stream is continuously monitored. If we detect any error
messages, we stopped sampling and inspected the issue. The measurements were considered valid
only when the issue was resolved, and the measurement taken during a diagnosis was flagged as
erroneous values.

160

S2. Wind rose diagrams

Figure S1. Wind rose diagrams showing the frequency of wind directions classified by wind
speed, during the hours from 9:00 am to 5:00 pm: a) Pasadena; b) Culver City; c) Downtown Los
Angeles (DTLA)

161

S3. Adjustment for background concentrations
Because of daily variations in ambient concentration levels, it was necessary to adjust for
background concentration levels to ensure accurate comparison of samples collected during
multiple days. Ideally, ambient concentration of all the pollutants should be measured
simultaneous at an urban background locations. However, it is typically not feasible to directly
measure background concentrations because of the requirement for additional equipment and a
dedicated technician for field operation. In absence of these resources, researchers often seek
alternative methods to estimate background concentrations (Brantley et al., 2014).
The most popular approach is to measure pollutants at an ambient site before and after each
sampling session. While promising, this approach does not accurately capture temporal variations
in background concentrations caused by atmospheric mixing or photochemical reactions
throughout the day. Another approach is to design an optimal sampling route which includes
multiple locations representative of as an ambient site (Van Poppel, Peters, & Bleux, 2013).
However, when designing an optimal sampling route is not permissible, or comparisons are being
made between two synchronized measurements (which is the case of this study), additional
strategies are needed. An alternative approach is to assume that the baseline of the measurement
(e.g. low percentile or moving average) is representative of background ambient concentrations.
In this study, background ambient concentrations were estimated using three algorithms:
rolling minimum, rolling 25
th
percentile, and spline of minimums. Previous studies have used and
documented the performance of these algorithms, and the spline of minimum showed the most
promising result for background estimation (for a comprehensive review, see Brantley et al. 2014).
The rolling minimums and the rolling 25
th
percentile were computed at 30 minutes intervals. The
spline of minimum was calculated using three steps: 1) apply a rolling 30s mean to smooth the
162

measurements; 2) identify the minimum concentration within discrete 10 min windows; and 3)
apply a smoothing algorithm (loess, thin plate spline, etc.) through the minimum concentrations.
These three algorithms were compared using time-series plots.
Figure S2 shows the results of the three algorithms applied to the PM2.5 measurements.
The red dots represent the hourly PM2.5 measurements obtained from the Los Angeles AQMD
(Air Quality Management District) site located near Downtown Los Angeles. The three algorithms
produced similar results, but the spline of minimum yielded the highest correlation with the
ambient PM2.5 measurement at the AQMD site (Pearson’s correlation test = 0.57, α = 0.05),
compared to the rolling minimum (Pearson’s correlation test = 0.55, α = 0.05) or the rolling 25th
percentile (Pearson’s correlation test = 0.56, α = 0.05).
Figure S3 and Figure S4 show a comparison of the three algorithms for BC and UFP
number concentrations, respectively. Because the AQMD site provides no ambient measurements
for these pollutants, the spline of minimum method was chosen as the most conservative
background estimation. The estimated background concentrations were then subtracted from the
measurements to obtain background-adjusted measurements.

163

Figure S2. Example of background estimation algorithms for PM2.5 on Nov. 11, 2015: (a) rolling minimum;
(b) rolling 25
th
percentile; and (c) spline of minimums. The red dot indicates ambient background
concentration measured at the Downtown Los Angeles AQMD site.
164

Figure S3. Example of background estimation algorithms for BC on Nov. 11, 2015: (a) rolling minimum; (b)
rolling 25
th
percentile; and (c) spline of minimums.

165

Figure S4. Example of background estimation algorithms for UFP on Nov. 11, 2015: (a) rolling minimum;
(b) rolling 25
th
percentile; and (c) spline of minimums.

166

S4. Checking for temporal autocorrelation
We checked the presence of autocorrelation with the pollutant measurements by plotting
autocorrelation function (ACF) and applying Durbin-Watson test. Prior to the autocorrelation
check, the measurement data were first aggregated with a 1-min sample interval to reduce noise
and to make time series appropriate for an autoregressive, AR(1) model.
To check for autocorrelation, the Durbin-Watson test was performed on the residuals of a
linear regression of the data values on time. Table S1 shows the resulting D-W statistic, 𝑑 ̂
, for each
pollutant during Non-CicLAvia and CicLAvia. The D-W table value for n > 200 and α = 0.05 with
2 degrees of freedom is 1.748 for the lower limit. None of the D-W statistics across the pollutants
exceeds this lower limit, indicating significant autocorrelation in the measurements. Aggregating
the data with longer time intervals did not eliminate the presence of autocorrelation; therefore, we
proceed to the next step and developed an autoregressive model to directly account for the
autocorrelation.

Table S1. D-W statistic 𝒅 ̂
of a linear regression model
Non-CicLAvia CicLAvia
N 𝑑 ̂
N 𝑑 ̂

PM 10 3381 0.83 1789 0.79
PM 2.5 3212 0.61 1729 0.58
UFP 3017 0.67 1717 0.94
BC 3444 1.12 1823 0.94

To determine the feasibility of applying an AR(1) model, an autocorrelation function
(ACF) was computed for the first 40 lags for each pollutant measurement. Figure S5 shows the
ACF for each pollutant with a 1-min sample interval during CicLAvia and Non-CicLAvia. The
ACF shows a decay representative of an AR(1) series and justifies the use of the AR(1) model.

167

After confirming the autoregressive process in the time series, AR(1) model was developed by
regressing the present measurement on a previous measurement using the following equation.

(1)

where yt is a present measurement recorded at time t; yt-1 is a previous measurement recorded at
time t-1; ϕ denotes constant multiple; μ is the mean of the measurements, and ϵ denotes the random
error. After developing the autoregressive model, the D-W test was applied to the residuals of the
AR(1) model. Table S2 is the resulting D-W statistic, 𝑑 ̂
, for each pollutant based on the residuals
of the AR(1) model. The D-W table value for n > 200 and α = 0.05 with 2 degrees of freedom is
1.789 for the upper limit. All of the D-W statistics across the pollutants are higher than this critical
value, indicating that the residual series are not significantly autocorrelated.
168

Figure S5. Autocorrelation function (ACF) for the first 40 lags. (a) PM 10, (b) PM 2.5, (c) UFP, (d) BC

169

Table S2. D-W statistic 𝒅 ̂
of a first-order autoregressive model, AR(1)
Non-CicLAvia CicLAvia
N 𝑑 ̂
N 𝑑 ̂

PM 10 3381 2.32 1789 2.27
PM 2.5 3212 2.31 1729 2.28
UFP 3017 2.21 1717 2.08
BC 3444 2.24 1823 2.25

The autocorrelation in the measurement does not affect mean values but only the statistical
tests which determine standard errors. To account for the underestimated standard error, we
adopted the approach proposed by Zwiers and von Storch (1995) and applied a modified t-test to
compare the mean values of the measurements. For the two time series, x and y, each with sample
size n and m, the modified t-test statistic is computed as follows:

(2)

The denominator of the modified test statistic has a pooled estimate of the standard variance, s
2
:

(3)

The denominator also includes “equivalent sample size”, ne and me, each corresponding to the
sample size n and m, effectively reduces the original sample size to obtain an improved estimate
of the standard error.

(4)

The equivalent sample size, ne and me are computed from a pooled estimate of the lag-1
correlation coefficient, r1:
170

(5)

The basic idea of the modified test statistic is that the underestimated standard error is corrected
by adjusting the sample size through a first-order autoregressive process. In order words, the above
procedures produce less biased noise-to-signal ratio by adjusting the underestimated variance in
the measurement due to correlation.

171

S5. Implementation of the modified t-test using R programming language
With high-resolution time series such as pollutant sample (typically 10 s time resolution),
neighboring values (in time) are more similar than distant values, and estimates of standard error
underestimated sampling variability due to inflation of the number of independent samples and
producing greater chances of statistically significant p values. To analyze time series data with
significant autocorrelation, Zwiers and von Storch (1995) developed a modified t-test which is an
extension of a traditional t-test. The modified t-test corrects for the underestimated standard error
in autocorrelated time series by reducing the sample size through a first-order autoregressive
process. Therefore, the modified t-test produces less biased noise-to-signal ratio by adjusting the
underestimated variance in the measurement due to correlation. Below is the logic model of the
modified t-test (Figure S6). The complete source code can be found at: https://github.com/andy-
hong/modTtest
172

Figure S6. Logic model of the modified t-test for coding implementation

173

S6. Effects of different buffer sizes on selected pollutants

Figure S7. Comparison of measured pollutants using different buffer distances for the Culver City CicLAvia
174

Figure S8. Comparison of measured pollutants using different buffer distances for the DTLA CicLAvia

175

S7. Summary of meteorological values between non-CicLAvia and CicLAvia Sundays

Table S3. Daily average values of selected weather parameters during the field measurement (9am-5pm)
Location Variables Non-CicLAvia CicLAvia p
Pasadena
Temperature (C) 22 22 1.00
Relative Humidity (%) 52 61 0.25
Wind Speed (m/s) 2.5 2.9 0.30
Wind Direction (degree) 316 272 0.30
Culver City
Temperature (C) 23 22 0.03 **
Relative Humidity (%) 74 64 <0.01 ***
Wind Speed (m/s) 2.3 2.3 0.60
Wind Direction (degree) 258 243 0.17
DTLA
Temperature (C) 25 24 0.35
Relative Humidity (%) 25 68 <0.01 ***
Wind Speed (m/s) 0.9 1.1 0.31
Wind Direction (degree) 212 228 0.37
Note: there was no precipitation during the sampling period. The hourly average wind direction was calculated using
vector-averaging of orthogonal components (u and v) of wind direction. Statistical significance (Wilcoxon rank-sum
test): *p <0.10; **p <0.05; ***p <0.01

176

S8. Summary of traffic and pollutant measurements (both adjusted and unadjusted)
Table S4. Changes in daily average values of traffic and measured pollutants
Location Type Variable
Unadjusted Background-adjusted
NC (mean) C (mean) %  p NC (mean) C (mean) %  p
Pasadena
MMP Traffic speed (km/h) 13.81 10.53 -24 0.027 ** - - - -
PM10 (μg/m
3
) 22.79 52.02 128 <0.001 *** 5.44 6.61 21 <0.01 ***
PM2.5 (μg/m
3
) 14.21 43.15 204 <0.001 *** 1.39 3.42 145 <0.01 ***
BC (μg/m
3
) 0.54 1.2 121 <0.001 *** 0.51 1.15 124 <0.01 ***
UFP (#/cm
3
) 16,333 15,810 -3 0.015 ** 3,702 5,906 60 <0.01 ***
PAH (ng/m
3
) 6.37 9.45 48 <0.001 *** 5.92 8.77 48 <0.01 ***
Ozone (ppb) 60.04 75.79 26 <0.001 *** 15.79 17.36 10 <0.01 ***
Walk PM10 (μg/m
3
) 22.38 74.16 231 <0.001 *** 8.6 8.46 -2 <0.001 ***
UFP (#/cm
3
) 52136 36456 -30 0.035 ** 28,119 17,778 -37 0.389
Culver
City
MMP Traffic volume (veh #/h) 402 402 0 0.160 - - - -
Traffic speed (km/h) 14 13 -3 0.805 - - - -
PM10 (μg/m
3
) 27.86 20.83 -25 <0.001 *** 7.85 5.99 -24 <0.01 ***
PM2.5 (μg/m
3
) 16.42 13.01 -21 <0.001 *** 2.40 1.96 -18 <0.01 ***
BC (μg/m
3
) 0.82 0.65 -22 0.072 * 0.79 0.61 -23 <0.01 ***
UFP (#/cm
3
) 6,202 7,141 15 0.002 *** 3,241 3,720 15 <0.01 ***
PAH (ng/m
3
) 6.18 10.58 71 <0.001 *** 5.52 9.68 75 <0.01 ***
Ozone (ppb) 47.32 39.82 -16 0.82 20.31 13.55 -33 0.91
Walk PM10 (μg/m
3
) 26.4 24.41 -8 0.008 *** 4.79 4.33 -10 <0.001 ***
UFP (#/cm
3
) 11,783 12,831 9 0.148 4,342 4,513 4 0.224
DTLA
MMP Traffic volume (veh #/h) 197 259 32 <0.001 *** - - - -
Traffic speed (km/h) 14 12 -12 0.064 * - - - -
PM10 (μg/m
3
) 21.11 21.66 3 <0.001 *** 7.40 6.07 -18 <0.01 ***
PM2.5 (μg/m
3
) 10.44 13.15 26 <0.001 *** 1.89 2.04 8 0.23
BC (μg/m
3
) 1.1 1.18 7 <0.001 *** 1.03 1.23 19 <0.01 ***
UFP (#/cm
3
) 11,143 10,983 -1 0.009 *** 4,283 4,409 3 0.08 *
PAH (ng/m
3
) 9.93 14.88 50 <0.001 *** 8.44 13.15 56 <0.01 ***
Ozone (ppb) 41.12 31.09 -24 0.008 *** 17.85 18.76 5 0.15
Walk PM10 (μg/m
3
) 16.31 24.98 53 <0.001 *** 17.07 6.71 -61 <0.01 ***
UFP (#/cm
3
) 30,487 24,812 -19 0.002 *** 10,502 5,323 -49 <0.05 **
Note: NC and C denote Non-CicLAvia and CicLAvia, respectively. Wilcoxon rank-sum test was used to compare the
means of traffic volume during 9am-5pm, and the modified t-test was used to compare the means of traffic speed and
pollutant concentrations. Statistical significance: *p <0.10; **p <0.05; ***p <0.01

177

S9. Ambient pollutant concentrations monitored at the Los Angeles SCAQMD site

Figure S9. Hourly trends of selected ambient pollutant concentrations at Los Angeles SCAQMD site: (a)
PM 10, (b) PM 2.5, (c) Ozone, (d) NO 2
Note. The shaded area indicates the typical sampling time from 9 am to 5 pm.

178

Table S5. Daily average values of selected ambient pollutant levels (9am-5pm) during the Pasadena event
Variable Time Non-CicLAvia CicLAvia p % 
PM 10 (μg/m
3
)
00:00 - 09:00 21.33 44.93 <0.01 *** 111
09:00 - 17:00 29.28 34.2 <0.10 * 17
17:00 - 24:00 21.68 25.41 <0.10 * 17
PM 2.5 (μg/m
3
)
00:00 - 09:00 10.06 35.78 <0.01 *** 256
09:00 - 17:00 13.81 27.88 <0.01 *** 102
17:00 - 24:00 9.14 11.29 0.22 23
O 3 (ppb)
00:00 - 09:00 28.72 29.89 0.96 4
09:00 - 17:00 51.44 64.88 <0.05 ** 26
17:00 - 24:00 33.86 45.57 <0.05 ** 35
NO 2 (ppb)
00:00 - 09:00 8.72 10.78 <0.10 * 24
09:00 - 17:00 6.94 9.38 <0.01 *** 35
17:00 - 24:00 9.64 6.71 0.14 -30
Note. Statistical significance (Wilcoxon rank-sum test): *p <0.10; **p <0.05; ***p <0.01

Table S6. Daily average values of selected ambient pollutant levels (9am-5pm) during the Culver City event
Variable Time
Non-
CicLAvia
CicLAvia p % 
PM 10 (μg/m
3
)
00:00 - 09:00 32.3 29.86 0.26 -8
09:00 - 17:00 41.98 35.86 <0.05 ** -15
17:00 - 24:00 27.71 21.23 <0.01 *** -23
PM 2.5 (μg/m
3
)
00:00 - 09:00 20.89 16.67 <0.05 ** -20
09:00 - 17:00 24.5 20.25 0.19 -17
17:00 - 24:00 15.14 10.71 0.25 -29
O 3 (ppb)
00:00 - 09:00 11.89 16.22 <0.10 * 36
09:00 - 17:00 54.44 46.25 <0.10 * -15
17:00 - 24:00 26.57 30.57 0.33 15
NO 2 (ppb)
00:00 - 09:00 16.61 14.22 0.15 -14
09:00 - 17:00 8 7.75 0.93 -3
17:00 - 24:00 10.07 7.14 <0.10 * -29
Note. Statistical significance (Wilcoxon rank-sum test): *p <0.10; **p <0.05; ***p <0.01

179

Table S7. Daily average values of selected ambient pollutant levels (9am-5pm) during the DTLA event
Variable Time
Non-
CicLAvia
CicLAvia p % 
PM 10 (μg/m
3
)
00:00 - 09:00 38.5 18.58 <0.01 *** -52
09:00 - 17:00 30.24 20.51 <0.01 *** -32
17:00 - 24:00 34.18 20.4 <0.01 *** -40
PM 2.5 (μg/m
3
)
00:00 - 09:00 13.94 13.67 0.82 -2
09:00 - 17:00 8.25 13.25 0.07 * 61
17:00 - 24:00 14.64 14.5 0.90 -1
O 3 (ppb)
00:00 - 09:00 5.11 28.78 <0.01 *** 463
09:00 - 17:00 53.44 37 <0.01 *** -31
17:00 - 24:00 23.43 25.33 0.74 8
NO 2 (ppb)
00:00 - 09:00 36.17 6.56 <0.01 *** -82
09:00 - 17:00 15.69 6.62 <0.01 *** -58
17:00 - 24:00 26.71 12.17 <0.01 *** -54
Note. Statistical significance (Wilcoxon rank-sum test): *p <0.10; **p <0.05; ***p <0.01

180

S10. Magnitude of changes in total traffic volume
As shown in the Table S8, the Culver City event experienced a total reduction of 40,459
vehicles. This is a 36% reduction (p < 0.001) from a normal Sunday traffic, and a statistically
significant changes in the daily average traffic volume. For the DTLA event, a total of 2,187
vehicles were increased during CicLAvia, resulting in a statistically significant increase (4%, p =
0.03) compared to Non-CicLAvia traffic (Table S9). While most of the traffic counters displayed
a reduction in traffic, it appears that similar magnitude of decrease in traffic occurred on the
CicLAvia route, offsetting the increase experienced on those traffic counters during the DTLA
event.

Table S8. Overall traffic patterns during the Culver City event (9am – 5pm) by traffic count location
Traffic
counter
Non-CicLAvia CicLAvia
Total Mean %  P
Total Mean SD Total Mean SD
1 12160 337.78 49.7 13025 361.81 48 865 24.03 7 <0.05 **
2 21694 602.61 79.75 17512 486.44 80 -4182 -116.17 -19 <0.01 ***
3 24154 670.94 78.33 24255 673.75 50 101 2.81 0 0.50
4 11256 312.67 50.8 12554 348.72 56 1298 36.05 12 <0.01 ***
5 3027 84.08 15.43 4999 138.86 24 1972 54.78 65 <0.01 ***
6 21422 595.06 98.66 0 0 0 -21422 -595.06 -100 <0.01 ***
7 19091 530.31 85.94 0 0 0 -19091 -530.31 -100 <0.01 ***
Total 112804 447.63 207.03 72345 287.08 240 -40459 -160.55 -36 <0.01 ***
Note. Statistical significance (Wilcoxon rank-sum test): *p <0.10; **p <0.05; ***p <0.01

181

Table S9. Overall traffic patterns during the DTLA event (9am – 5pm) by traffic count location
Traffic
counter
Non-CicLAvia CicLAvia
Total Mean %  P
Total Mean SD Total Mean SD
1 6496 180.44 40.15 9283 257.86 54 2787 77.42 43 <0.01 ***
2 6835 189.86 31.15 8815 244.86 34 1980 55 29 <0.01 ***
3 7050 195.83 34.28 9542 265.06 61 2492 69.23 35 <0.01 ***
4 7843 217.86 55 9894 274.83 88 2051 56.97 26 <0.01 ***
5 12662 351.72 60.98 15334 425.94 84 2672 74.22 21 <0.01 ***
6 1704 47.33 8.94 3172 88.11 23 1468 40.78 86 <0.01 ***
7 6420 178.33 28.74 0 0 0 -6420 -178.33
-
100
<0.01 ***
8 4843 134.53 30.51 0 0 0 -4843 -134.53
-
100
<0.01 ***
Total 53853 186.99 88.57 56040 194.58 151 2187 7.59 4 0.03 **
Note. Statistical significance (Wilcoxon rank-sum test): *p <0.10; **p <0.05; ***p <0.01

182

“
This work has provided a scientific and empirical understanding of the links between the
built environment, travel behavior, and air pollution. The first chapter laid a conceptual foundation
of the dissertation by focusing on three important pathways: the built environment and travel
behavior, travel behavior and air pollution exposure, and built environment and air pollution.
In the second chapter, I explored the link between the built environment and physical
activity through experimental research design to isolate the health effects of light rail introduction.
The results indicate that the impacts of the light rail on active travel were greater for persons who
previously had low walking and physical activity, and the results were the opposite for more active
persons. This suggests that people who are sedentary will benefit from having new public transit

183

system near their homes, but people who are already active may not benefit much from having a
transit service nearby.
The third chapter examined the difference in traffic exposure while driving a car and taking
a light rail transit. Using simultaneous measurement of in-car and in-transit exposure, I found that
traffic exposure was significantly altered by ventilation status and, to a lesser extent, traffic
microenvironment and other vehicle-specific factors, such as fan strength, vehicle speed, and
vehicle age. This result suggests that mode shift from car to light rail transit will be particularly
beneficial to those owning an older vehicle with a sub-par ventilation system.
In the fourth chapter, I examined how the car-free street event, referred to as CicLAvia,
affects localized air pollution exposure. Using mobile measurements collected on both event and
non-event days, most pollutants did not show statistically significant changes despite an increase
in traffic congestion generated by the event. The findings suggest that more careful traffic
management will be needed for car-free streets programs to mitigate the potential risks of air
pollution exposure. The results also suggest that continuous technological improvements in
gasoline-powered vehicles now allow minimal air quality impacts from changes in traffic
congestion in Los Angeles.
Overall, the three empirical essays demonstrated that the underlying personal and
behavioral attributes contribute to the differential outcomes with regard to physical activity and
air pollution exposure. The results of the three essays consistently show that the built environment
creates dynamic interactions between travel behavior and air pollution exposure. This interrelated
nature of the relationship between the built environment, travel behavior, and air pollution implies
that the health-related effects are not only affected by the “place” but also by the "people” who
live and work in the neighborhood, calling for a more comprehensive approach to policy and
184

planning. Below, I make three recommendations for future policy and research that are related to
the findings of each chapter.

Research in the area of the built environment and health is still in an early stage. In the
early phase, active living research has dominated much of the theoretical and methodological
development, with particular emphasis on physical activity and obesity (Blackwell, 2009; Jackson,
Dannenberg, & Frumkin, 2013). A separate stream of research has emerged from environmental
health, addressing much broader, ranging from air pollution exposure to mental health. For
example, public health researchers have recently introduced a new concept called “Exposome”—
defined as the totality of human environmental exposure over the life course (Wild, 2005). As
noted by Lioy and Rappaport (2011), this is a powerful concept because it helps identify the
etiology of diseases by understanding a person’s lifetime history of all exposures from both
external sources (e.g. air, water, and food) and internal sources (e.g. inflammation, infection, and
microbiome).
As research moves forward in multiple directions, it is important to acknowledge the
benefits of collaboration among researchers in diverse fields. For example, we have recently seen
a beneficial collaboration between urban planning and physical activity research (Giles-Corti &
Whitzman, 2012; Handy et al., 2002), which has helped bring the active living research to the
mainstream planning and policy agenda, and culminated the development of health impact
assessment (HIA) framework (Dannenberg et al., 2006).
As the development of legal and institutional framework for incorporating health into
mainstream planning and policy is underway, more fine-tuning of health polices and land use

185

regulations at the neighborhood level should appear on the future research agenda. In the second
chapter, I have shown that the goal of transit agency may be slightly out of tune with the goal of
public health department. The traditional transit-oriented development (TOD) strategy that strives
to minimize the distance to transit is somewhat in conflict with the goal of increasing physical
activity through transit. If the goal of transit is not only to increase ridership but also to encourage
physical activity for both sedentary and active population, transit agencies will need to reconsider
the definition of transit-oriented development (TOD) and the related zoning regulations (e.g.
density bonus around station area). Studies are increasingly showing that people may be indifferent
about half-mile or one-mile in terms of their willingness to walk or bike to transit (Guerra, Cervero,
& Tischler, 2012; Ker & Ginn, 2003), but the difference between half mile and one mile for
maintaining daily dose of physical activity could be substantial in a long term (Langlois, Wasfi,
Ross, & El-Geneidy, 2016). Therefore, the goal of future policy and research would be to find the
optimal distance between home and transit station, which will satisfy the dual goal of increasing
ridership and meeting the daily recommended levels of physical activity (i.e. 20 minutes of
physical activity per day). This will require evidence-based approach, rather than relying on
normative arguments, to enable fine-tuning of goals and priorities among different agencies to
achieve satisfying changes in the built environment and health relationship.
Future policy and research should also focus on developing a common framework to bring
activity-based transportation research and exposure science together. In the fourth chapter, I
demonstrated that the impact of road closure on pollutant concentrations differed depending on
local traffic conditions. This suggests that spatial variation of pollutant concentration will likely
be modified by activity patterns of travelers. Modeling microscopic travel behavior and estimating
subsequent exposure patterns is a challenging task, and requires cross-disciplinary collaboration
186

to achieve meaningful outcomes. For example, activity-based approach currently being developed
in transportation modelers (Pendyala, 2009; Pinjari & Bhat, 2011) has wider implications for
environmental health research because it helps determine daily exposure profile of one person that
considers both static and dynamic environmental exposure (Beckx et al., 2009; Buonanno, Stabile,
& Morawska, 2014; Steinle et al., 2013). Exposure scientists have begun to apply this activity-
based approach to exposure modeling and achieved some notable success (Dons, Poppel, Kochan,
Wets, & Int, 2014; Hatzopoulou & Miller, 2010). However, to make the model useful for robust
policy and planning analysis, microscopic human behavioral patterns must be captured and
quantified. With the help of new mobile sensing technology, it will be possible to quantify the
differences in behavior and to develop a subsequent model to accurately estimate exposure that
occur on the microscopic level (de Nazelle et al., 2013; Nieuwenhuijsen, 2016). Combined efforts
in activity-based modeling and ubiquitous sensing technology will likely enable development of
new tools and methods to advance the built environment and health research.
Another area of research that merits continuous attention concerns how relationship
between the built environment and health affects population differently. In the third chapter, I have
demonstrated that mode shift from car to light rail will be particularly beneficial to those owning
an older vehicle. This further justifies targeting policy inventions to poor families experiencing
social disadvantage in order to reduce or eliminate disproportionate burden of environmental
pollution. Even with advanced technology and evaluation tools, it is important to recognize
environmental injustice and health disparities that occur among low-income and minority
neighborhoods. One example of institutional efforts to address chronic environmental justice
issues can be drawn from California. California Environment Protection Agency (CalEPA) is
leading the development of the California Communities Environmental Health Screening Tool

187

(CalEnviroScreen) – a new tool that allows researchers to identify communities burdened by
multiple sources of pollution and face cumulative health and socioeconomic vulnerabilities
(OEHHA, 2014). This tool was created in response to the Environmental Justice Action Plan,
which called for the development of guidance on cumulative impacts to reduce environmental
pollution in communities that are most burdened (CalEPA, 2004). Despite some of the flaws in
the tool, I believe this is a step in the right direction in terms of addressing environmental justice
and health disparities that stem from underlying social disadvantage and marginalization.
While new directions in urban planning and active living research has motivated renewed
interests in public health roots in urban planning, much work remains to be done in order to develop
a comprehensive approach to understand how policy and planning influence the built environment
and ultimately health outcomes. My hope and expectation is that this dissertation will help advance
our understanding of the relationship between the built environment and health, and make positive
impacts on health and wellbeing through effective, and yet, equitable policy and planning
interventions.

188

Aarts, H., Paulussen, T., & Schaalma, H. (1997). Physical exercise habit: on the conceptualization and
formation of habitual health behaviours. Health Education Research, 12(3), 363–374.
Adams, H. S., Nieuwenhuijsen, M. J., Colvile, R. N., McMullen, M. A. S., & Khandelwal, P. (2001). Fine
particle (PM2.5) personal exposure levels in transport microenvironments, London, UK. The
Science of the Total Environment, 279(1-3), 29–44.
Advocacy Advance. (2014). First Mile, Last Mile: How Federal Transit funds can improve access to
transit for people who walk and bike.
Ajzen, I. (1991). The Theory of Planned Behavior. Organizational Behavior and Human Decision
Processes, 50, 179–211.
Ajzen, I. (2002). Residual Effects of Past on Later Behavior: Habituation and Reasoned Action
Perspectives. Personality and Social Psychology Review, 6, 107.
http://doi.org/10.1207/S15327957PSPR0602
Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood
Cliffs, NJ: Prentice-Hall.
Albalate, D., & Bel, G. (2009). What local policy makers should know about Urban road charging:
Lessons from worldwide experience. Public Administration Review, 69(5), 962–975.
http://doi.org/10.1111/j.1540-6210.2009.02045.x
Badland, H., & Schofield, G. (2005). Transport, urban design, and physical activity: an evidence-based
update. Transportation Research Part D: Transport and Environment, 10(3), 177–196.
http://doi.org/10.1016/j.trd.2004.12.001
Badoe, D., & Miller, E. (2000). Transportation–land-use interaction: empirical findings in North America,
and their implications for modeling. Transportation Research Part D: Transport and Environment,
5(4), 235–263.
Bamberg, S., Ajzen, I., & Schmidt, P. (2003). Choice of travel mode in the theory of planned behavior:
The roles of past behavior, habit, and reasoned action. Basic and Applied Social Psychology, 25(3),
175–187. http://doi.org/10.1016/j.tranpol.2010.08.006
Bassett, D. R., Pucher, J., Buehler, R., Thompson, D. L., & Crouter, S. E. (2008). Walking, cycling, and

189
obesity rates in Europe, North America, and Australia. Journal of Physical Activity & Health, 5(6),
795–814.
Bauman, A. E., Sallis, J. F., Dzewaltowski, D. a, & Owen, N. (2002). Toward a better understanding of
the influences on physical activity. American Journal of Preventive Medicine, 23(2), 5–14.
http://doi.org/10.1016/S0749-3797(02)00469-5
Beckx, C., Int Panis, L., Arentze, T., Janssens, D., Torfs, R., Broekx, S., & Wets, G. (2009). A dynamic
activity-based population modelling approach to evaluate exposure to air pollution: Methods and
application to a Dutch urban area. Environmental Impact Assessment Review, 29(3), 179–185.
http://doi.org/10.1016/j.eiar.2008.10.001
Ben-Akiva, M. E., & Bierlaire, M. (1999). Discrete choice methods and their applications to short term
travel decisions. In R. W. Hall (Ed.), Handbook of transportation science (pp. 5–34). New York:
Springer Science+Business Media.
Ben-Akiva, M. E., & Lerman, S. R. (1985). Discrete choice analysis: theory and application to predict
travel demand. Cambridge, MA: MIT Press.
Bentler, P. M., & Speckart, G. (1979). Models of Attitude-Behavior Relations. Psychological Review,
86(5), 452–464.
Berkowicz, R., Ketzel, M., Jensen, S., Hvidberg, M., & Raaschounielsen, O. (2008). Evaluation and
application of OSPM for traffic pollution assessment for a large number of street locations.
Environmental Modelling & Software, 23(3), 296–303. http://doi.org/10.1016/j.envsoft.2007.04.007
Besser, L., & Dannenberg, A. L. (2005). Walking to public transit: steps to help meet physical activity
recommendations. American Journal of Preventive Medicine, 29(4).
http://doi.org/10.1016/j.ampre.2005.06.010
Bigazzi, A. Y., & Figliozzi, M. a. (2014). Review of Urban Bicyclists’ Intake and Uptake of Traffic-
Related Air Pollution. Transport Reviews, 34(2), 221–245.
http://doi.org/10.1080/01441647.2014.897772
Bijnen, F. (1998). Age, period, and cohort effects on physical activity among elderly men during 10 years
of follow-up: the Zutphen Elderly Study. The Journals of Gerontology Series A: Biological Sciences
and Medical Sciences, 53(3), 235–241.
Blackwell, A. G. (2009). Active Living Research and the movement for healthy communities. American
Journal of Preventive Medicine, 36(2 Suppl), S50–2. http://doi.org/10.1016/j.amepre.2008.10.004
Bland, J. M., & Altman, D. G. (1994). Statistical notes: Regression towards the mean. BMJ: British
Medical Journal, 308, 1499. http://doi.org/10.1136/bmj.308.6942.1499
Boarnet, M. G., & Greenwald, M. J. (2000). Land Use, Urban Design, and Nonwork Travel.
Transportation Research Record: Journal of the Transportation Research Board, 1722, 27–37.
190

Boarnet, M. G., & Sarmiento, S. (1998). Can Land-use Policy Really Affect Travel Behaviour? A Study
of the Link between Non-work Travel and Land-use Characteristics. Urban Studies, 35(7), 1155–
1169. http://doi.org/10.1080/0042098984538
Borrego, C., Tchepel, O., Costa, A., Martins, H., Ferrieira, J., & Miranda, A. (2006). Traffic-related
particulate air pollution exposure in urban areas. Atmospheric Environment, 40(37), 7205–7214.
Bors, P., Dessauer, M., Bell, R., Wilkerson, R., Lee, J., & Strunk, S. L. (2009). The Active Living by
Design national program: community initiatives and lessons learned. American Journal of
Preventive Medicine, 37(6 Suppl 2), S313–21. http://doi.org/10.1016/j.amepre.2009.09.027
Bowman, J., & Ben-Akiva, M. E. (2000). Activity-based disaggregate travel demand model system with
activity schedules. Transportation Research Part A: Policy and Practice, 35, 1–28.
Boyce, D. (2002). Is the Sequential Travel Forecasting Paradigm Counterproductive? Journal of Urban
Planning and Development, 128(4), 169–183.
Brantley, H. L., Hagler, G. S. W., Kimbrough, E. S., Williams, R. W., Mukerjee, S., & Neas, L. M.
(2014). Mobile air monitoring data-processing strategies and effects on spatial air pollution trends.
Atmospheric Measurement Techniques, 7(7), 2169–2183. http://doi.org/10.5194/amt-7-2169-2014
Brauer, M., Hoek, G., van Vliet, P., Meliefste, K., Fischer, P., Gehring, U., … Brunekreef, B. (2003).
Estimating long-term average particulate air pollution concentrations: application of traffic
indicators and geographic information systems. Epidemiology, 14(2), 228–39.
Briggs, D. J. (2000). A regression-based method for mapping traffic-related air pollution: application and
testing in four contrasting urban environments. The Science of the Total Environment, 253(1-3),
151–167.
Briggs, D. J., de Hoogh, K., Morris, C., & Gulliver, J. (2008). Effects of travel mode on exposures to
particulate air pollution. Environment International, 34(1), 12–22.
Brown, B. B., & Werner, C. M. (2007). A new rail stop: tracking moderate physical activity bouts and
ridership. American Journal of Preventive Medicine, 33(4), 306–9.
http://doi.org/10.1016/j.amepre.2007.06.002
Brown, B. B., & Werner, C. M. (2008). Before and After a New Light Rail Stop: Resident Attitudes,
Travel Behavior, and Obesity. Journal of the American Planning Association, 75(1), 5–12.
http://doi.org/10.1080/01944360802458013
Brownson, R. C., Boehmer, T. K., & Luke, D. A. (2005). Declining rates of physical activity in the
United States: What are the contributors? Annual Review of Public Health, 26, 421–43.
Brownson, R. C., Hoehner, C. M., Day, K., Forsyth, A., & Sallis, J. F. (2009). Measuring the built
environment for physical activity: state of the science. American Journal of Preventive Medicine,
36(4 Suppl), S99–123.e12.

191
Buonanno, G., Stabile, L., & Morawska, L. (2014). Personal exposure to ultrafine particles: The influence
of time-activity patterns. Science of the Total Environment, 468-469, 903–907.
http://doi.org/10.1016/j.scitotenv.2013.09.016
Bussel, J. B., Leviton, L. C., & Orleans, C. T. (2009). Active living by design: perspectives from the
Robert Wood Johnson Foundation. American Journal of Preventive Medicine, 37(6 Suppl 2), S309–
12. http://doi.org/10.1016/j.amepre.2009.09.019
Cairns, S., Sloman, L., Newson, C., Anable, J., Kirkbridge, A., & Goodwin, P. (2008). Smarter choices:
Assessing the potential to achieve traffic reductions using “soft measures.” Transport Reviews,
28(5), 593–618. http://doi.org/10.1080/01441640801892504
CalEPA. (2004). Environmental Justice Action Plan. Sacramento, CA.
Cao, X. (Jason), Handy, S. L., & Mokhtarian, P. L. (2006). The Influences of the Built Environment and
Residential Self-Selection on Pedestrian Behavior: Evidence from Austin, TX. Transportation, 33,
1–20.
Cao, X. (Jason), Mokhtarian, P. L., & Handy, S. L. (2009). Examining the Impacts of Residential Self-
selection on Travel Behaviour: A Focus on Empirical Findings. Transport Reviews, 29(3), 359–395.
Carlson, J. A., Sallis, J. F., Conway, T. L., Saelens, B. E., Frank, L. D., Kerr, J., … King, A. C. (2012).
Interactions between psychosocial and built environment factors in explaining older adults’ physical
activity. Preventive Medicine, 54(1), 68–73. http://doi.org/10.1016/j.ypmed.2011.10.004
Cartenì, A., Cascetta, F., & Campana, S. (2015). Underground and ground-level particulate matter
concentrations in an Italian metro system. Atmospheric Environment, 101, 328–337.
http://doi.org/10.1016/j.atmosenv.2014.11.030
Carto DB. (n.d.). Retrieved April 6, 2016, from http://cartodb.com/
Cervero, R., & Kockelman, K. (1997). Traffic demand and the 3Ds: Density, diversity, and design.
Transportation Research Part D: Transport and Environment, 2(3), 199–219.
Chaix, B., Kestens, Y., Perchoux, C., Karusisi, N., Merlo, J., & Labadi, K. (2012). An interactive
mapping tool to assess individual mobility patterns in neighborhood studies. American Journal of
Preventive Medicine, 43(4), 440–50. http://doi.org/10.1016/j.amepre.2012.06.026
Chakrabarti, B., Fine, P. M., Delfino, R., & Sioutas, C. (2004). Performance evaluation of the active-flow
personal DataRAM PM2.5 mass monitor (Thermo Anderson pDR-1200) designed for continuous
personal exposure measurements. Atmospheric Environment, 38(20), 3329–3340.
http://doi.org/10.1016/j.atmosenv.2004.03.007
Chan, L. ., Lau, W. ., Lee, S. ., & Chan, C. . (2002). Commuter exposure to particulate matter in public
transportation modes in Hong Kong. Atmospheric Environment, 36(21), 3363–3373.
http://doi.org/10.1016/S1352-2310(02)00318-7
192

Chan, L. Y., Lau, W. L., Zou, S. C., Cao, Z. X., & Lai, S. C. (2002). Exposure level of carbon monoxide
and respirable suspended particulate in public transportation modes while commuting in urban area
of Guangzhou , China, 36, 5831–5840.
CicLAvia. (n.d.). Our history. Retrieved March 2, 2016, from http://www.ciclavia.org/history
City of Pasadena. (2015). Transportation Data Management System. Retrieved March 8, 2016, from
http://pasadena.ms2soft.com/tcds/tsearch.asp?loc=Pasadena&mod=TCDS
Colvile, R. ., Hutchinson, E. ., Mindell, J. ., & Warren, R. . (2001). The transport sector as a source of air
pollution. Atmospheric Environment, 35(9), 1537–1565. http://doi.org/10.1016/S1352-
2310(00)00551-3
Corburn, J. (2004). Confronting the challenges in reconnecting urban planning and public health.
American Journal of Public Health, 94(4), 541–6.
Corburn, J., Osleeb, J., & Porter, M. (2006). Urban asthma and the neighbourhood environment in New
York City. Health and Place, 12(2), 167–179. http://doi.org/10.1016/j.healthplace.2004.11.002
Crane, R., & Crepeau, R. (1998). Does neighborhood design influence travel?: A behavioral analysis of
travel diary and GIS data. Transportation Research Part D: Transport and Environment, 3(4), 225–
238.
Cummins, S., Curtis, S., Diez-Roux, A. V, & Macintyre, S. (2007). Understanding and representing
“place” in health research: a relational approach. Social Science & Medicine (1982), 65(9), 1825–38.
http://doi.org/10.1016/j.socscimed.2007.05.036
Dannenberg, A. L., Bhatia, R., Cole, B. L., Dora, C., Fielding, J. E., Kraft, K., … Tilson, H. H. (2006).
Growing the field of health impact assessment in the United States: an agenda for research and
practice. American Journal of Public Health, 96(2), 262–70.
http://doi.org/10.2105/AJPH.2005.069880
Dannenberg, A. L., Jackson, R. J., Frumkin, H., Schieber, R. A., Pratt, M., Kochtitzky, C., & Tilson, H.
H. (2003). The impact of community design and land-use choices on public health: a scientific
research agenda. American Journal of Public Health, 93(9), 1500. Retrieved from
http://ajph.aphapublications.org/cgi/content/abstract/93/9/1500
de Hartog, J. J., Boogaard, H., Nijland, H., & Hoek, G. (2010). Do the health benefits of cycling outweigh
the risks? Environmental Health Perspectives, 118(8), 1109–16.
de Nazelle, A. J., Fruin, S. A., Westerdahl, D., Martinez, D., Ripoll, A., Kubesch, N., & Nieuwenhuijsen,
M. (2012). A travel mode comparison of commuters’ exposures to air pollutants in Barcelona.
Atmospheric Environment, 59, 151–159. http://doi.org/10.1016/j.atmosenv.2012.05.013
de Nazelle, A. J., Nieuwenhuijsen, M. J., Antó, J. M., Brauer, M., Briggs, D., Braun-Fahrlander, C., …
Lebret, E. (2011). Improving health through policies that promote active travel: a review of

193
evidence to support integrated health impact assessment. Environment international (Vol. 37).
de Nazelle, A. J., & Rodriguez, D. A. (2009). Tradeoffs in incremental changes towards pedestrian-
friendly environments: Physical activity and pollution exposure. Transportation Research Part D:
Transport and Environment, 14(4), 255–263.
de Nazelle, A. J., Rodriguez, D. A., & Crawford-Brown, D. (2009). The built environment and health:
impacts of pedestrian-friendly designs on air pollution exposure. Science of the Total Environment,
407(8), 2525–35.
de Nazelle, A. J., Seto, E., Donaire-Gonzalez, D., Mendez, M., Matamala, J., Nieuwenhuijsen, M. J., &
Jerrett, M. (2013). Improving estimates of air pollution exposure through ubiquitous sensing
technologies. Environmental Pollution, 176, 92–99. http://doi.org/10.1016/j.envpol.2012.12.032
Domencich, T., & McFadden, D. (1975). Urban Travel Demand. Amsterdam: North-Holland.
Dons, E., Int Panis, L., Van Poppel, M., Theunis, J., Willems, H., Torfs, R., & Wets, G. (2011). Impact of
time–activity patterns on personal exposure to black carbon. Atmospheric Environment, 45(21),
3594–3602. http://doi.org/doi:10.1016/j.atmosenv.2011.03.064
Dons, E., Poppel, M. Van, Kochan, B., Wets, G., & Int, L. (2014). Implementation and validation of a
modeling framework to assess personal exposure to black carbon. Environment International, 62,
64–71. http://doi.org/10.1016/j.envint.2013.10.003
EPA. (2000). Meteorological Monitoring Guidance for Regulatory Modeling Applications. Research
Triangle Park, NC. Retrieved from http://www.epa.gov/scram001/guidance/met/mmgrma.pdf
ESRI. (2011). ArcGIS Desktop: Release 10. Redlands, CA: Environmental Systems Research Institute.
Evans, G. W., & Kantrowitz, E. (2002). Socioeconomic Status and Health: The Potential Role of
Environmental Risk Exposure. Annual Review of Public Health, 23(1), 303–331.
http://doi.org/10.1146/annurev.publhealth.23.112001.112349
Evans, R. G., Bare, M. L., & Marmor, T. R. (Eds.). (1994). Why are some people healthy and others not?:
The determinants of health populations. New York, NY: Aldine de Gruyter.
Evenson, K. (2006). Girls’ perception of physical environmental factors and transportation: reliability and
association with physical activity and active transport to school. International Journal of Behavioral
Nutrition and Physical Activity, 16, 1–16. http://doi.org/10.1186/1479-Received
Evenson, K. R., Murray, D. M., Birnbaum, A. S., & Cohen, D. A. (2010). Examination of perceived
neighborhood characteristics and transportation on changes in physical activity and sedentary
behavior: The Trial of Activity in Adolescent Girls. Health & Place, 16(5), 977–85.
http://doi.org/10.1016/j.healthplace.2010.06.002
Ewing, R. (1999). Pedestrian-and transit-friendly design: a primer for smart growth. Washington, DC.
Ewing, R., Tian, G., Goates, J., Zhang, M., Greenwald, M. J., Joyce, A., … Greene, W. (2014). Varying
194

influences of the built environment on household travel in 15 diverse regions of the United States.
Urban Studies, 1–9. http://doi.org/10.1177/0042098014560991
Forsyth, A., Oakes, J. M., Schmitz, K. H., & Hearst, M. (2007). Does Residential Density Increase
Walking and Other Physical Activity? Urban Studies, 44(4), 679–697.
http://doi.org/10.1080/00420980601184729
Frank, L. D., Engelke, P., & Schmid, T. (2003). Health and Community Design: The Impacts of the Built
Environment on Physical Activity. Washington, D.C.: Island Press.
Frank, L. D., Jr, B. S., & Bachman, W. (2000). Linking land use with household vehicle emissions in the
central puget sound: methodological framework and findings. Transportation Research Part D,
5(3), 173–196.
Frank, L. D., Sallis, J. F., Conway, T. L., Chapman, J. E., Saelens, B. E., & Bachman, W. (2006). Many
pathways from land use to health: associations between neighborhood walkability and active
transportation, body mass index, and air quality. Journal of the American Planning Association,
72(1), 75–87.
Fredricks, A. J., & Dossett, D. L. (1983). Attitude-behavior relations: A comparison of the Fishbein-
Ajzen and the Bentler-Speckart models. Journal of Personality and Social Psychology, 45(3), 501–
512. http://doi.org/10.1037/0022-3514.45.3.501
Freedson, P. S., Melanson, E., & Sirard, J. (1998). Calibration of the Computer Science and Applications,
Inc. accelerometer. Medicine and Science in Sports and Exercise, 30(5), 777–81.
Freeland, A. L., Banerjee, S. N., Dannenberg, A. L., & Wendel, A. M. (2013). Walking associated with
public transit: moving toward increased physical activity in the United States. American Journal of
Public Health, 103(3), 536–42. http://doi.org/10.2105/AJPH.2012.300912
Friedman, M. S., Powell, K. E., Hutwagner, L., Graham, L. M., & Teague, W. G. (2001). Impact of
Changes in Transportation and Commuting Behaviors During the 1996 Summer Olympic Games in
Atlanta on Air Quality and Childhood Asthma. The Journal of the American Medical Association,
285(7), 897–905. http://doi.org/10.1001/jama.285.7.897
Fruin, S. A., Hudda, N., Sioutas, C., & Delfino, R. J. (2011). Predictive model for vehicle air exchange
rates based on a large, representative sample. Environmental Science & Technology, 45(8), 3569–
3575. http://doi.org/10.1021/es103897u
Fruin, S. A., Westerdahl, D., Sax, T., Sioutas, C., & Fine, P. M. (2008). Measurements and predictors of
on-road ultrafine particle concentrations and associated pollutants in Los Angeles. Atmospheric
Environment, 42(2), 207–219. http://doi.org/10.1016/j.atmosenv.2007.09.057
Frumkin, H., Frank, L. D., & Jackson, R. J. (2004). Urban Sprawl and Public Health: Designing,
Planning, and Building for Healthy Communities. Washington, D.C.: Island Press.

195
Gardner, B. (2009). Modelling motivation and habit in stable travel mode contexts. Transportation
Research Part F: Traffic Psychology and Behaviour, 12(1), 68–76.
http://doi.org/10.1016/j.trf.2008.08.001
Gärling, T., & Axhausen, K. W. (2003). Introduction: Habitual travel choice. Transportation, 30(1), 1–
11. http://doi.org/10.1023/A:1021230223001
Garshick, E., Laden, F., Hart, J. E., & Caron, A. (2003). Residence near a major road and respiratory
symptoms in U.S. Veterans. Epidemiology, 14(6), 728–36.
Gauderman, W. J., Vora, H., McConnell, R., Berhane, K., Gilliland, F., Thomas, D., … Jerrett, M.
(2007). Effect of exposure to traffic on lung development from 10 to 18 years of age: a cohort study.
The Lancet, 369(9561), 571–577. http://doi.org/10.1016/S0140-6736(07)60037-3
Giles-Corti, B., & Donovan, R. J. (2002). The relative influence of individual, social and physical
environment determinants of physical activity. Social Science & Medicine, 54(12), 1793–812.
Giles-Corti, B., & Whitzman, C. (2012). Active living research: partnerships that count. Health & Place,
18(1), 118–20. http://doi.org/10.1016/j.healthplace.2011.09.010
Gillies, J. a, Gertler, a W., Sagebiel, J. C., & Dippel, W. a. (2001). On-Road Particulate Matter ( PM 2 . 5
and PM 10 ) Emissions in the Sepulveda Tunnel , Los Angeles ,. Environmental Science &
Technology, 35(6), 1054–1063.
Glanz, K., Rimer, B. K., & Viswanath, K. (2008). Health behavior and health education. Health
Education Behavior, 30(2), 7–9.
Gluckman, P. D., Hanson, M. a., Cooper, C., & Thornburg, K. L. (2008). Effect of in utero and early-life
conditions and adult health and disease. The New England Journal of Medicine, 359, 1523–1524;
author reply 1524. http://doi.org/10.1056/NEJMra0708473
Greaves, S. P. (2006). Variability of personal exposure to fine particulates for urban commuters inside
automobiles. Journal of the Transportation Research Board, (1987), 161–170.
Greenwald, M. J., & Boarnet, M. G. (2001). Built Environment as Determinant of Walking Behavior:
Analyzing Nonwork Pedestrian Travel in Portland, Oregon. Transportation Research Record:
Journal of the Transportation Research Board, 1780, 33–42.
Guerra, E., Cervero, R., & Tischler, D. (2012). Half-mile circle: Does it best represent transit station
catchments? Transportation Research Record, 2276(August), 101–109. http://doi.org/10.3141/2276-
12
Gulliver, J., & Briggs, D. (2004). Personal exposure to particulate air pollution in transport
microenvironments. Atmospheric Environment, 38(1), 1–8.
Gulliver, J., & Briggs, D. (2011). STEMS-Air: A simple GIS-based air pollution dispersion model for
city-wide exposure assessment. The Science of the Total Environment, 409(12), 2419–29.
196

http://doi.org/10.1016/j.scitotenv.2011.03.004
Gunier, R. B., Hertz, A., Von Behren, J., & Reynolds, P. (2003). Traffic density in California:
socioeconomic and ethnic differences among potentially exposed children. Journal of Exposure
Analysis and Environmental Epidemiology, 13(3), 240–246. http://doi.org/10.1038/sj.jea.7500276
Hagger, M. (2001). The influence of self-efficacy and past behaviour on the physical activity intentions of
young people. Journal of Sports Sciences, 19(9), 711–725.
Hagler, G. S. W. (2011). Post-processing Method to Reduce Noise while Preserving High Time
Resolution in Aethalometer Real-time Black Carbon Data. Aerosol and Air Quality Research, 11,
539–546. http://doi.org/10.4209/aaqr.2011.05.0055
Hamilton, L. (2012). Regression diagnostics. In Statistics with Stata: Version 12 (pp. 209–228). Boston,
MA: Cengage Learning.
Handy, S. L. (1993). Regional Versus Local Accessibility : Implications for Nonwork Travel.
Transportation Research Record: Journal of the Transportation Research Board, (234), 58–66.
Handy, S. L., Boarnet, M. G., Ewing, R., & Killingsworth, R. E. (2002). How the built environment
affects physical activity: views from urban planning. American Journal of Preventive Medicine,
23(2 Suppl), 64–73.
Handy, S. L., Cao, X. (Jason), & Mokhtarian, P. L. (2006). Self-selection in the relationship between the
built environment and walking: empirical evidence from northern California. Journal of the
American Planning Association, 72(1), 55–74.
Hatzopoulou, M., & Miller, E. J. (2010). Linking an activity-based travel demand model with traffic
emission and dispersion models: Transport’s contribution to air pollution in Toronto. Transportation
Research Part D: Transport and Environment, 15(6), 315–325.
http://doi.org/10.1016/j.trd.2010.03.007
Health Effects Institute. (2010). Traffic-related air pollution: a critical review of the literature on
emissions, exposure, and health effects. Boston, MA. Retrieved from
http://pubs.healtheffects.org/view.php?id=334
Hertel, O., Hvidberg, M., Ketzel, M., Storm, L., & Stausgaard, L. (2008). A proper choice of route
significantly reduces air pollution exposure--a study on bicycle and bus trips in urban streets.
Science of the Total Environment, 389(1), 58–70.
Hoek, G., Brunekreef, B., Goldbohm, S., Fischer, P., & van den Brandt, P. A. (2002). Association
between mortality and indicators of traffic-related air pollution in the Netherlands: a cohort study.
The Lancet, 360(9341), 1203–1209.
Hong, A., Schweitzer, L., Yang, W., & Marr, L. C. (2015). Impact of Temporary Freeway Closure on
Regional Air Quality: A Lesson from Carmageddon in Los Angeles, United States. Environmental

197
Science & Technology, 49(5), 3211–3218. http://doi.org/10.1021/es505185c
Houston, D. (2014). Implications of the modifiable areal unit problem for assessing built environment
correlates of moderate and vigorous physical activity. Applied Geography, 50, 40–47.
http://doi.org/10.1016/j.apgeog.2014.02.008
Houston, D., Boarnet, M., & Spears, S. (2015). Final Report - Evaluating the Benefits of Light Rail
Transit. Sacramento, CA.
Houston, D., Luong, T. T., & Boarnet, M. G. (2014). Tracking daily travel: Assessing discrepancies
between GPS-derived and self-reported travel patterns. Transportation Research Part C: Emerging
Technologies, 48, 97–108. http://doi.org/10.1016/j.trc.2014.08.013
Hudda, N., Eckel, S. P., Knibbs, L. D., Sioutas, C., Delfino, R. J., & Fruin, S. A. (2012). Linking In-
Vehicle Ultrafine Particle Exposures to On-Road Concentrations. Atmospheric Environment
(Oxford, England : 1994), 59, 578–586. http://doi.org/10.1016/j.atmosenv.2012.05.021
Hudda, N., & Fruin, S. A. (2013). Models for Predicting the Ratio of Particulate Pollutant Concentrations
Inside Vehicles to Roadways. Environmental Science & Technology.
http://doi.org/10.1021/es401500c
Hudda, N., Kostenidou, E., Sioutas, C., Delfino, R. J., & Fruin, S. A. (2011). Vehicle and driving
characteristics that influence in-cabin particle number concentrations. Environmental Science &
Technology, 45(20), 8691–7. http://doi.org/10.1021/es202025m
Int Panis, L., de Geus, B., Vandenbulcke, G., Willems, H., Degraeuwe, B., Bleux, N., … Meeusen, R.
(2010). Exposure to particulate matter in traffic: A comparison of cyclists and car passengers.
Atmospheric Environment, 44(19), 2263–2270.
Jackson, R. J. (2003). The impact of the built environment on health: an emerging field. American
Journal of Public Health, 93(9), 1382–4. Retrieved from
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1447976&tool=pmcentrez&rendertype=
abstract
Jackson, R. J., Dannenberg, A. L., & Frumkin, H. (2013). Health and the built environment: 10 years
after. American Journal of Public Health, 103(9), 1542–4.
http://doi.org/10.2105/AJPH.2013.301482
Jerrett, M., Arain, M. a, Kanaroglou, P., Beckerman, B., Crouse, D., Gilbert, N. L., … Finkelstein, M. M.
(2007). Modeling the intraurban variability of ambient traffic pollution in Toronto, Canada. Journal
of Toxicology and Environmental Health. Part A, 70(3-4), 200–12.
http://doi.org/10.1080/15287390600883018
Jerrett, M., Burnett, R., Willis, A., Krewski, D., Goldberg, M., DeLuca, P., & Finkelstein, N. (2003).
Spatial Analysis of the Air Pollution-–Mortality Relationship in the Context of Ecologic
198

Confounders. Journal of Toxicology and Environmental Health, Part A, 66(16-19), 1735–1778.
http://doi.org/10.1080/15287390306438
Jimenez, J., Claiborn, C., Larson, T., Gould, T., Kirchstetter, T. W., & Gundel, L. (2007). Loading Effect
Correction for Real-Time Aethalometer Measurements of Fresh Diesel Soot. Journal of the Air &
Waste Management Association, 57(7), 868–873. http://doi.org/10.3155/1047-3289.57.7.868
Joh, K., Nguyen, M. T., & Boarnet, M. G. (2011). Can Built and Social Environmental Factors Encourage
Walking among Individuals with Negative Walking Attitudes? Journal of Planning Education and
Research, 32(2), 219–236. http://doi.org/10.1177/0739456X11427914
Kam, W., Cheung, K., Daher, N., & Sioutas, C. (2011). Particulate matter (PM) concentrations in
underground and ground-level rail systems of the Los Angeles Metro. Atmospheric Environment,
45(8), 1506–1516. http://doi.org/10.1016/j.atmosenv.2010.12.049
Kanaroglou, P. S., Jerrett, M., Morrison, J., Beckerman, B., Arain, M. A., Gilbert, N. L., & Brook, J. R.
(2005). Establishing an air pollution monitoring network for intra-urban population exposure
assessment: A location-allocation approach. Atmospheric Environment, 39(13), 2399–2409.
http://doi.org/10.1016/j.atmosenv.2004.06.049
Kaur, S., Clark, R., Walsh, P., Arnold, S., Colvile, R., & Nieuwenhuijsen, M. (2006). Exposure
visualisation of ultrafine particle counts in a transport microenvironment. Atmospheric Environment,
40(2), 386–398.
Kaur, S., Nieuwenhuijsen, M., & Colvile, R. (2005a). Pedestrian exposure to air pollution along a major
road in Central London, UK. Atmospheric Environment, 39(38), 7307–7320.
http://doi.org/10.1016/j.atmosenv.2005.09.008
Kaur, S., Nieuwenhuijsen, M., & Colvile, R. (2005b). Personal exposure of street canyon intersection
users to PM, ultrafine particle counts and carbon monoxide in Central London, UK. Atmospheric
Environment, 39(20), 3629–3641.
Kaur, S., Nieuwenhuijsen, M. J., & Colvile, R. N. (2007). Fine particulate matter and carbon monoxide
exposure concentrations in urban street transport microenvironments. Atmospheric Environment,
41(23), 4781–4810. http://doi.org/10.1016/j.atmosenv.2007.02.002
Ker, I., & Ginn, S. (2003). Myths and realities in walkable catchments: the case of walking and transit.
Road and Transport Research, 12(2), 69–79.
Kerr, J., Norman, G. J., Adams, M. a, Ryan, S., Frank, L. D., Sallis, J. F., … Patrick, K. (2010). Do
neighborhood environments moderate the effect of physical activity lifestyle interventions in adults?
Health & Place, 16(5), 903–8. http://doi.org/10.1016/j.healthplace.2010.05.002
Khattak, A., & Rodriguez, D. (2005). Travel behavior in neo-traditional neighborhood developments: A
case study in USA. Transportation Research Part A: Policy and Practice, 39(6), 481–500.

199
Kim, B. M., Teffera, S., & Zeldin, M. D. (2000). Characterization of PM2.5 and PM10 in the South Coast
Air Basin of southern California: Part. 2--Temporal variations. Journal of the Air & Waste
Management Association (1995), 50(12), 2045–2059.
http://doi.org/10.1080/10473289.2000.10464244
King, A. C., Stokols, D., Talen, E., Brassington, G. S., & Killingsworth, R. (2002). Theoretical
approaches to the promotion of physical activity: forging a transdisciplinary paradigm. American
Journal of Preventive Medicine, 23(02), 15–25.
Kirchstetter, T. W., & Novakov, T. (2007). Controlled generation of black carbon particles from a
diffusion flame and applications in evaluating black carbon measurement methods. Atmospheric
Environment, 41(9), 1874–1888. http://doi.org/10.1016/j.atmosenv.2006.10.067
Knibbs, L. D., Cole-Hunter, T., & Morawska, L. (2011). A review of commuter exposure to ultrafine
particles and its health effects. Atmospheric Environment, 45(16), 2611–2622.
Knibbs, L. D., de Dear, R. J., & Atkinson, S. E. (2009). Field study of air change and flow rate in six
automobiles. Indoor Air, 19(4), 303–313. http://doi.org/10.1111/j.1600-0668.2009.00593.x
Knibbs, L. D., de Dear, R. J., & Morawska, L. (2010). Effect of cabin ventilation rate on ultrafine particle
exposure inside automobiles. Environmental Science & Technology, 44(9), 3546–51.
http://doi.org/10.1021/es9038209
Künzli, N., McConnell, R., Bates, D., Bastain, T., Hricko, A., Lurmann, F., … Peters, J. (2003).
Breathless in Los Angeles: The Exhausting Search for Clean Air. American Journal of Public
Health. http://doi.org/10.2105/AJPH.93.9.1494
Kwan, M.-P. (2012). The uncertain geographic context problem. Annals of the Association of American
Geographers, 102(5).
Kwan, M.-P. (2013). Beyond Space (As We Knew It): Toward Temporally Integrated Geographies of
Segregation, Health, and Accessibility. Annals of the Association of American Geographers, 103(5),
1078–1086. http://doi.org/10.1080/00045608.2013.792177
Lachapelle, U., Frank, L. D., Saelens, B. E., Sallis, J. F., & Conway, T. L. (2011). Commuting by public
transit and physical activity: where you live, where you work, and how you get there. Journal of
Physical Activity & Health, 8(Suppl 1), S72–S82.
Lachapelle, U., & Noland, R. B. (2012). Does the commute mode affect the frequency of walking
behavior? The public transit link. Transport Policy, 21, 26–36.
http://doi.org/10.1016/j.tranpol.2012.01.008
Ladabaum, U., Mannalithara, A., Myer, P. A., & Singh, G. (2014). Obesity, abdominal obesity, physical
activity, and caloric intake in US adults: 1988 to 2010. American Journal of Medicine, 127(8), 717–
727.e12. http://doi.org/10.1016/j.amjmed.2014.02.026
200

LADOT. (2013). 24 Hours Traffic Volume. Los Angeles, CA. Retrieved from http://navigatela.lacity.org/
Langlois, M., Wasfi, R. A., Ross, N. A., & El-Geneidy, A. M. (2016). Can transit-oriented developments
help achieve the recommended weekly level of physical activity? Journal of Transport and Health.
http://doi.org/10.1016/j.jth.2016.02.006
Larson, T., Henderson, S. B., & Brauer, M. (2009). Mobile monitoring of particle light absorption
coefficient in an urban area as a basis for land use regression. Environmental Science & Technology,
43(13), 4672–8.
Latoski, S. P., Dunn Jr, W. M., Wagenblast, B., Randall, J., & Walker, M. D. (2003). Managing travel for
planned special events. USDOT. Washington, D.C. Retrieved from
http://tmcpfs.ops.fhwa.dot.gov/cfprojects/uploaded_files/Front
Matter.doc\nhttp://www.ops.fhwa.dot.gov/publications/fhwaop04010/handbook.pdf\nhttp://ops.fhw
a.dot.gov/eto_tim_pse/preparedness/pse/handbook.htm
Lee, I., Shiroma, E., Lobelo, F., & Puska, P. (2012). Effect of physical inactivity on major non-
communicable diseases worldwide: an analysis of burden of disease and life expectancy. The
Lancet, 6736(panel 2), 1–11.
Lenschow, P. (2001). Some ideas about the sources of PM10. Atmospheric Environment, 35, 23–33.
http://doi.org/10.1016/S1352-2310(01)00122-4
Lioy, P. J., & Rappaport, S. M. (2011). Exposure Science and the Exposome: An Opportunity for
Coherence in the Environmental Health Sciences. Environmental Health Perspectives, 119(11),
a466–a467. http://doi.org/10.1289/ehp.1104387
Liu, F., Evans, J. E. (Jay), & Rossi, T. (2012). Recent Practices in Regional Modeling of Nonmotorized
Travel. Transportation Research Record: Journal of the Transportation Research Board, 2303, 1–8.
http://doi.org/10.3141/2303-01
Lund, H. (2003). Testing the Claims of New Urbanism: Local access, pedestrian travel, and neighboring
behaviors. Journal of the American Planning Association, 69(4), 414–429. http://doi.org/Article
Lurmann, F., Avol, E., & Gilliland, F. (2015). Emissions reduction policies and recent trends in Southern
California’s ambient air quality. Journal of the Air and Waste Management Association, 65(June
2016), 324–335. http://doi.org/10.1080/10962247.2014.991856
MacDonald, J. M., Stokes, R. J., Cohen, D. a, Kofner, A., & Ridgeway, G. K. (2010). The effect of light
rail transit on body mass index and physical activity. American Journal of Preventive Medicine,
39(2), 105–112. http://doi.org/10.1016/j.amepre.2010.03.016
Marshall, J. D., Brauer, M., & Frank, L. D. (2009). Healthy neighborhoods: walkability and air pollution.
Environmental Health Perspectives, 117(11), 1752–9. http://doi.org/10.1289/ehp.0900595
Marshall, J. D., McKone, T. E., Deakin, E., & Nazaroff, W. W. (2005). Inhalation of motor vehicle

201
emissions: Effects of urban population and land area. Atmospheric Environment, 39(2), 283–295.
http://doi.org/10.1016/j.atmosenv.2004.09.059
Matson, U., Ekberg, L. E., & Afshari, a. (2004). Measurement of Ultrafine Particles: A Comparison of
Two Handheld Condensation Particle Counters. Aerosol Science and Technology, 38(5), 487–495.
http://doi.org/10.1080/02786820490462200
Matthews, C. E., Hagströmer, M., Pober, D. M., & Bowles, H. R. (2012). Best practices for using
physical activity monitors in population-based research. Medicine and Science in Sports and
Exercise, 44(1 Suppl 1), S68–76. http://doi.org/10.1249/MSS.0b013e3182399e5b
McCormack, G. R., Giles-Corti, B., & Bulsara, M. (2008). The relationship between destination
proximity, destination mix and physical activity behaviors. Preventive Medicine, 46(1), 33–40.
http://doi.org/10.1016/j.ypmed.2007.01.013
McDonald, B. C., Dallmann, T. R., Martin, E. W., & Harley, R. A. (2012). Long-term trends in nitrogen
oxide emissions from motor vehicles at national, state, and air basin scales. Journal of Geophysical
Research Atmospheres, 117(17), 1–11. http://doi.org/10.1029/2012JD018304
McDonald, B. C., Gentner, D. R., Goldstein, A. H., & Harley, R. A. (2013). Long-term trends in motor
vehicle emissions in u.s. Urban areas. Environmental Science & Technology, 47(17), 10022–31.
http://doi.org/10.1021/es401034z
McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.),
Frontiers in Econometrics (pp. 105–142). New York: Academic Press.
McMurry, P., Zhang, X., & Lee, C. (1996). Issues in aerosol measurement for optics assessments. Journal
of Geophysical Research-Atmospheres, 101(D14), 19189–19197. http://doi.org/10.1029/95JD02342
McNeill, L. H., Kreuter, M. W., & Subramanian, S. V. (2006). Social Environment and Physical activity:
A review of concepts and evidence. Social Science and Medicine, 63, 1011–1022.
http://doi.org/10.1016/j.socscimed.2006.03.012
Mokhtarian, P. L. (2008). Examining the impacts of residential self-selection on travel behavior: A focus
on methodologies. Transportation Research, 42(3), 204–228.
Morello-Frosch, R., Zuk, M., Jerrett, M., Shamasunder, B., & Kyle, A. D. (2011). Understanding the
cumulative impacts of inequalities in environmental health: Implications for policy. Health Affairs,
30(5), 879–887. http://doi.org/10.1377/hlthaff.2011.0153
Morency, C., Trépanier, M., & Demers, M. (2011). Walking to transit: An unexpected source of physical
activity. Transport Policy, 18(6), 800–806. http://doi.org/10.1016/j.tranpol.2011.03.010
Newman, J. P., & Bernardin, V. L. (2010). Hierarchical ordering of nests in a joint mode and destination
choice model. Transportation, 37(4), 677–688.
Nieuwenhuijsen, M. J. (2016). Urban and transport planning, environmental exposures and health-new
202

concepts, methods and tools to improve health in cities. Environmental Health, 15(S1), 38.
http://doi.org/10.1186/s12940-016-0108-1
Norman, P., Conner, M., & Bell, R. (2000). The Theory of Planned Behaviour and exercise: Evidence for
the moderating role of past behaviour. British Journal of Health Psychology, 5(3), 249–261.
http://doi.org/10.1348/135910700168892
NRC. (1998). Research Priorities for Airborne Particulate Matter: I. Immediate Priorities and a Long-
Range Research Portfolio. Washington, D.C.: National Academies Press.
http://doi.org/10.17226/10957
O’Shaughnessy, P., & Cavanaugh, J. E. (2015). Performing T-tests to Compare Autocorrelated Time
Series Data Collected from Direct-Reading Instruments. Journal of Occupational and
Environmental Hygiene, 12(11), 743–752. http://doi.org/10.1080/15459624.2015.1044603
OEHHA. (2014). CalEnviroScreen version 2.0. Retrieved May 20, 2016, from
http://oehha.ca.gov/calenviroscreen/report/calenviroscreen-version-20
Oke, T. R. (1988). Boundary Layer Climates. U.K.: Routledge.
Ott, W., Klepeis, N., & Switzer, P. (2008). Air change rates of motor vehicles and in-vehicle pollutant
concentrations from secondhand smoke. Journal of Exposure Science & Environmental
Epidemiology, 18(3), 312–325. http://doi.org/10.1038/sj.jes.7500601
Ouellette, J., & Wood, W. (1998). Habit and intention in everyday life: the multiple processes by which
past behavior predicts future behavior. Psychological Bulletin, 124(1), 54–74.
Pendyala, R. (2009). Challenges and opportunities in advancing activity-based approaches for travel
demand analysis. In R. Kitamura, T. Yoshii, & T. Yamamoto (Eds.), The Expanding Sphere of
Travel Behaviour Research (pp. 303–335). Bingley, UK: Emerald Group Publishing.
Peng, R. D., & Dominici, F. (2008). Statistical Methods for Environmental Epidemiology with R: A Case
Study in Air Pollution and Health. Use R! New York, NY: Springer.
Pickett, K. E., & Pearl, M. (2001). Multilevel analyses of neighbourhood socioeconomic context and
health outcomes: a critical review. Journal of Epidemiology and Community Health, 55(2), 111–22.
Pinjari, A. R., & Bhat, C. (2011). Activity-based travel demand analysis. In A. de Palma, R. Lindsey, E.
Quinet, & R. Vickerman (Eds.), A Handbook of Transport Economics (pp. 213–248). UK: Edward
Elgar Publishing.
Pinjari, A. R., Pendyala, R. M., Bhat, C. R., & Waddell, P. a. (2011). Modeling the choice continuum: an
integrated model of residential location, auto ownership, bicycle ownership, and commute tour
mode choice decisions. Transportation, 38(6), 933–958.
Pokharel, S. S., Bishop, G. A., Stedman, D. H., & Slott, R. (2003). Emissions Reductions as a Result of
Automobile Improvement. Environmental Science and Technology, 37(22), 5097–5101.

203
http://doi.org/10.1021/es026340x
Quiros, D. C., Lee, E. S., Wang, R., & Zhu, Y. (2013). Ultrafine particle exposures while walking,
cycling, and driving along an urban residential roadway. Atmospheric Environment, 73, 185–194.
http://doi.org/10.1016/j.atmosenv.2013.03.027
R Development Core Team. (2014). A language and environment for statistical computing. Vienna,
Austria: R Foundation for Statistical Computing. Retrieved from http://www.r-project.org
Rabl, A., & de Nazelle, A. J. (2012). Benefits of shift from car to active transport. Transport Policy.
http://doi.org/10.1016/j.tranpol.2011.09.008
Ramachandran, G., Adgate, J. L., Pratt, G. C., & Sexton, K. (2003). Characterizing Indoor and Outdoor
15 Minute Average PM 2.5 Concentrations in Urban Neighborhoods. Aerosol Science and
Technology, 37(1), 33–45. http://doi.org/10.1080/02786820300889
Rissel, C., Curac, N., Greenaway, M., & Bauman, A. (2012). Physical activity associated with public
transport use--a review and modelling of potential benefits. International Journal of Environmental
Research and Public Health, 9(7), 2454–78. http://doi.org/10.3390/ijerph9072454
Rodríguez, D. A., Khattak, A. J., & Evenson, K. R. (2006). Can new Urbanism encourage physical
activity?: Comparing a new Urbanist neighborhood with conventional suburbs. Journal of the
American Planning Association, 72(1), 43–54.
Rojas-Rueda, D., de Nazelle, A. J., Tainio, M., & Nieuwenhuijsen, M. J. (2011). The health risks and
benefits of cycling in urban environments compared with car use: health impact assessment study.
Bmj, 343(aug04 2), d4521–d4521. http://doi.org/10.1136/bmj.d4521
Rojas-Rueda, D., de Nazelle, A. J., Teixidó, O., & Nieuwenhuijsen, M. J. (2012). Replacing car trips by
increasing bike and public transport in the greater Barcelona metropolitan area: A health impact
assessment study. Environment International, 49, 100–109.
http://doi.org/10.1016/j.envint.2012.08.009
Saelens, B. E., Sallis, J. F., & Frank, L. D. (2003). Environmental correlates of walking and cycling:
findings from the transportation, urban design, and planning literatures. Annals of Behavioral
Medicine : A Publication of the Society of Behavioral Medicine, 25(2), 80–91.
Saint-Maurice, P. F., Welk, G. J., Russell, D. W., & Huberty, J. (2014). Moderating influences of baseline
activity levels in school physical activity programming for children: the Ready for Recess project.
BMC Public Health, 14(1), 103. http://doi.org/10.1186/1471-2458-14-103
Sallis, J. F., Cervero, R. B., Ascher, W., Henderson, K. a K. A., Kraft, M. K. K., & Kerr, J. (2006). An
ecological approach to creating active living communities. Annual Review of Public Health, 27(1),
297–322. http://doi.org/10.1146/annurev.publhealth.27.021405.102100
Sallis, J. F., Conway, T. L., Dillon, L. I., Frank, L. D., Adams, M. A., Cain, K. L., & Saelens, B. E.
204

(2013). Environmental and demographic correlates of bicycling. Preventive Medicine, 57(5), 456–
60. http://doi.org/10.1016/j.ypmed.2013.06.014
Sallis, J. F., Floyd, M. F., Rodríguez, D. a, & Saelens, B. E. (2012). Role of built environments in
physical activity, obesity, and cardiovascular disease. Circulation, 125(5), 729–37.
http://doi.org/10.1161/CIRCULATIONAHA.110.969022
Sallis, J. F., Frank, L. D., Saelens, B. E., & Kraft, M. K. (2004). Active transportation and physical
activity: opportunities for collaboration on transportation and public health research. Transportation
Research-Part A Policy and Practice, 38(4), 249–268.
Schwartz, J., Litonjua, A., Suh, H., Verrier, M., Zanobetti, A., Syring, M., … Gold, D. R. (2005). Traffic
related pollution and heart rate variability in a panel of elderly subjects. Thorax, 60(6), 455–61.
Schweitzer, L., & Zhou, J. (2010). Neighborhood Air Quality, Respiratory Health, and Vulnerable
Populations in Compact and Sprawled Regions. Journal of the American Planning Association,
76(3), 363–371. http://doi.org/10.1080/01944363.2010.486623
Sener, I. N., & Bhat, C. R. (2011). Modeling the spatial and temporal dimensions of recreational activity
participation with a focus on physical activities. Transportation, 39(3), 627–656.
Setton, E., Marshall, J. D., Brauer, M., Lundquist, K. R., Hystad, P., Keller, P., & Cloutier-Fisher, D.
(2011). The impact of daily mobility on exposure to traffic-related air pollution and health effect
estimates. Journal of Exposure Science & Environmental Epidemiology, 21(1), 42–48.
http://doi.org/10.1038/jes.2010.14
Shu, S., Batteate, C., Cole, B., Froines, J., & Zhu, Y. (2015). Air quality impacts of a CicLAvia event in
Downtown Los Angeles, CA. Environmental Pollution, (February), 0–7.
http://doi.org/10.1016/j.envpol.2015.09.010
Siegel, J. D., Cea, J., Fernández, J. E., Rodriguez, R. E., & Boyce, D. (2006). Comparisons of Urban
Travel Forecasts Prepared with the Sequential Procedure and a Combined Model. Networks and
Spatial Economics, 6(2), 135–148.
Sirard, J., Melanson, E., Li, L., & Freedson, P. (2000). Field evaluation of the Computer Science and
Applications, Inc. physical activity monitor. Medicine & Science in Sports & Exercise, 7164(8),
695–700.
Spears, S., Houston, D., & Boarnet, M. G. (2013). Illuminating the unseen in transit use: A framework for
examining the effect of attitudes and perceptions on travel behavior. Transportation Research Part
A: Policy and Practice, 58, 40–53. http://doi.org/10.1016/j.tra.2013.10.011
StataCorp. (2013). Stata Statistical Software: Release 13. College Station, TX: StataCorp LP.
Steinle, S., Reis, S., & Sabel, C. E. (2013). Quantifying human exposure to air pollution-Moving from
static monitoring to spatio-temporally resolved personal exposure assessment. The Science of the

205
Total Environment, 443, 184–93. http://doi.org/10.1016/j.scitotenv.2012.10.098
Stokes, R. J., MacDonald, J., & Ridgeway, G. (2008). Estimating the effects of light rail transit on health
care costs. Health & Place, 14(1), 45–58. http://doi.org/10.1016/j.healthplace.2007.04.002
Stone, B., Mednick, A. C., Holloway, T., & Spak, S. N. (2007). Is compact growth good for air quality?
Journal of the American Planning Association, 73(4), 404–418.
http://doi.org/10.1080/01944360701653235
Thai, A., McKendry, I., & Brauer, M. (2008). Particulate matter exposure along designated bicycle routes
in Vancouver, British Columbia. Science of the Total Environment, 405(1-3), 26–35.
Thøgersen, J. (2006). Understanding repetitive travel mode choices in a stable context: A panel study
approach. Transportation Research Part A: Policy and Practice, 40(8), 621–638.
http://doi.org/10.1016/j.tra.2005.11.004
Timperio, A., Ball, K., Salmon, J., Roberts, R., Giles-Corti, B., Simmons, D., … Crawford, D. (2006).
Personal, family, social, and environmental correlates of active commuting to school. American
Journal of Preventive Medicine, 30(1), 45–51. http://doi.org/10.1016/j.amepre.2005.08.047
TRB. (2011). Policy Options for Reducing Energy Use and Greenhouse Gas Emissions from U.S.
Transportation. Transportation Research Board. Washington, D.C. Retrieved from
http://www.trb.org/Main/Blurbs/165535.aspx
Troiano, R. P., Berrigan, D., Dodd, K. W., Mâsse, L. C., Tilert, T., & McDowell, M. (2008). Physical
activity in the United States measured by accelerometer. Medicine and Science in Sports and
Exercise, 40(1), 181–8. http://doi.org/10.1249/mss.0b013e31815a51b3
Tukey, J. W. (1977). Exploratory data analysis. Boston, MA: Addison-Wesley.
United Nations. (2014). World urbanization prospects: The 2014 revision. New York.
US DOT. (2014). Statewide transportation planning for healthy communities. Washington, DC.
Van Acker, V., Van Wee, B., & Witlox, F. (2010). When Transport Geography Meets Social Psychology:
Toward a Conceptual Model of Travel Behaviour. Transport Reviews, 30(2), 219–240.
Van Poppel, M., Peters, J., & Bleux, N. (2013). Methodology for setup and data processing of mobile air
quality measurements to assess the spatial variability of concentrations in urban environments.
Environmental Pollution (Barking, Essex : 1987), 183, 224–33.
http://doi.org/10.1016/j.envpol.2013.02.020
Vardoulakis, S. (2003). Modelling air quality in street canyons: a review. Atmospheric Environment,
37(2), 155–182.
Verplanken, B., Aarts, H., Van Knippenberg, A., & Moonen, A. (1998). Habit versus planned behaviour:
A field experiment. British Journal of Social Psychology, 37, 111–128. http://doi.org/9554090
Verplanken, B., & Melkevik, O. (2008). Predicting habit: The case of physical exercise. Psychology of
206

Sport and Exercise, 9(1), 15–26. http://doi.org/10.1016/j.psychsport.2007.01.002
Visser, M., & Pluijm, S. (2002). Physical activity as a determinant of change in mobility performance:
The Longitudinal Aging Study Amsterdam. Journal of the American Geriatrics Society, 50, 1774–
1781.
Wasfi, R. A., Dasgupta, K., Eluru, N., & Ross, N. A. (2015). Exposure to walkable neighbourhoods in
urban areas increases utilitarian walking: Longitudinal study of Canadians. Journal of Transport and
Health. http://doi.org/10.1016/j.jth.2015.08.001
Westerdahl, D., Fruin, S. A., Sax, T., Fine, P. M., & Sioutas, C. (2005). Mobile platform measurements of
ultrafine particles and associated pollutant concentrations on freeways and residential streets in Los
Angeles. Atmospheric Environment, 39(20), 3597–3610.
http://doi.org/10.1016/j.atmosenv.2005.02.034
Whitlow, T. H., Hall, A., Zhang, K. M., & Anguita, J. (2011). Impact of local traffic exclusion on near-
road air quality: findings from the New York City “Summer Streets” campaign. Environmental
Pollution, 159(8-9), 2016–2027. http://doi.org/10.1016/j.envpol.2011.02.033
WHO. (2009). Global health risks: mortality and burden of disease attributable to selected major risks.
Geneva: World Health Organization.
Wild, C. P. (2005). Complementing the genome with an “exposome”: The outstanding challenge of
environmental exposure measurement in molecular epidemiology. Cancer Epidemiology
Biomarkers and Prevention, 14(8), 1847–1850. http://doi.org/10.1158/1055-9965.EPI-05-0456
Willsher, K. (2015, October 3). Paris’s first attempt at car-free day brings big drop in air and noise
pollution. The Guardian. London, UK. Retrieved from
http://www.theguardian.com/world/2015/oct/03/pariss-first-attempt-at-car-free-day-brings-big-drop-
in-air-and-noise-pollution
Wilson, J. G., Kingham, S., Pearce, J., & Sturman, A. P. (2005). A review of intraurban variations in
particulate air pollution: Implications for epidemiological research. Atmospheric Environment,
39(34), 6444–6462.
Woodruff, T. J., Parker, J. D., Darrow, L. A., Slama, R., Bell, M. L., Choi, H., … Wilhelm, M. (2009).
Methodological issues in studies of air pollution and reproductive health. Environmental Research,
109(3), 311–320. http://doi.org/10.1016/j.envres.2008.12.012
World Health Organization. (2014a). Ambient (outdoor) air quality and health. Retrieved February 10,
2016, from http://www.who.int/mediacentre/factsheets/fs313/en/
World Health Organization. (2014b). Ambient air pollution database. World Health Organization.
Retrieved from http://www.who.int/phe/health_topics/outdoorair/databases/cities/en/
Yordy, G., & Lent, R. (1993). Predicting aerobic exercise participation: social cognitive, reasoned action,

207
and planned behavior models. Journal of Sport & Exercise Psychology, 15, 363–374.
Zenk, S., Schulz, A., & Matthews, S. (2011). Activity space environment and dietary and physical
activity behaviors: a pilot study. Health & Place, 17(5), 1150–1161.
Zheng, Y. (2008). The benefit of public transportation: physical activity to reduce obesity and ecological
footprint. Preventive Medicine, 46(1), 4–5. http://doi.org/10.1016/j.ypmed.2007.11.019
Zuurbier, M., Hoek, G., Oldenwening, M., Lenters, V., Meliefste, K., van den Hazel, P., & Brunekreef, B.
(2010). Commuters’ exposure to particulate matter air pollution is affected by mode of transport,
fuel type, and route. Environmental Health Perspectives, 118(6), 783–9.
Zwiers, F. W., & Von Storch, H. (1995). Taking serial correlation into account in tests of the mean.
Journal of Climate, 8(2), 336–351. http://doi.org/10.1175/1520-
0442(1995)008<0336:TSCIAI>2.0.CO;2

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The built environment, tour complexity, and active travel

PDF

Environmental justice in real estate, public services, and policy

PDF

Urban air pollution and environmental justice: three essays

PDF

Active travel, outdoor leisure, and neighborhood environment: path analysis, Los Angeles County

PDF

The flexible workplace: regional tendencies and daily travel implications

PDF

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

PDF

Towards health-conscious spaces: building for human well-being and performance

PDF

Toxicity of urban particulate matter: long-term health risks, influences of surrounding geography, and diurnal variation in chemical composition and the cellular oxidative stress response

Asset Metadata

Creator Hong, E-Sok Andy (author)

Core Title Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Publication Date 08/01/2016

Defense Date 06/14/2016

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

Tag active transportation,Air pollution,built environment,car-free street,mode switch,nonmotorized,OAI-PMH Harvest,physical activity,traffic environment,ventilation

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Boarnet, Marlon G. (committee chair), Fruin, Scott (committee member), Schweitzer, Lisa (committee member)

Creator Email andyhong@gmail.com,esokhong@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-288857

Unique identifier UC11281090

Identifier etd-HongESokAn-4691.pdf (filename),usctheses-c40-288857 (legacy record id)

Legacy Identifier etd-HongESokAn-4691.pdf

Dmrecord 288857

Document Type Dissertation

Format application/pdf (imt)

Rights Hong, E-Sok Andy

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