University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Investigating Daily Effects of Activity Space-based Built Environment Exposures on Physical
Activity Behaviors in Hispanic Women during Pregnancy and Early Postpartum

by

Li Yi

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POPULATION, HEALTH AND PLACE)

May 2022

Copyright © 2022 Li Yi

ii
Dedication

To my parents, wife, and furry paws.

iii
Acknowledgements
I am grateful to my mentors and co-advisors, Dr. John Wilson and Dr. Genevieve Dunton, and
my dissertation committee member, Dr. Rima Habre, for their support, guidance, and patience. I
am also grateful to my other dissertation committee member, Dr. Geoff Boeing, for his
commitment and thoughtful contribution to the work of this dissertation. I am thankful for the
guidance from Dr. Sandrah Eckel, Dr. Teresa Bastain, Dr. Carrie Breton, Dr. Tyler Mason, Dr.
Brian Finch at various stages of this dissertation work. I am grateful for the support from my
colleagues and friends in the PHP program and REACH lab, especially Yan Xu, Dr. Johanna
Avelar, Lois Park, Kate Vavra-Musser, Emily Serman, Avery Everhart, Dr. Daniel Chu, Dr.
Sydney O’Conner, Dr. Chih-Hsiang Yang, Dr. Wei-Lin Wang, Shirlene Wang, Bridgette Do,
Christine Naya, and Dr. Martina Kanning. I am grateful for the data that these studies are based
on provided to me by Wangjing Ke (madresGPS app developer), Jane Cabison, Marisela
Rosales, and the larger MADRES team. I am also grateful to the MADRES participants and their
families and our clinic partners for their time and effort. My appreciation goes to Prof. Elizabeth
Plater-Zyberk, Dr. Dennis Hector, Prof. Joanna Lombard, Dr. Scott Brown, Victor Dover, Jason
King, and Pam Stacy, who led me into my current research field. Lastly, my deepest gratitude to
my parents – Lijuan Yi and Zhenfa Lin, for their firm support of my education and life
endeavors, and my wife – Ying Peng, and my furry paws – Mango and Nimei for their day and
night accompany, support, and encouragement throughout this epic journey.
iv
Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ....................................................................................................................... iii
List of Tables ............................................................................................................................... viii
List of Figures ................................................................................................................................. ix
Abbreviations ................................................................................................................................... x
Abstract ........................................................................................................................................... xi
Chapter 1 Introduction ..................................................................................................................... 1
1.1. Disparities of Obesity-Related Outcomes in Hispanic Women ......................................... 1
1.2. Critical Gaps in Research ................................................................................................... 2
1.3. Analytical Challenges ......................................................................................................... 4
1.3.1. Daily Mobility in Health and Place Research ........................................................... 4
1.3.2. Uncertain Geographic Context Problem (UGCoP) ................................................... 5
1.3.3. Within-Person Effects of BE Exposures ................................................................... 6
1.3.4. Moderating Mechanisms in Behavioral Pathways .................................................... 7
1.4. Mobile Sensing in PA Research ......................................................................................... 8
1.5. Conceptual Framework and Dissertation Outline .............................................................. 9
Chapter 2 Time-Activity and Daily Mobility Patterns during Pregnancy and
Early Postpartum .................................................................................................................... 12
2.1. Related Work .................................................................................................................... 12
2.2. Methods ............................................................................................................................ 16
2.2.1. Study Design ............................................................................................................ 16
2.2.2. Data Collection ........................................................................................................ 17
2.2.3. Data processing ........................................................................................................ 18
2.2.4. Statistical Analysis .................................................................................................. 27
v
2.3. Results .............................................................................................................................. 30
2.3.1. Data Completeness .................................................................................................. 30
2.3.2. Descriptive Statistics of Covariates ......................................................................... 31
2.3.3. Descriptive Statistics for Time-Activity and Daily Mobility Patterns .................... 33
2.3.4. Base GLMM Results ............................................................................................... 39
2.3.5. Final GLMM Results ............................................................................................... 42
2.4. Discussion ......................................................................................................................... 47
2.4.1. Time-Activity and Mobility Patterns of Pregnant Women ..................................... 47
2.4.2. Changes in Time-Activity and Mobility Patterns during Pregnancy and
Postpartum .................................................................................................................. 49
2.4.3. Additional Predictors of Time-Activity and Mobility Patterns ............................... 50
2.4.4. Implications for Future Studies ............................................................................... 51
2.4.5. Study Strengths and Limitations .............................................................................. 52
2.5. Conclusions ...................................................................................................................... 54
Chapter 3 Assessing Dynamic Daily Exposures to Built-Environment Characteristics
during Pregnancy and Early Postpartum using Smartphone Location Data .......................... 55
3.1. Related Work .................................................................................................................... 55
3.2. Methods ............................................................................................................................ 58
3.2.1. Design and Overview .............................................................................................. 58
3.2.2. Data Collection ........................................................................................................ 59
3.2.3. Data Processing ....................................................................................................... 60
3.2.4. Statistical Analysis .................................................................................................. 65
3.3. Results .............................................................................................................................. 67
3.3.1. Data Completeness .................................................................................................. 67
3.3.2. Daily BE Exposure Estimates during Pregnancy and Early Postpartum ................. 67
3.3.3. Within-person Correlations in Daily BE Exposures across Time ........................... 70
vi
3.3.4. Correlations of BE Exposures across methods (Static versus Dynamic) ................ 70
3.3.5. Impact of Daily Mobility on Potential for Exposure Misclassification
using the Static Method .............................................................................................. 71
3.4. Discussion ......................................................................................................................... 81
3.4.1. Women’s Daily BE Exposures during Pregnancy and Early Postpartum ............... 82
3.4.2. Similarities of Exposure Estimates between Static and Dynamic Methods ............ 83
3.4.3. Exposure Misclassification Introduced by the Static Method to Due to Daily
Mobility ...................................................................................................................... 85
3.4.4. Study Strengths and Limitations .............................................................................. 86
3.5. Conclusions ...................................................................................................................... 88
Chapter 4 The Association between Built Environment Characteristics and Physical
Activity during Pregnancy and Early Postpartum .................................................................. 89
4.1. Related Work .................................................................................................................... 89
4.2. Methods ............................................................................................................................ 92
4.2.1. Study Design and Participants ................................................................................. 92
4.2.2. BE Measures ............................................................................................................ 93
4.2.3. PA Outcomes ........................................................................................................... 96
4.2.4. Covariates ................................................................................................................ 97
4.2.5. Statistical Analysis .................................................................................................. 98
4.3. Results ............................................................................................................................ 100
4.3.1. Descriptive Statistics ............................................................................................. 100
4.3.2. Predictors of Day-Level PA Outcomes in Pregnant Women ................................ 103
4.3.3. Moderators of Day-Level Associations between BE and PA ................................ 105
4.4. Discussion ....................................................................................................................... 111
4.4.1. Day-Level Associations between BE and PA ....................................................... 112
4.4.2. Differences in Day-Level BE-PA Associations by Temporal Factors .................. 113
vii
4.4.3. Differences in Day-Level BE-PA Associations by Individual
Sociodemographics and Neighborhood Characteristics ........................................... 115
4.4.4. Study Strengths and Limitations ............................................................................ 117
4.5. Conclusions .................................................................................................................... 119
Chapter 5 Conclusions ................................................................................................................. 121
5.1. Contributions and Connections ...................................................................................... 121
5.2. Limitations and Directions for Future Studies ............................................................... 125
5.3. Major Takeaways ........................................................................................................... 128
References ................................................................................................................................... 130
Appendix A: Confidence flagging algorithm for signal loss scenarios using the Maternal
And Developmental Risks from Environmental and Social stressors (MADRES)
real-time and personal sampling study Global Positioning System (GPS) data .................. 147
Appendix B: Rules used to classify contexts of stays detected with the Maternal And
Developmental Risks from Environmental and Social stressors (MADRES) real-time
and personal sampling study Global Positioning System (GPS) data .................................. 148
Appendix C: An illustration of the GPS imputation rules based on existing data and home
locations detected from context classification. .................................................................... 149
Appendix D: Individual sociodemographics, neighborhood characteristics, and temporal
factors included in the study along with their measures and data sources ........................... 150
Appendix E: Zero-inflated Generalized Linear Mixed-Effects Model (GLMM) results for
changes in time-activity patterns by pregnancy and postpartum period .............................. 151
Appendix F: Zero-inflated Generalized Linear Mixed-Effects Model (GLMM) results for
changes in daily mobility patterns by pregnancy and postpartum period ............................ 152
Appendix G: Sensitivity analyses results of between-subject (BS) and within-subject (WS)
effects of built-environment (BE) exposures on day-level moderate-to-vigorous physical
activity (MVPA) outcomes .................................................................................................. 153
Appendix H: Significant interactions between day-level built-environment (BE) exposures
and temporal, individual and neighborhood factors in predicting day-level
moderate-to-vigorous physical activity (MVPA) outcomes ................................................ 155

viii
List of Tables
Table 2.1. Descriptive statistics of participant characteristics at baseline (a) and person-day
level temporally varying factors by 1
st
and 3
rd
trimesters, and 4-6 months postpartum (b). ......... 32
Table 2.2. Summary of total number of visits to multiple spatial contexts and indoor/outdoor
microenvironments and total number of pedestrian and vehicular trips made. ............................. 34
Table 2.3. Day-level summary of time spent in spatial contexts, indoor/outdoor
microenvironments, and number of pedestrian/vehicular trips made. ........................................... 35
Table 2.4. Zero-inflated Generalized Mixed-Effects Model (GLMM) results for time
spent in five non-home spatial contexts adjusted for covariates. .................................................. 43
Table 2.5. Zero-inflated Generalized Mixed-Effects Model (GLMM) results for time
spent in home and non-home contexts and microenvironments adjusted for covariates. ............. 44
Table 2.6. Zero-inflated Generalized Mixed-Effects Model (GLMM) results for
time spent in pedestrian and vehicular trips and number of pedestrian and vehicular trips
performed adjusted for covariates. ................................................................................................ 45
Table 3.1. Built-environment (BE) measures, corresponding data sources, resolutions, and
processing steps. ............................................................................................................................ 64
Table 3.2. Descriptive statistics of daily built-environment (BE) exposure estimates for
participants using the route buffer (RB), kernel density estimation (KDE), and residential
network buffer (RNB) methods. .................................................................................................... 68
Table 3.3. Intraclass-Correlation Coefficients (ICCs) of daily exposure estimates
derived using the kernel density estimation (KDE) and route buffer (RB) methods. ................... 71
Table 3.4. Pearson correlation coefficients between day-level static versus dynamic
exposure estimates for the same built-environment (BE) variable during the 1
st
and 3
rd

trimesters and 4-6 months postpartum periods. ............................................................................. 72
Table 4.1. Descriptive statistics for physical activity (PA) outcomes, built-environment
(BE) predictors, and covariates of the study participants. ........................................................... 102
Table 4.2. Between-subject (BS) and within-subject (WS) effects of exposures to built-
environment (BE) characteristics derived using the 250 m kernel density estimation (KDE)
and 250 m route buffer (RB) methods on daily moderate-to-vigorous physical activity
(MVPA) outcomes. ...................................................................................................................... 104

ix
List of Figures
Figure 1.1. The conceptual framework within which topics of this dissertation research are
organized. ...................................................................................................................................... 10
Figure 2.1. Global Positioning System (GPS) pre-processing steps of Maternal And
Developmental Risks from Environmental and Social Stressors (MADRES) real-time and
personal sampling study GPS data. ............................................................................................... 19
Figure 2.2. Geoprocessing steps executed to detect stays from Global Positioning System
(GPS) data, classify their contexts based on land use, and their indoor/outdoor
microenvironments based on building footprints. ......................................................................... 23
Figure 2.3. Geoprocessing steps to detect trips and classify their modes based on mean and
standard deviation of Global Positioning System (GPS) observations in trips. ............................ 26
Figure 2.4. Distributions of top five origin-destination combinations by pedestrian- and
vehicle-based trip modes. .............................................................................................................. 40
Figure 2.5. Base Generalized Mixed-Effects Model results of variations in time-activity
and daily mobility patterns by 1st and 3rd trimesters of pregnancy and
4-6 months postpartum. ................................................................................................................. 41
Figure 3.1. An illustration of two activity space metrics – route buffer (RB) and kernel density
estimation (KDE) applied in this study ......................................................................................... 61
Figure 3.2. The directions of potential residential network buffer (RNB) exposure
misclassifications for built-environment (BE) measures compared to kernel density
estimation (KDE) estimates. .......................................................................................................... 73
Figure 4.1. The significant interactions between built-environment (BE) characteristics and
temporal factors, individual sociodemographics, and neighborhood characteristics in
predicting day-level moderate-to-vigorous physical activity (MVPA) outcomes. ...................... 106

x
Abbreviations
BE Built Environment

EMA Ecological Momentary Assessment

GIS Geographic Information Systems

GLMM Generalized Linear Mixed-Effects Model

GPS Global Positioning System

ICC Intraclass Correlation Coefficient

LMM Linear Mixed-Effects Model

KDE Kernel Density Estimation

MADRES Maternal and Development Risks from Environmental and Social Stressors

MCU Measured Contextual Unit

MVPA Moderate-to-Vigorous Physical Activity

NDVI Normalized Differenced Vegetation Index

PA Physical Activity

RB Route Buffer

RNB Residential Network Buffer

SES Socio-Economic Status

TAGA Time-Aligned GPS Accelerometry Dataset

TCU True Contextual Unit

UGCoP Uncertain Geographic Context Problem

USC University of Southern California

xi
Abstract
Increasing rates of maternal obesity during and after pregnancy are associated with numerous
short- and long-term health risks and outcomes in women postpartum. Hispanic women of
childbearing age have disproportionally high obesity risks and low physical activity (PA) levels
compared to non-Hispanic white populations. This dissertation study used 4-day objective,
highly resolved smartphone location and accelerometry-assessed activity data collected from 62
Hispanic women of childbearing age in urban Los Angeles, CA, during pregnancy and the early
postpartum period to investigate the associations among women’s daily mobility patterns
(Chapter 2), dynamic built-environment (BE) exposures (Chapter 3), and PA outcomes (Chapter
4). Results of these three empirical case studies revealed exceedingly low parks and open space
exposure for this group at their daily activity locations and along travel paths during and after
pregnancy, which had negative impacts on their day-to-day PA outcomes. In addition, important
modifiers (e.g., late pregnancy, early postpartum, high pre-pregnancy BMI, low neighborhood
safety) of women’s daily mobility and BE exposures, and their associations with PA outcomes
were identified. Lastly, measurement error and bias resulting from applying traditional
residential-based measures were evaluated and their implications for uncovering the relationships
between BE exposures and PA outcomes were investigated. Future studies should conduct
qualitative analyses of the BE features (e.g., parks) in which women’s PA were performed,
mitigate selective daily mobility bias, and apply real-time surveying techniques such as
ecological momentary assessment (EMA) to elucidate psychosocial pathways from BE
exposures to PA outcomes. Future PA promotion interventions for pregnant women should target
at-risk pregnancy periods and sub-population groups to improve their efficacy, especially for
those of low socioeconomic status and specific racial/ethnic minority groups.

1
Chapter 1 Introduction
Increasing rates of maternal obesity during and after pregnancy are associated with numerous
short-term and long-term health risks and outcomes for women postpartum (Algoblan, Alalfi,
and Khan 2014; Fan et al. 2013; Fowles, and Sterling 2011; Fraser et al. 2011; Soltani and Fraser
2000; Walker, Walter et al. 2015). Rates of pregnancy-related obesity risks and health outcomes
are disproportionally high in Hispanic women (Brawarsky et al. 2005; Chasan-Taber et al. 2008;
Headen et al. 2012; Ogden et al. 2016). Research indicates physical activity (PA), a common
etiological factor of obesity, is disproportionally low among Hispanic pregnant women (August
and Sorkin 2011; Hughes, McDowell, and Brody 2008; Arredondo et al. 2016), which may be
explained by women’s consistent exposures to the built environment (BE) barriers for PA. This
chapter describes the critical gaps and analytical problems in past studies examining the
associations between BE exposure and women’s PA during and after pregnancy, presents a
conceptual framework that aims to tackle these gaps, and further elucidates the impact of BE on
PA in women of childbearing age, particularly for those of low-income or in specific
racial/ethnic minority groups.
1.1. Disparities of Obesity-Related Outcomes in Hispanic Women
Obesity prevalence is rapidly growing in the US, with the latest 2016 National Health and
Nutrition Examination Survey showing 39.8% of US adults were obese (Hales et al. 2018).
Obesity rates for US women of childbearing age have doubled over the past 30 years (Ogden et
al. 2016). Evidence indicates more than 40% of women gained weight during pregnancy that
exceeded recommended levels (Rasmussen, Catalano, and Yaktine 2009). Increasing rates of
maternal obesity during and after pregnancy pose serious health concerns for mothers. Higher
pre-pregnancy weight and excessive weight gain during pregnancy lead to increased risks of

2
postpartum weight retention (Soltani and Fraser 2000; Walker, Fowles, and Sterling 2011),
which can result in long-term health outcomes such as diabetes, cancer, and cardiovascular
diseases (Algoblan, Alalfi, and Khan 2014; Fan et al. 2013; Fraser et al. 2011; Walter et al.
2015). Rates of pregnancy-related obesity risks and health outcomes are disproportionally high
in Hispanic women. Latest data indicates that 40% of Hispanic women of childbearing age were
obese compared to 31% for non-Hispanic white women of this age and 51% of them gained
more weight than recommended during pregnancy (Brawarsky et al. 2005; Chasan-Taber et al.
2008; Headen et al. 2012; Ogden et al. 2016). Energy-balance behaviors such as PA and diet are
common etiological factors of obesity (Taveras et al. 2010; Singh et al. 2008). Although few
existing studies have examined PA in Hispanic women of children bearing age, some evidence
indicates Hispanic women overall was less likely to meet the PA guidelines than non-Hispanic
white population (August and Sorkin 2011; Hughes, McDowell, and Brody 2008; Arredondo et
al. 2016). Understanding drivers of causes of disparities in PA outcomes among minority women
is a critical step towards reducing the disproportionate obesity risks borne by these groups.
1.2. Critical Gaps in Research
According to Sallis’s socioecological model, exposure to the BE could be linked to
obesity-related risks and health outcomes through influencing individual-level energy-balance
behaviors (Sallis et al. 2006). BE exposures typically refer to objectively assessed measures of
access to destinations (e.g., parks) and features (e.g., street trees) that might influence behaviors,
as well as measures characterizing urban form (e.g., walkability) within certain spatial extents
(e.g., buffers of home address points) and time periods (e.g., days) (Sallis et al. 2012; Booth,
Pinkston, and Poston 2005). For pregnant women, past studies report mixed results of
associations between BE features and characteristics and their PA behaviors. For instance, a

3
Norwegian study showed positive associations between both objectively measured and perceived
access to parks and open space in the residential neighborhood and moderate-to-vigorous PA
(MVPA) minutes during and after pregnancy (Richardsen et al. 2016). However, similar
associations between total green space area within residential census tracts and PA outcomes
during pregnancy were not significant in a New Zealand study (Nichani et al. 2016). In addition,
past studies indicate associations between BE features and PA outcomes only exist at a specific
pre- and postpartum period. For instance, a US study indicates total greenspace area within
residential census tracts was positively associated with women’s MVPA minutes during the 3
rd

trimester, while the same study also reports distance from home to transit and PA facilities were
negatively associated with MVPA minutes at postpartum periods (Porter et al. 2019). However,
there is little research on how the effects of BE on PA behaviors operate during the pre- and
postpartum periods. Since PA behaviors of women may be subject to dramatic changes during
the pre- and postpartum periods due to growing family and childcare responsibilities (Borodulin,
Evenson, and Herring 2009; O’Brien et al. 2017), as well as biophysiological changes, it is
important to examine the contributions of various BE features or characteristics to PA among
this population and how their associations vary across periods for these populations.
Moreover, besides the disproportionate rates of obesity experienced by Hispanic women
of childbearing age, previous studies also show that women from disadvantaged minority and
low-income groups are disproportionately exposed to BE barriers that deter PA, including low
access to recreational facilities and parks and open space, low quality pedestrian infrastructure
such as under-maintained sidewalks, and less aesthetically pleasing streetscapes (Lovasi et al.
2009; Perez, Ruiz, and Berrigan 2019; Sallis et al. 2011; Singh et al. 2008). In addition, past
research on low-income populations also indicates how the effects of various BE elements might

4
operate differently among different socioeconomic status (SES) groups. For example, a study on
exercise among low-income African American women reports high membership costs of
recreational facilities in their neighborhoods as one of the major deterrents for them to utilize
such facilities to exercise (Krans and Chang 2011). As a result, research to examine BE features
and the characteristics associated with PA behaviors among low-income Hispanic pregnant
women can further elucidate the behavioral pathways that link BE exposures to gestational
weight gain and postpartum weight retention in this population, and to formulate evidence-based
behavioral interventions, policies, and guidelines to combat the high obesity prevalence.
1.3. Analytical Challenges
1.3.1. Daily Mobility in Health and Place Research
In addition to a lack of research on influences of BE exposures on PA behaviors during
and after pregnancy, several analytical and inferential problems arise in studies drawing
inferences between BE exposures and PA outcomes. The most prominent is the lack of
integration of individual daily mobility into exposure assessment. Many of the published studies
use the residence-based neighborhood (e.g., radius buffer around the home location or census
tract or block group) as the measured contextual unit (MCU) of analysis, in which BE features
and characteristics are assessed with arbitrarily defined spatial extents that may or may not be the
most relevant or influential for the specific outcome under study (Jankowska, Schipperijn, and
Kerr 2015; James et al. 2016). This residence-based approach assumes the effect of BE exposure
on health behaviors and outcomes primarily or exclusively operates around the home and ignores
areas of contact or exposures that occur outside of the home based on individuals’ every day
mobility patterns. This may be true if the study population spends most of its time in the home
neighborhood; however, empirical evidence suggests otherwise. For instance, several studies

5
(e.g., Evenson et al. 2009; Nethery, Brauer, and Janssen 2009; Ouidir et al. 2015; Zhu et al.
2019) report their participants are highly mobile during data collection periods and spend a large
proportion of time at non-home locations. As a result, the residence-based approach might result
in exposure misclassification and attenuate or potentially bias effect estimates in subsequent
health analyses (Jankowska, Schipperijn, and Kerr 2015; James et al. 2016).
1.3.2. Uncertain Geographic Context Problem (UGCoP)
The lack of integration of daily mobility into BE exposure assessment is part of the
uncertain geographic contexts problem (UGCoP) (Robertson and Feick 2018; Kwan 2012). The
problem "arises because of the spatial uncertainty (e.g., buffer sizes) in the actual areas that exert
contextual influences on the individuals being studied, and the temporal uncertainty of the timing
and duration in which individuals experienced these contextual influences" (Kwan 2012). It is
also very difficult for any individual study that uses area-based BE contextual variables to
explain individual behaviors to fully overcome this problem because the complete and perfect
knowledge of “true causally relevant” BE contexts for PA behavior is unknown (Kwan 2012).
The mitigation of UGCoP requires more accurate measurement and estimation the "true
causally relevant" geographic context (true contextual unit or TCU) than is accomplished with
residence-based approach (Kwan 2012; Matthews and Yang 2013). To achieve this, a growing
body of research has delineated MCUs through generating daily activity spaces (i.e., activity
locations visited, or paths traversed in their daily lives) and measure BE exposures within
activity space-based MCUs (Perchoux, Chaix, and Kestens 2019; Matthews and Yang 2013;
Rainham et al. 2010; Sherman et al. 2005). In compared to the residence-based approach, the
activity space-based approach is able to improve accuracy of exposure assessment by linking BE
exposures and PA outcomes in space and time beyond home contexts (Yi et al. 2019).

6
Nevertheless, there remains the uncertainty of choices of spatial (e.g., buffer size, kernel
bandwidth) and temporal (e.g., time unit, duration of exposures) parameters in delineating the
MCU (Smith, Foley, and Panter 2019; Yi et al. 2019). To further mitigate the UGCoP, recent
review studies have recommended generation of multiple area-based environment attributes
based on several combinations of spatial (e.g., buffer size, kernel bandwidth) and temporal (e.g.,
hour, day, weekday) parameters for use in sensitivity tests to understand the influence of these
choices on obtained exposure-response effect estimates, further shedding light on the “true
causally relevant” spatial and temporal extent of the exposure (Lee and Kwan 2019; Yi et al.
2019). Additionally, these reviews have also recommended that future studies consider durations
of exposure in assessments of BE (Lee and Kwan 2019; Yi et al. 2019). However, few GPS-
based studies have been able to adopt these recommendations to incorporate daily mobility and
start to tackle the UGCoP in their analyses.
1.3.3. Within-Person Effects of BE Exposures
In addition to study biases, most often, past studies assessed BE exposures (i.e.,
greenness) and PA outcomes (i.e., walking minutes) at the person-level (Smith, Foley, and Panter
2019). The person-level analysis offers insights of effects of inter-individual (i.e., between-
person) variations in BE exposures on PA outcomes. For example, whether living in a walkable
downtown is associated with higher walking activity compared to living in an automobile-
dependent suburb. However, repeated measures designs where individuals are assessed at
multiple points in time can address questions related to the effects of intra-individual (i.e.,
within-person or day-to-day) variations in exposures due to the impact of daily mobilities of
study participants on outcomes. For instance, whether the variations in exposures to parks and
open space within daily activity spaces of individuals are associated with their day-level PA

7
outcomes. Therefore, to provide a more comprehensive understanding on contextual influences
of BE on PA behaviors, it is very important to examine the effects of both within and between
person variations in exposures to BE features and characteristics on their day-to-day PA
outcomes (Dunton 2018). However, very few studies have examined this relationship to date.
1.3.4. Moderating Mechanisms in Behavioral Pathways
Lastly, several studies (e.g., Diez Roux and Mair 2010; Sallis et al. 2009; McNeill,
Kreuter, and Subramanian 2006; Perez et al. 2016; and Larsen et al. 2013) have examined the
moderating effect of neighborhood social environment (e.g., social cohesion, crime) on
relationships between BE and PA behaviors. For instance, a study on Hispanic women shows
perceived safety from crime modified the associations between sidewalk conditions and outdoor
MVPA minutes (Perez et al. 2016). Nevertheless, most of the studies of this type only examined
the residential neighborhood environment and did not investigate possible moderators of the
effects of BE exposure on PA behaviors encountered within activity spaces or in non-residential
contexts. Moreover, many previous studies have linked temporal factors (e.g., weekdays versus
weekend days, pregnancy and postpartum periods) to pregnant women’s PA behaviors
(Borodulin, Evenson, and Herring 2009; da Silva et al. 2019; Jenum et al. 2013; Renault et al.
2012; Schmidt et al. 2006; Sinclair et al. 2019). For example, a study of 1,482 US women
reported that pregnant women’s PA decreased over pregnancy periods (Borodulin et al. 2008).
Another US study of 688 pregnant women showed how PA rebounded during postpartum
periods compared to pregnancy (Borodulin, Evenson, and Herring 2009). Therefore, it is
important to understand how these temporal factors can potentially moderate the effect of BE
exposures on PA behaviors, especially in low-income, Hispanic health disparity populations.

8
1.4. Mobile Sensing in PA Research
The recent development of mobile sensing technology such as Global Positioning System
(GPS), accelerometry and ecological momentary assessment (EMA) offer new opportunities to
collect novel datasets to fill some of aforementioned gaps. Among them, GPS provides users
with real-time geo-locations, which can then be imported into Geographic Information Systems
(GIS) to generate individual activity spaces (i.e., MCUs) and measure BE exposures within them
(Yi et al. 2019). The accelerometer measures the physical acceleration experienced by an object
and can be worn by study participants to record their epoch-level steps taken and intensity of PA
(Troiano et al. 2008). Lastly, EMA is an intensive survey technique that allows participants to
self-report on symptoms, affect, behavior, cognition, and environment contexts close in time to
experience (Moskowitz and Young 2006). Mobile phone-based EMA applications can be
programmed at a customized daily frequency over study periods to collect both data of self-
reported environment contexts and energy-balance behavioral outcomes (Dunton 2017).
Past studies have combined GPS, accelerometry and EMA in answering a variety of
questions regarding effects of BE exposures on PA outcomes. To start, epoch-level GPS and
accelerometry data can be aligned by timestamps to generate time-aligned GPS accelerometry
(TAGA) data so that minute-level relationships between BE and PA behaviors can be examined
(Yi et al. 2019). For example, one study used TAGA dataset to study minute-level influences of
greenness exposure level on PA outcomes (Almanza et al. 2012). In addition, GPS and
accelerometer data can be linked with EMA data by timestamps and aggregated at day- or
person-levels to examine effects of BE exposures on daily or personal PA outcomes. For
example, a study might generate activity spaces from daily GPS tracks and calculate densities of
sport facilities within these spaces to examine associations between daily accessibility to sports

9
facilities and day-level energy expenditure. In addition, the same study might also measure day-
level energy expenditures using movement counts from accelerometry (Shrestha et al. 2019).
However, no studies on pregnant women have integrated GPS technology with daily movement
patterns collected by accelerometry to derive daily activity spaces and assess BE exposures
within these activity spaces to examine the effects of BE exposures on PA outcomes.
1.5. Conceptual Framework and Dissertation Outline
This dissertation research aimed to address aforementioned gaps by investigating
associations between daily time activity and mobility patterns, BE exposures, and PA behaviors
in a group of predominantly low-income Hispanic women of childbearing age, while accounting
for potential analytical and inferential errors and biases and considering potential moderation
mechanisms in the BE-PA relationship. The topics of this dissertation research are organized
within the conceptual framework summarized in Figure 1.1. This framework is adapted from
Sallis’s socio-ecological model framework that describes the interplay of the BE, PA behavior,
and obesity (Sallis et al. 2012). The new model focuses on individual mobility in examining
effects of BE exposures on PA outcomes. In this model, women’s daily mobility patterns (e.g.,
time spent at home, daily path areas) at pre- and postpartum periods are hypothesized to be
important determinants of their dynamic daily exposures to BE characteristics such as parks and
open space access, walkability, which in turn may influence women’s PA behaviors (e.g., sports
activities, walking to the bus stop) through multiple behavioral pathways (e.g., the visual
exposure to a park along the sidewalk, the past experience of visiting a nearby exercise facility).

10

Figure 1.1. The conceptual framework within which topics of this dissertation research are organized.

11
Building on this conceptual framework, in three sequential parts (i.e., three empirical case
studies), this dissertation examines the associations between women’s daily time-activity
mobility patterns, BE exposures, and PA behaviors during pregnancy and the early postpartum
period. Among them, the first study (Chapter 2) tackles the “Daily Mobility Pattern” part of the
framework by examining changes in women’s daily time-activity and mobility patterns across
pregnancy and the postpartum period, and their potential individual and neighborhood level
determinants. The second study (Chapter 3) tackles the “Built Environment Exposure” part of the
framework by examining women’s dynamic daily exposures to BE characteristics derived using
activity space methods and evaluated the BE exposure measurement error using static compared
to dynamic methods. The third study (Chapter 4) tackles the “Behavioral Pathways” and
“Physical Activity Behaviors” parts by examining the associations of dynamic daily BE
exposures and PA outcomes and tested potential moderators in these associations. Since all three
studies are stand-alone manuscripts, each of these chapters includes an introduction as well as
related work, methods, results, discussion, and conclusions. The fifth and final Chapter provides
a summary of major findings, the contribution of this work to the current literature and
limitations, future research directions, and major takeaways.

12
Chapter 2 Time-Activity and Daily Mobility Patterns during Pregnancy and
Early Postpartum
Pregnant women’s daily time-activity and mobility patterns determine their environmental
exposures and related health effects (Balakrishnan et al. 2015; Blanchard et al. 2018; Dadvand et
al. 2012; Hannam et al. 2013; Nethery, Brauer, and Janssen 2009; Zhu et al. 2019). Most studies
ignore these and assess pregnancy exposures using static residential measures exposures (Porter
et al. 2019; Nichani et al. 2016; Richardsen et al. 2016). This chapter applied 4-day continuous
geo-location monitoring in 62 pregnant Hispanic women during the pregnancy and early post-
partum periods to derive their daily time-activity and mobility patterns and examine whether
these patterns differed by individual sociodemographics, temporal factors, and neighborhood
characteristics.
2.1. Related Work
Chemical and physical environment exposures including air pollution, lack of access to
parks and green space, and low walkability, have been associated with poorer health behaviors
and increased risk of health problems in pregnant women and their offspring (McEachan et al.
2016; Porter et al. 2019; Leggett et al. 2018). However, prior studies examining the influence of
environmental exposures on health behaviors (e.g., PA, diet) and disease outcomes (e.g., asthma,
obesity, diabetes) in pregnant women have mainly applied the residence-based assessment
approach (i.e., measuring physical environment features and characteristics at or near residences)
to estimate individual, personal exposures (Porter et al. 2019; Nichani et al. 2016; Richardsen et
al. 2016). This approach assumes outdoor environmental exposures around home residences are
surrogates of daily “true causally relevant contexts” (true contextual units or TCUs) that
influence behaviors or outcomes of interest (Robertson and Feick 2018). Nonetheless, this

13
assumption has two limitations – it assumes participants always stay within their residential
neighborhoods when they might be highly mobile daily or they may change their residential
addresses during and after pregnancy, and that all exposures occur in outdoor environments
whereas quite often they occur mainly indoors or in transit, resulting in exposure
misclassification or measurement error (Kwan 2012; Delmelle et al. 2022).
Indeed, past studies on time-activity and mobility patterns (hereafter referred to as time-
activity patterns) of women during pregnancy have validated these concerns (Zhu et al. 2019;
Ouidir et al. 2015; Blanchard et al. 2018; Payam Dadvand et al. 2012).

For example, a study in
Shanghai, China, reported that pregnant women on average spent over a third of their time in
work locations within three-day observation periods during the 2
nd
trimester (Zhu et al. 2019).
Another study conducted in France reported a median of almost 12 non-home h/day for pregnant
women during a 3-week observation period in the 1
st
trimester (Ouidir et al. 2015). As a result,
the failure to capture or model the non-home contribution to environmental exposures in past
studies might lead to under- or over-estimation of exposures and therefore mask their true
relationships with health behaviors or outcomes (Yi et al. 2019). Very few studies of pregnant
women have incorporated time-activity patterns into environmental exposure assessments largely
due to either feasibility or burden-related challenges with tracking or capturing these patterns at
high spatiotemporal resolutions in large population-based studies.
Moreover, unlike other populations, pregnant women have increased demands to prepare
for childbirth, increased fatigue, difficulty physically moving around, and poor sleep, which
might influence or lead to dramatic variations in their time-activity patterns across the pregnancy
and postpartum periods (Varshavsky et al. 2020). For example, a Canadian study on time-activity
patterns of pregnant women has reported that more time was spent at home during the 3
rd

14
trimester of pregnancy compared to the 1
st
trimester (Nethery, Brauer, and Janssen 2009).
Another US study has found that in-vehicle travel times were longer during the early stages
compared to later stages of the pregnancy (Wu et al. 2013). While very few studies have
examined changes in time-activity patterns of women across pregnancy, to the best of my
knowledge, none have extended the investigation to the early postpartum period. Given that the
timing of the environmental exposures during these critical windows of time could have different
effects on fetal development, early childhood and postpartum maternal health (Porter et al. 2019;
Nichani et al. 2016; Richardsen et al. 2016), it is important to better understand time-activity
patterns and how they might change over pregnancy and early postpartum periods.
Although limited, a small number of studies have implemented various approaches to
collect mobility data for pregnant women and integrate time-activity patterns into exposure
assessments. Among them, most have relied on self-reported mobility surveys or diaries
(Balakrishnan et al. 2015; Blanchard et al. 2018; Dadvand et al. 2012; Hannam et al. 2013;
Nethery, Brauer, and Janssen 2009; Zhu et al. 2019). This approach is relatively cost-efficient
with low technical barriers and thus may suit population-based studies with large sample sizes;
however, the subjective nature of self-reported survey data also makes the approach prone to
recall bias and measurement error. Additionally, it is difficult to collect highly space- and time-
resolved data using diaries or surveys. Recently, a growing body of research has started to apply
GPS technology to objectively capture the mobility of participants (Ouidir et al. 2015; Nethery,
Brauer, and Janssen 2009; Ha et al. 2020). The geolocation coordinates collected from the GPS
device can be imported into the GIS software, in which spatial clusters and trip detection
algorithms can be applied to derive time-activity (i.e., time spent in specific contexts, and
indoor/outdoor microenvironments) and mobility (i.e., modes and durations of trips) patterns of

15
study participants (Thierry, Chaix, and Kestens 2013; Kirby, Delmelle, and Eberth 2017). Also,
GPS data and these derived time-activity patterns can be integrated with fine-scale (e.g., 10-s)
personal air pollution monitoring or other wearables data to construct highly individualized,
contextualized, and space-time resolved exposure models (Ouidir et al. 2015). Finally, activity
spaces derived from GPS data can be integrated with other geospatial data layers (e.g., crime,
parks and open space, walkability scores) to understand actual exposure to BE (Yi et al. 2019).
To address the above gaps, this study combines GPS technology and geospatial analysis
to describe time-activity patterns in a subset of 62 low-income, Hispanic women participating in
the Maternal and Development Risks from Environmental and Social Stressors (MADRES)
cohort study, during 4-day observation periods in the 1
st
and 3
rd
trimesters of pregnancy and at 4
to 6 months postpartum. By analyzing highly time-resolved (i.e., 10-s epoch) GPS data from the
customized smartphone app for MADRES study (madresGPS), this study aimed to answer the
following questions:
1) What are typical time-activity (i.e., time spent in multiple contexts, and
indoor/outdoor microenvironments) and mobility patterns (i.e., trips performed, their duration,
and mode) of women during pregnancy and during the early postpartum period?
2) Do daily time-activity and mobility patterns change over time, across pregnancy and
early postpartum periods?
3) Do temporal (e.g., weekdays versus weekend days), individual sociodemographic, and
residential neighborhood factors explain some of the variation in these patterns?
This study hypothesized that women’s time spent at their home residences would
increase, and time spent in non-home contexts and in transit would decrease as pregnancy
progresses from the 1
st
to the 3
rd
trimesters, and such trends may continue into the postpartum

16
period. Moreover, this study hypothesized time-activity and mobility patterns may differ by other
temporal, individual sociodemographic, and residential neighborhood factors.
2.2. Methods
2.2.1. Study Design
Data for this study comes from the Real-Time and Personal Sampling sub-study of the
MADRES cohort (Bastain et al. 2019). This study uses an intensive longitudinal, observational
panel study design and examines the daily effects of environmental exposures and social
stressors on maternal pre- and post-partum obesity-related biobehavioral responses (O’Connor et
al. 2019). A total of 65 Hispanic, women with lower incomes, were drawn from the larger
MADRES prospective cohort study which recruited participants from prenatal care providers in
Los Angeles, CA, serving predominantly medically-underserved populations (Bastain et al.
2019). To be eligible for the larger MADRES study, a participant needed to be 18 years old with
a singleton pregnancy and be at less than 30 weeks’ gestation at time of recruitment. In addition,
participants who were HIV positive, had physical, mental, or cognitive disabilities that prevented
participation, or were currently incarcerated were excluded from the study. Recruitment of 65
Hispanic women occurred on a rolling basis between 2016 and 2018 from one county hospital
prenatal clinic (N=16) and one non-profit community health clinic (N=49). Additional eligibility
criteria for this sub-study are described in further detail in a separate manuscript (O’Connor et al.
2019). The University of Southern California (USC) Institutional Review Board approved all
study procedures and participants signed an informed consent before enrolling into the study.

17
2.2.2. Data Collection
2.2.2.1. GPS Based Location Information
GPS data were continuously collected from participants at 10-s intervals for four days
(two weekdays and two weekend days) during the 1
st
and 3
rd
trimester of pregnancy and at 4-6
months postpartum (O’Connor et al. 2019). MADRES researchers designed a custom
smartphone application (madresGPS app) for Android operating systems to collect highly
resolved and encrypted GPS data (O’Connor et al. 2019). Study coordinators configured the
application on dedicated study smartphones (Samsung MotoG phone) to gather geographic
coordinates and geolocation/motion metadata (O’Connor et al. 2019). The application logged
instantaneous GPS location and sensor data every 10 s from the smartphone’s multiple built-in
location finding features (cell tower triangulation, WiFi networks, and GPS) and motion sensors.
Along with the timestamp, metadata such as the number of satellites in use/view, geolocation
accuracy, source of GPS, velocity (if GPS source), and network connection status (if network
source) were recorded (O’Connor et al. 2019).
2.2.2.2. EMA data
EMA data were self-reported through a commercially available application (MovisensXS
app) built for Android operation systems, which was pre-installed on the same study phone used
to collect GPS data. The EMA survey was prompted at random times during each five pre-
specified sampling windows (i.e., wake-up to 10 a.m.; 11 a.m. to 1 p.m.; 2 p.m. to 4 p.m.; 5 p.m.
to 7 p.m., and 8 p.m. to bedtime) within the same four-day time GPS data collection windows
during the three study periods (O’Connor et al. 2019). Survey questions included physical and
social contexts at the prompt time, current affective and physical feeling states, current perceived

18
stress, and past 2-h exposure to a list of daily stressors. The complete list of EMA survey
questions is described in further detail in a separate manuscript (O’Connor et al. 2019).
2.2.2.3. Retrospective surveys and medical record abstraction
Sociodemographic data including maternal height and weight race/ethnicity, enrollment
age, education, parity, and country of origin were assessed in prenatal interviewer-administered
questionaries with the women. Weight and height were also measured at study visits.
Retrospective surveys were completed at various study timepoints to gather residential and
occupational histories and assessed psychosocial stressors. Working status was also collected via
questionnaires in the 1
st
trimester, 3
rd
trimester, and 6 months postpartum and perceived
neighborhood cohesion and safety score was gathered in the 2
nd
trimester (chosen to represent
pregnancy) and 6 months postpartum questionnaires. Additionally, residential locations at
screening were geocoded and used to generate residential neighborhood characteristics in this
work.
2.2.3. Data processing
2.2.3.1. GPS processing
The major processing steps of raw GPS data are described in Figure 2.1. In total, this study
collected 6,948,118 GPS observations for 62 of the 65 participants. Raw observations collected
outside of the 4-day designated monitoring period (during device set up and return) were
dropped (Ndropped=1,893,013). Then, this study dropped a small number of observations with
erroneously logged zero values of latitude and longitude (Ndropped=28,848). After that, this study
devised a logic to drop the least accurate source of geolocation data for every 10-s epoch when
two sources of data (GPS/Network) were available as follows. This logic was informed by
comparing the time-series of GPS versus network source coordinates in relation to the

19

Figure 2.1. Global Positioning System (GPS) pre-processing steps of Maternal And
Developmental Risks from Environmental and Social Stressors (MADRES) real-time and
personal sampling study GPS data.
1
Data collection dates included two weekdays and two weekend days.
2
Three periods were first trimester, third trimester, and four-to-six months postpartum.
3
Network source included observations recorded by WiFi and cellular networks.
4
Signal loss scenarios were defined as ≥ 1 min time windows with same timestamps.

5
Outliers were defined as observations with a distance > 450 m from the median latitude/longitude coordinates
(corresponding to the maximum realistically possible distance moved in 10 s based on a speed of 45 m/s or 100
mph) and replaced with the median coordinates within the moving window.

20
participants’ potential movement in space and time. Based on preliminary analyses, the GPS
source usually exhibited the fastest update frequency compared to the network especially when
individuals were mobile. When individuals appeared to be stationary (i.e., at home), both sources
seemed to be frequently updating, but the network source generally exhibited higher accuracy.
Of particular note, participants were asked to connect study smartphones to their home WiFi
networks, when possible, to complete study-related EMA surveys on the same smartphones,
resulting in a high likelihood of phones connecting to home WiFi networks (and thus network
source geolocation data being available often when stationary/at home) during this study
(averaging 12.8% per observation day). When both GPS and network sources were available,
they were examined, and the less accurate source was dropped (Ndropped=542,213).
In circumstances when the signal from either GPS or network source was lost for ≥1 min
(i.e., signal loss scenario), the app was designed to log the latest known position for that source
along with the latest update (or confirmation) time, both of which will be repeated and will not
change for the duration of time the signal was lost. Once signal loss scenarios were identified
(per source of data), the update frequency and positional accuracy of the geolocation data from
both sources were compared and the less accurate source was dropped (Ndropped=81,987). For
time windows when either the GPS or network source was updating (real sensor data logging
timestamp changed) but the other was not because of signal loss, the connected source was kept.
Then, under circumstances when both sources lost signal, the one that indicated no movement
(no change in latitude and longitude) from the previous to the next interval of time when signal
was available was dropped.
Next, this study rounded timestamps to the nearest 10 s and retained the first observation
within a 10-s window (Ndropped=53,530). This step was performed to allow us to align and

21
integrate GPS data with other simultaneously collected accelerometry and personal air pollution
exposure monitoring data (in subsequent, ongoing analyses). In addition, it also ensures roughly
similar temporal spacing and density of GPS data per participant to enable between-person
comparisons of environmental exposures derived using this GPS data and the kernel density
algorithm (Jankowska et al. 2017).
Finally, a moving median filter was applied to remove outliers in windows of
approximately 1-min duration (7 observations at a 10-s epoch) to correct extreme outliers that
might occasionally be present in the data (Wang, Gao, and Juan 2017). Outliers were defined as
observations with a distance >450 m from the median latitude/longitude coordinates
(corresponding to the maximum realistically possible distance moved in 10 s based on a speed of
45 m/s or 100 mph) and were replaced with the median coordinates within the moving window.
The final processed dataset consisted of 4,375,774 observations across 552 person-days.
Throughout the GPS data processing, this study created flags to indicate data quality or
identify records affected by any assumptions or decisions made, which were used to inform
sensitivity analyses. For example, this study created day-level GPS data completeness flags (i.e.,
≤6 h, ≤10 h, ≤16 h), which were then used to evaluate whether time-activity and daily mobility
patterns were sensitive to day-level GPS data completeness. This study also created flags
indicating confidence in whether an individual likely stayed at the logged location or moved
during signal loss windows (see Appendix A). These flags were based on the plausibility of the
distance moved within the window and total duration of the window. More specifically, higher
confidence levels were assigned to signal loss windows with shorter duration (e.g., ≤120 min)
and more reasonable distances traveled (e.g., ≤45 m/s times the time elapsed between the last
known location before signal loss and the new location after signal loss). For analyses in this

22
study, signal loss windows were removed (Ndropped=523,112) that I either could not make a
judgement on (i.e., with no distance/duration data) or had extremely low confidence (e.g.,
distance traveled >45 m/s times the time elapsed, >120 min in duration) on whether a participant
likely remained in the same position when the signal was lost.
2.2.3.2. Stay-trip detection
The processed 10-s GPS data were imported into geographic information system software
ArcGIS 10.8 (Esri, 2020) to first identify trips and stays and then classify stays based on their
spatial contexts and indoor/outdoor microenvironment. Figure 2.2 describes the steps to process
GPS tracks for each person and study period combination (i.e., 4-day GPS tracks were treated as
one sequential time-series). In order to identify trips and stays, the “Activity Place Detection
Algorithm” ArcGIS toolbox developed by Thierry, Chaix, and Kestens (2013) was used, which
builds a kernel density surface (50 m bandwidth or search radius) from GPS points and extracts
local maxima from the surface as candidates for classification as stays. In comparison to methods
that analyze data points sequentially, the kernel-based method has been shown to have better
global performance (i.e., better agreement between number of stops detected versus true stops),
higher spatial accuracy (i.e., shorter Euclidean distance between a detected stop and the true
stop), and lower sensitivity to bandwidth choices (i.e., 50, 100 m) (Thierry, Chaix, and Kestens
2013). Minimum duration for a stay candidate to become a stay was ≥5 min, and two
consecutive stay candidates within proximity to each other needed to be separated by at least a
≥5 min timespan to be kept as separate stays. After stays and their respective start and stop times
were detected, GPS points recorded between two consecutive stays were connected into trips
(path trajectories) by sequences of timestamp and smoothed (snapped to road networks), and
their start and stop times were recorded. This essentially means that stays also act as trip origin

23

Figure 2.2. Geoprocessing steps executed to detect stays from Global Positioning System (GPS) data, classify their contexts based on
land use, and their indoor/outdoor microenvironments based on building footprints.

24
and destination points when trips occur. A minimum duration of ≥5 min was needed for a loop
path trajectory (i.e., one that started and ended at the same location) to be kept as a trip.
2.2.3.3. Context classification of stays
This study next classified the stays (i.e., trip origins and destinations) into one of seven
spatial contexts (home residential, non-home residential, commercial and services, parks and
open space, schools and public facilities, industrial and office spaces, and other). Spatial
parameters and data sources used in context classification are fully documented in Appendix B.
The stay with the longest duration in a study period (4-day monitoring period in 1
st
and 3
rd

trimesters, and at 4-6 months postpartum) was designated as the residential home context given
participants might have changed residence or lived with family or relatives across study periods.
Non-home contexts were classified based on their spatial relationships with the Southern
California Association of Governments existing land use (2016) data (see Appendix B). A 15 m
buffer was applied to existing land use boundaries to account for potential combinations of
indoor/outdoor activities within a stay and considering the average width of sidewalks in urban
Los Angeles. Additionally, an indoor/outdoor microenvironment was assigned to each stay point
by examining its spatial relationship with Los Angeles County building footprints (2014). A 1 m
buffer was applied to existing building footprints to account for scenarios when indoor activities
occurred mainly in the corners of the building (i.e., corner apartments, stairwells, laundry
rooms), resulting in a detected stay point that fell outside the building footprint polygon, which
could then be misclassified as outdoor.
2.2.3.4. Missing GPS data imputation using home context
Home residential locations detected via the stay-trip detection algorithm were then used
to impute some of the missing records in the processed GPS data. More specifically, participants

25
self-reported their sleep and waking times prior to each study period to help configure suitable
timing and frequencies for the EMA survey. This sleep and wake time data was used to divide
each four-day study period into day (from waking to sleep time in a data collection day) and
night (from sleep time in a data collection day to the waking time on the next day) windows. For
night windows, identified visit-level home location was used to impute missing data (if ≥ 60 min
of GPS logged data that was within ≤100 m of the home location). If this rule was not met, the
median coordinates logged during the night were used to fill in any remaining missingness that
night. If no GPS data was available during the night, then imputation was not attempted. As for
day windows, this study filled in missing records with home coordinates when available if the
day was identified as a home day (i.e., all EMA survey prompts within the day reported current
physical context as either “Home-Indoor” or “Home-Outdoor”). The entire workflow of the
missingness imputation process is documented in Appendix C. The imputed GPS records
(N=306,915) were classified as “home-residential” and merged with processed epoch-level GPS
data to produce a final time-activity pattern dataset that records location coordinates and contexts
of each stay, its start and stop time, as well as method of classification (i.e., via algorithm or
imputation). In addition, flags were created which labelled days with < 6 h of GPS data (post-
imputation) as invalid days.
2.2.3.5. Trip mode detection
A trip mode classification algorithm was also applied (Figure 2.3) to classify all trips into
either a walking- or vehicle-based mode. Both distance-based trip speed (i.e., sum of Euclidean
distances of consecutive GPS records in a trip divided by duration of time elapsed) and total
distance traveled (i.e., sum of Euclidean distances of consecutive GPS records in a trip) were
considered in the decision-making process. Previous studies have reported a walking speed range

26

Figure 2.3. Geoprocessing steps to detect trips and classify their modes based on mean and
standard deviation of Global Positioning System (GPS) observations in trips.
1
Trips start time was identified as the end of previous stay and trip end time was identified as the start of the next
consecutive stay.
2
Epoch-level distance-based speed (vtrip) was calculated by dividing the Euclidean distance traveled (dtrip)
between two consecutive epochs with time elapsed (ttrip).

of 1.00–1.8 m/s for women during pre- and post-pregnancy periods (Byrne et al. 2011; Gilleard
2013; Loh et al. 2019). In this study, a relatively high threshold of 2 m/s (4.5 mph) was treated as

27
the theoretically possible maximum walking speed for women to account for GPS data noise in
areas that might obstruct or interfere with GPS signals (e.g., neighborhoods with multi-level
residences or abundant and dense tree canopies). Given similarities in average speed between a
true walking scenario and a slow driving one that could occur during Los Angeles rush hours or
when passing through areas with frequent traffic lights, a condition was added such that a trip
required a standard deviation of speed that was smaller than 1 m/s to be classified as walking-
based. This criterion was based on observed patterns in the data showing that walking trips
typically have a much smaller standard deviation in speed than slow driving trips comprising
sudden acceleration, deceleration, and frequent stops. Furthermore, for a trip to be vehicle-based,
it also needed to exceed the maximum possible distance a human can travel via walking (i.e., 3
m/s x trip duration). Lastly, for a limited number of trips (N=99) that exhibited patterns with
multiple modes (e.g., walking to the parking lot, driving, riding the metro and walking), the
criteria was relaxed and only used the mean speed to determine the primary trip mode (i.e.,
vehicle-based: ≥2 m/s; walking-based: <2 m/s). For these trips, lower confidence was assigned to
their detected trip modes so that they could be excluded for sensitivity analyses purposes.
2.2.4. Statistical Analysis
2.2.4.1. Descriptive analysis
Mean, medians, proportions, or standard deviations for covariates and time-activity and
daily mobility outcomes were calculated for the 1
st
and 3
rd
trimesters of pregnancy and the 4-6
months postpartum periods. The number of stays were summarized by context and
microenvironment and aggregated into day-level time-activity patterns (min/d at each spatial
context and within indoor/outdoor microenvironments). Meanwhile, the number of trips were
summarized by trip mode and aggregated into day-level mobility patterns (min/d and N/d in trip

28
of vehicular or pedestrian mode). Non-valid days (<6 h of GPS data) were eliminated to reduce
potential biases of estimating day-level outcomes.
2.2.4.2. Generalized mixed effects models
To account for the interdependency of the nested data structure in the current study
(Level 1-days nested within Level 2-persons), generalized linear mixed-effects models
(GLMMs) with participant-level random intercepts were used. Additionally, negative binomial
family functions were fitted because outcomes had over-dispersed distributions, which log-
transformed the outcome during analyses. Lastly, a zero-inflated portion was added to all models
due to the presence of excessive zero values except for the model that examined min/d at home
residential context. These zero-inflated models estimated a participant’s probability of having
zero min/d at a given context (not visiting the context) or at a given trip mode (not performing
the trip).
2.2.4.3. Model building strategy
For each outcome, GLMM models were first fitted to test whether the derived time-
activity and mobility patterns changed over time during pregnancy and postpartum (hereinafter
referred to as the Base GLMM model). Then, individual sociodemographic, neighborhood, and
additional temporal factors were further included to explore whether these factors can further
explain these time-activity and mobility patterns. This was accomplished by first entering all
other covariates to construct the fully-adjusted model if the univariate analysis (one covariate at a
time) reported a p<.1 (hereinafter referred to as the Fully-adjusted GLMM model). Lastly,
covariates with reduced explanatory power (i.e., p-value became >.05 in the fully adjusted
model) were dropped in the final model to ensure model parsimony (hereinafter referred to as the
Final GLMM model). Following the recommended practice, covariates were kept the same for

29
the count and zero parts of each zero-inflated GLMM (Brooks et al. 2017). Additionally, the day-
level total GPS data collection hours were always included as a covariate to adjust for the
varying amount of GPS data collected possibly due to individual device wearing behaviors or
other factors.
2.2.4.4. Model covariate selection
A list of temporal factors, individual-level sociodemographic, and neighborhood-level
characteristics were included as covariates in the models. Past studies have associated these
covariates with time-activity patterns of pregnant women (e.g., Abatzoglou 2013; Blanchard et
al. 2018; P. Dadvand et al. 2012; Hannam et al. 2013; Nethery, Brauer, and Janssen 2009;
Nethery et al. 2008; Wu et al. 2013; Zhu et al. 2019). Temporal factors included weekend versus
weekday (weekend=1), daily average temperature in degrees Celsius, and study period (1
st

trimester, 3
rd
trimester [reference group], and 4-6 months postpartum). The 3
rd
trimester was
chosen as the reference group since most prior pregnancy studies examining the relationship
between environmental exposures and maternal or birth outcomes usually characterize
environmental exposure based on location at a single point in time late in pregnancy or at
delivery (Porter et al. 2019; Banay et al. 2017; Pennington et al. 2017). Since the 3
rd
trimester is
closest to infant delivery, this study contrasted changes in time-activity and mobility patterns
over time relative to this commonly used assumption. Individual sociodemographic
characteristics from MADRES questionnaires were included, including age, education (less than
or equal to high school diploma), marital status (married/living together,
single/divorced/separated/widowed, or declined to answer/missing response), birth country
(foreign- versus US-born), parity (first born versus second or greater birth) as well as
employment status at each period. Body Mass Index (BMI) categories were also calculated

30
(recoded as normal versus overweight/obese) based on height and weight measured during pre-
natal visits. Additionally, individual-level neighborhood cohesion and safety scores were
included from questionnaires administered during pregnancy and postpartum (Sampson,
Raudenbush, and Earls 1997). Neighborhood characteristics were assigned to participants’
residences based on the 2010 census block group boundary within which their home residences
were situated. These included the National Walkability Index Score from the Environmental
Protection Agency (EPA) EnviroAtlas and the Deprivation Index Score from the Neighborhood
Atlas (Kind and Buckingham 2018; Pickard et al. 2015). A full list of covariates measures and
corresponding data sources is documented in Appendix D.
2.2.4.5. Sensitivity analysis
Finally, sensitivity analyses were run by excluding days with <10 h or <16 h of GPS data
to examine the influence of GPS completeness on observed associations, and by replacing study
periods with binary (pregnancy versus postpartum) and continuous time (continuous week from
conception to post-birth) variables, and by testing non-linear (quadratic) terms. The R 4.0.2 (R
Core Team, 2020) and glmmTMB package (version 1.0.2.1) were used for generalized mixed-
effects modeling (Brooks et al. 2017). Exponentiated effect estimates which are interpreted on a
multiplicative scale were reported for all models. Reversed odds ratios (i.e., odds to accumulate
any minutes at a given time activity or mobility pattern outcome) of zero-inflated models were
calculated for easier interpretation.
2.3. Results
2.3.1. Data Completeness
A total of 65 participants were initially enrolled in the study, of which 62 provided at
least one valid GPS observation day (≥6 h of data) across three study periods. Within these 62

31
participants, 35 had at least one valid day in all three study periods; 17 in two of the three
periods; and 10 in one of the three periods. The final analytical sample comprised a total of 552
valid days of GPS data from 62 participants across the 1
st
(N=205 person-days) and 3
rd
trimesters
(N=180 person-days) of pregnancy and 4–6 months postpartum (N=167 person-days). Each
participant provided an average of 8.9 (SD=3.00; Range: 3.00-12.00) valid GPS days across the
three periods. The average number of hours of GPS observations collected on valid days was
21.7 h (SD= 5.00; Range: 6.2-24.00). The average number of hours was highest in the 3
rd

trimester (Mean=22.3h; SD=4.4; Range: 6.5-24.00) followed by the 4-6 months postpartum
period (Mean=21.8h; SD=4.7; Range: 7.00-24.00) and the 1
st
trimester (Mean=21.1; SD=5.6;
Range: 6.2-24.00). Almost half of the valid person-days (49.3%) were weekend days across the
three periods.
2.3.2. Descriptive Statistics of Covariates
Descriptive statistics for the participant- and day-level covariates are shown in Tables
2.1. Participants’ mean age at study entry was 29 years (SD=6.1; Range: 18-45). All the
participants were Hispanic, and more than half were born outside of the US (53.2%). About one-
third (32.3%) had some college or above education, and 80.6% were either married or living
together with their partners at study entry. One in three (36.4%) was employed during the 1
st

trimester compared to 39.6% during the 3
rd
trimester, and 19.6% at 4-6 months postpartum. At
recruitment, 25.8% were pregnant with their first child, 74.2% were overweight or obese
according to their pre-pregnancy BMI. The recruited participants lived in neighborhoods with an
average walkability index score of 14.4 (SD=2.00, Range: 9.3-19; on 1-20 scale; where 1=least
walkable) and average deprivation index score of 6.5 (SD=1.7, Range: 2.00-9.00; on 1-10 scale;
where 1=least deprived). The average neighborhood safety and cohesion score (on 1-5 scale;

32
where 1=least safe and cohesive) self-reported by women was 3.1 (SD=.7, Range: 1.00-4.4) at
the 1
st
and 3
rd
(SD=.7; Range: 1.00-5.00) trimesters, and 3.3 (SD=.9, Range: 1.4-4.8) at 4-6
months postpartum.

Table 2.1. Descriptive statistics of participant characteristics at baseline (a) and
person-day level temporally varying factors by 1
st
and 3
rd
trimesters, and 4-6
months postpartum (b).
(a) Baseline statistics

Overall
(N=62 Participants)
Age at consent (years)

Mean (SD) 29 (6.1)
Median [Min, Max] 28 [18, 45]
Education

High school or less 42 (67.7%)
Some college/Graduate 20 (32.3%)
Marital status

Married/Living together 50 (80.6%)
Single/Divorced/Separated/Widowed 10 (16.1%)
Missing 2 (3.2%)
Acculturation

US-Born Hispanic 29 (46.8%)
Foreign-Born Hispanic 33 (53.2%)
Maternal parity

First-born 16 (25.8%)
Already had child 46 (74.2%)
Pre-pregnancy BMI category

Normal 16 (25.8%)
Overweight/Obesity 46 (74.2%)
Neighborhood Walkability Score
Mean (SD) 14.4 (2.00)
Median [Min, Max] 14 [9.3, 19]
Neighborhood Deprivation Score
Mean (SD) 6.5(1.7)
Median [Min, Max] 7.00 [2.00, 9.00]
Missing 2 (3.2%)

33
Table 2.1. (Cont.)

(b) Temporally varying statistics

1st Trimester
(N=205
person-days)
3rd Trimester
(N=180
person-days)
4-6 Months
Postpartum
(N=167
person-days)
Overall
(N=552
person-days)
Valid GPS observation (h/day)

Mean (SD) 21 (5.6) 22 (4.4) 22 (4.7) 22 (5.00)
Median [Min, Max] 24 [6.2, 24] 24 [6.5, 24] 24 [7.00, 24] 24 [6.2, 24]
Average Daily Temperature (°C)
Mean (SD) 21 (4.2) 21 (4.4) 19 (4.4) 20 (4.4)
Median [Min, Max] 21 [8.00, 31] 21 [9.00, 31] 20 [5.2, 28] 20 [5.2, 31]
Missing 19 (9.3%) 0 (0%) 0 (0%) 19 (3.4%)
Type of day

Weekday 104 (50.7%) 91 (50.6%) 85 (50.9%) 280 (50.7%)
Weekend 101 (49.3%) 89 (49.4%) 82 (49.1%) 272 (49.3%)

1st Trimester
(N=55
Participants)
3rd Trimester
(N=48
Participants)
4-6 Months
Postpartum
(N=46
Participants)
Overall
(N=149
Participants)
Employment status

Unemployed 35 (63.6%) 28 (58.3%) 29 (63.0%) 92 (61.7%)
Employed 20 (36.4%) 19 (39.6%) 9 (19.6%) 48 (32.2%)
Missing 0 (0%) 1 (2.1%) 8 (17.4%) 9 (6.0%)
Neighborhood Cohesion and Safety Score

Mean (SD) 3.1 (.7) 3.1 (.7) 3.3 (.9) 3.1 (.8)
Median [Min, Max] 3.00 [1.00, 4.4] 3.1 [1.00, 5.00] 3.2 [1.4, 4.8] 3.0 [1.00, 5.00]
Missing 3 (5.5%) 0 (0%) 8 (17.4%) 11 (7.4%)

Note: BMI = Body Mass Index, GPS = Global Positioning System. SD = Standard deviation.
2.3.3. Descriptive Statistics for Time-Activity and Daily Mobility Patterns
2.3.3.1. Time-Activity Patterns
The descriptive statistics for derived time-activity patterns (N stays and min/day by
context and indoor/outdoor microenvironment) are shown in Tables 2.2 and 2.3. Overall, 2,621
stays were detected from 552 valid GPS person-days across three study periods (Table 2.2). The

34
Table 2.2. Summary of total number of visits to multiple spatial contexts and indoor/outdoor microenvironments and total number of
pedestrian and vehicular trips made.

1st Trimester
(Nstay=947; Ntrip=682)
3rd Trimester
(NStay=914; Ntrip=692)
4-6 Months
Postpartum
(NStay=760;
NTrip=551)
Overall
(NStay=2,621;
NTrip=1,925)
N(%) N(%) N(%) N(%)
Spatial contexts
Home residential 412 (43.5%) 363 (39.7%) 337 (44.3%) 1,112 (42.4%)
Non-home residential 64 (6.8%) 60 (6.6%) 79 (10.4%) 203 (7.7%)
Commercial and
Services
281 (29.7%) 283 (31.0%) 193 (25.4%) 757 (28.9%)
Industrial and Office
Spaces
84 (8.9%) 105 (11.5%) 64 (8.4%) 253 (9.7%)
Schools and Public
Facilities
52 (5.5%) 61 (6.7%) 57 (7.5%) 170 (6.5%)
Parks and Open Spaces 22 (2.3%) 17 (1.9%) 12 (1.6%) 51 (1.9%)
Other 32 (3.4%) 25 (2.7%) 18 (2.4%) 75 (2.9%)
Indoor/outdoor
microenvironment

Home Indoor 363 (38.3%) 336 (36.8%) 302 (39.7%) 1,001 (38.2%)
Non-Home Indoor 220 (23.2%) 253 (27.7%) 168 (22.1%) 641 (24.5%)
Home Outdoor 49 (5.2%) 27 (3.0%) 35 (4.6%) 111 (4.2%)
Non-Home Outdoor 291 (30.7%) 288 (31.5%) 230 (30.3%) 809 (30.9%)
Out of Los Angeles
County
24 (2.5%) 10 (1.1%) 25 (3.3%) 59 (2.3%)
Trip modes
Pedestrian trips 175 (25.7%) 185 (26.7%) 120 (21.8%) 480 (24.9%)
Vehicular trips 507 (74.3%) 507 (73.3%) 431 (78.2%) 1,445 (75.1%)

35
Table 2.3. Day-level summary of time spent in spatial contexts, indoor/outdoor microenvironments, and number of
pedestrian/vehicular trips made.

1
st
Trimester
(N=205 person-days)
3
rd
Trimester
(N=180 person-days)
4-6 Months
Postpartum
(N=167 person-days)
Overall
(N=552 person-days)
Spatial Contexts
Home Residential (h/day)

Mean (SD) 16.8 (6.6) 17.5 (6.6) 17.6 (6.3) 17.3 (6.6)
Median [Min, Max] 18.8 [0, 24.00] 19.5 [0, 24.00] 19.4 [0, 24.00] 19.2 [0, 24.00]
Missing 2 (1.0%) 0 (0%) 1 (.6%) 3 (.5%)
All Non-Home Contexts (min/day)

Mean (SD) 205 (324) 219 (328) 190 (295) 205 (316)
Median [Min, Max] 58.00 [0, 1440] 81.4 [0, 1440] 73.2 [0, 1440] 73.2 [0, 1440]
Non-Home Residential (min/day)

Mean (SD) 51.9 (158) 40.8 (139) 68.7 (164) 53.1 (154)
Median [Min, Max] 0 [0, 1260] 0 [0, 1040] 0 [0, 831] 0 [0, 1260]
Missing 46 (22.4%) 26 (14.4%) 34 (20.4%) 106 (19.2%)
Commercial and Services (min/day)

Mean (SD) 68.2 (109) 84.2 (134) 47.7 (68.9) 67.4 (109)
Median [Min, Max] 16.2 [0, 561] 22.2 [0, 619] 9.50 [0, 349] 16.2 [0, 619]
Missing 40 (19.5%) 25 (13.9%) 28 (16.8%) 93 (16.8%)
Schools and Public Facilities
(min/day)

Mean (SD) 21.1 (71.8) 26.2 (70.8) 23.4 (66.1) 23.6 (69.7)
Median [Min, Max] 0 [0, 480] 0 [0, 517] 0 [0, 521] 0 [0, 521]
Missing 49 (23.9%) 25 (13.9%) 32 (19.2%) 106 (19.2%)

36
Table 2.3. (Cont.)

1
st
Trimester
(N=205 person-days)
3
rd
Trimester
(N=180 person-days)
4-6 Months
Postpartum
(N=167 person-days)
Overall
(N=552 person-days)
Industrial and Office Spaces (min/day)

Mean (SD) 103 (304) 93.2 (269) 72.5 (241) 90.4 (274)
Median [Min, Max] 0 [0, 1440] 0 [0, 1440] 0 [0, 1440] 0 [0, 1440]
Missing 44 (21.5%) 25 (13.9%) 29 (17.4%) 98 (17.8%)
Parks and Open Spaces (min/day)

Mean (SD) 11.8 (55.9) 5.57 (30.1) 8.86 (55.4) 8.73 (48.3)
Median [Min, Max] 0 [0, 384] 0 [0, 275] 0 [0, 517] 0 [0, 517]
Missing 53 (25.9%) 29 (16.1%) 36 (21.6%) 118 (21.4%)

Indoor/outdoor microenvironment
Home Outdoor (min/day)

Mean (SD) 150 (389) 129 (389) 138 (391) 139 (389)
Median [Min, Max] 0 [0, 1440] 0 [0, 1440] 0 [0, 1440] 0 [0, 1440]
Missing 49 (23.9%) 26 (14.4%) 37 (22.2%) 112 (20.3%)
Non-Home Outdoor (min/day)

Mean (SD) 109 (248) 117 (272) 112 (253) 113 (257)
Median [Min, Max] 15.5 [0, 1440] 12.00 [0, 1440] 12.00 [0, 1440] 12.3 [0, 1440]
Missing 40 (19.5%) 23 (12.8%) 24 (14.4%) 87 (15.8%)

Daily Mobility
Trip (min/day)

Mean (SD) 60.2 (73.3) 66.6 (69.4) 64.7 (76.6) 63.7 (73.00)
Median [Min, Max] 40.00 [0, 387] 49.6 [0, 363] 37.8 [0, 351] 44.2 [0, 387]

37
Table 2.3. (Cont.)

1
st
Trimester
(N=205 person-days)
3
rd
Trimester
(N=180 person-days)
4-6 Months
Postpartum
(N=167 person-days)
Overall
(N=552 person-days)
Pedestrian-based Trip (min/day)

Mean (SD) 16.2 (30.8) 17.9 (31.7) 14.9 (29.8) 16.4 (30.8)
Median [Min, Max] 0 [0, 205] 0 [0, 186] 0 [0, 166] 0 [0, 205]
Missing 45 (22.0%) 23 (12.8%) 30 (18.0%) 98 (17.8%)
Vehicular-based Trip (min/day)

Mean (SD) 57.3 (67.6) 58.1 (63.7) 60.9 (72.00) 58.7 (67.6)
Median [Min, Max] 36.3 [0, 372] 41.6 [0, 356] 35.3 [0, 351] 39.3 [0, 372]
Missing 35 (17.1%) 22 (12.2%) 23 (13.8%) 80 (14.5%)
Trip (N/day)

Mean (SD) 3.33 (3.86) 3.84 (3.97) 3.30 (3.61) 3.49 (3.82)
Median [Min, Max] 2.00 [0, 18.00] 3.00 [0, 17.00] 2.00 [0, 16.00] 2.00 [0, 18.00]
Pedestrian-based Trip (N/day)

Mean (SD) 1.09 (1.97) 1.18 (1.75) .876 (1.37) 1.06 (1.73)
Median [Min, Max] 0 [0, 13.00] 0 [0, 8.00] 0 [0, 6.00] 0 [0, 13.00]
Missing 45 (22.0%) 23 (12.8%) 30 (18.0%) 98 (17.8%)
Vehicular-based Trip (N/day)

Mean (SD) 2.98 (2.97) 3.21 (3.32) 2.99 (3.31) 3.06 (3.19)
Median [Min, Max] 2.00 [0, 12.00] 2.00 [0, 15.00] 2.00 [0, 15.00] 2.00 [0, 15.00]
Missing 35 (17.1%) 22 (12.2%) 23 (13.8%) 80 (14.5%)
Notes: SD = Standard Deviation.

38
1
st
trimester (N=947 stays) had larger numbers of different stays detected compared to the 3
rd

trimester (N=914 stays) and 4-6 months postpartum (N=760 stays). Among all stays, 42.4%
(N=1,112) were at home, with an average duration of 17.3 h/day (SD=6.6 h/day). Commercial
and services locations were the most popular destinations (28.9% of all stays; N=757
stays; Mean=1.1 h/day; SD=1.8 h/day) among all non-home contexts, followed by non-home
residential locations, industrial and office spaces, and schools and public facilities, each of which
constituted 5~10% of all stays (Table 2.2). Lastly, women in this panel study rarely visited parks
and open space (1.9% of all stays; N=51 stays; Mean=8.73 min/day; SD=48.3 min/day).
In terms of descriptive trends across the three study periods, the number of visits to
industrial and office spaces, and to commercial and services locations increased from the 1
st

trimester to the 3
rd
trimester but decreased at 4-6 months postpartum. However, women’s visits
to non-home residential places increased at 4-6 months postpartum compared to the 3
rd
trimester
(10.4 versus 6.6% of all stays). Additionally, women’s visits to parks and open space showed a
decreasing trend from the 1
st
to the 3
rd
trimester of pregnancy and onto the 4-6 months
postpartum (2.3, 1.9, and 1.6%, respectively of all stays in these time periods).
Approximately one in three (35.1%) of stays detected across the three study periods occurred in
outdoor microenvironments including locations outside of the home (e.g., porch, lawns,
sidewalks) (4.2% of all stays; Mean=2.1 h/day; SD=3.4 h/d) and at non-home outdoor locations
(e.g., parks, sports venues, sidewalks) (30.9% of all stays; Mean=1.9 h/day;SD=4.3 h/day).
Overall, the 3
rd
trimester had the lowest fraction of stays (39.8%) at home (both indoor and
outdoor) and the highest fraction of stays (27.7%) at non-home indoor microenvironments (Table
2.3).

39
39
2.3.3.2. Daily Mobility Patterns
The summary statistics for derived mobility patterns (N and min/d for trips, and N/d by
trip mode) are reproduced in Tables 2.2 and 2.3. Overall, participants took 1,925 trips over the
duration of the study spread across 552 person-days, one in four of these trips (24.9%) was
pedestrian-based (N=489; Mean=16.4 min; SD=30.8 min). The number of trips made varied
slightly between the 1
st
and 3
rd
trimesters (N=682 versus N=692) and decreased at 4-6 months
postpartum (N=551). This pattern was replicated across all trip modes.
Figure 2.4 shows the most popular trip origins and destinations by mode and purpose. For
pedestrian-based trips (N=489), around 1 in 5 (21.7%, N=103) were between different
commercial and services locations, followed by walking within the same commercial and
services locations (14.5%; N=71). For vehicle-based trips (N=1445), about 2 in 5
(37.8%; N=546) were between home and commercial and services locations, followed by
commuting between different commercial and services locations (18.5%; N=268) and between
home and non-home residential locations (9.0%, N=130).
2.3.4. Base GLMM Results
The base GLMM results examining whether day-level time-activity and mobility patterns
varied across the three study periods are illustrated in Figure 2.5. The odds of visiting
commercial and services locations were 58% lower at 4-6 months postpartum compared to the
3
rd
trimester (OR=.42, 95%CI: .23–.76) (Figure 2.5). No other stay contexts (in terms of
frequency or duration of time spent within them) were significantly different across the three
time periods. These results did not change in sensitivity analyses using days with ≥10 h or ≥16 h
of GPS data, binary (pregnancy versus postpartum) and continuous (week number from
conception to post-birth) time. variables, and non-linear (quadratic) time terms. Moreover, the

40

Figure 2.4. Distributions of top five origin-destination combinations by pedestrian- and vehicle-based trip modes.

41

Figure 2.5. Base Generalized Mixed-Effects Model results of variations in time-activity and daily mobility patterns by 1st and 3rd
trimesters of pregnancy and 4-6 months postpartum.
Note: IRR = Incidence Rate Ratio. OR = Odds Ratio.
1
IRR can be interpreted as: if mothers visit a particular context or perform trips with a particular mode, their min/d spent increase (if IRR>1) or decrease
(if IRR<1), compared to the reference time point (i.e., 3
rd
trimester).
2
OR can be interpreted as: mothers in a time point decrease (if OR<1) or increase (if OR>1) the odds of visiting a particular context or performing trips
with a particular mode, compared to the reference time point (i.e., 3
rd
trimester).

42
odds of staying outdoors and time spent outdoors did not vary significantly across the three study
periods (Figure 2.5). Lastly, in terms of mobility patterns, the odds of taking a vehicular trip
were 56% lower at 4-6 months postpartum compared to the 3
rd
trimester at the day level
(OR=.44, 95%CI: .21–.92) (Figure 2.5). These results did not change in sensitivity analyses
restricting trips to those with ≥ 5% of epoch-level GPS data within trip segments detected. The
odds of performing any trips overall or in pedestrian-mode did not vary by study period in these
base models using min/d spent in trips or N/d trips taken. The full model results of base GLMM
can be found in Appendices E and F.
2.3.5. Final GLMM Results
2.3.5.1. Three Study Periods
The results of the final GLMM exploring whether individual sociodemographic,
neighborhood, and other temporal factors such as weekdays vs weekend days additionally
explained the women’s time-activity and daily mobility patterns are summarized in Tables 2.4,
2.5, and 2.6. All significant results found in the base models examining variation over time
remained in the fully adjusted models for odds of taking commercial and services locations, and
for odds of performing any vehicle-based trip). Additionally, the final GLMM results showed
that when women visited non-home residential locations in the 4-6 months postpartum period,
their mean min/day spent there increased by 83% (Incidence Rate Ratio or IRR=1.83, 95%CI:
1.03-3.25) compared to when they visited this same context in the 3
rd
trimester (see Table 2.4).
2.3.5.2. Weekdays versus Weekends
Other temporally varying factors including weekdays versus weekend days and daily
temperature were not significantly associated with duration of time (min/day) spent at the home

43
Table 2.4. Zero-inflated Generalized Mixed-Effects Model (GLMM) results for time spent in
five non-home spatial contexts adjusted for covariates.
Predictors
Non-Home
Residential
(min/d)
Commercial
and Services
(min/d)
Industrial
and Office
Spaces
(min/d)
Schools and
Public
Facilities
(min/d)
Parks and
Open Spaces
(min/d)
Count Model Incidence Rate Ratio (95%CI)
Period: 1st Trimester
1.14
(.65 – 1.98)
1.01
(.79 – 1.30)
1.37
(.89 – 2.10)
.81
(.49 – 1.34)
1.07
(.37 – 3.11)
Period: 4-6 Months
Postpartum
1.83 *
(1.03 – 3.25)
.88
(.66 – 1.17)
1.51
(.96 – 2.36)
.91
(.56 – 1.46)
.45
(.16 – 1.31)
Valid GPS
observation (h/day)
1.01
(.94 – 1.09)
1.07 **
(1.02 – 1.13)
1.03
(.97 – 1.09)
1.05
(.96 – 1.15)
.95
(.80 – 1.13)
Type of day: Weekend
1.64 *
(1.05 – 2.56)

3.02 **
(1.32 – 6.92)
Average air
temperature (°C)
1.04
(.99 – 1.10)

Maternal parity: Already
had child

.63 **
(.46 – .86)

Employment status:
Employed

1.43 **
(1.11 – 1.85)
2.01 *
(1.06 – 3.79)
2.25 ***
(1.40 – 3.64)

Maternal age at consent

.99
(.97 – 1.01)

Neighborhood safety
and
cohesion score

.83 *
(.71 – .97)

Zero-Inflated Model Odds Ratio (95%CI)
Period: 1st Trimester
.95
(.53 – 1.79)
.60
(.33 – 1.09)
.63
(.34 – 1.18)
.88
(.45 – 1.72)
1.35
(.50 – 3.85)
Period: 4-6 Months
Postpartum
1.05
(.59 – 2.00)
.37 **
(.19 – .72)
.63
(.32 – 1.27)
.96
(.48 – 1.96)
1.2
(.40 – 3.70)
Valid GPS
observation (hay/day)
.34 *
(.12 – .97)
.28 *
(.10 – .85)
.20
(.04 – 1.01)
.14 **
(.03 – .59)
.22
(.02 – 2.70)
Type of day: Weekend
.74
(.45 – 1.22)

1.37
(.59 – 3.23)
Average air
temperature (°C)
1.00
(1.00 – 1.06)

Maternal parity: Already
had child

.53
(.18 – 1.59)

Employment status:
Employed

.62
(.29 – 1.33)
2.33 *
(1.11 – 5.00)
.61
(.28 – 1.35)

Maternal age at consent

1.10 *
(1.01 – 1.19)

Neighborhood safety
and cohesion score

1.16
(.77 – 1.82)

*p<.05. **p<.01. ***p<.001. Exponentiated parameter estimates are shown. Reversed odds ratio (i.e., odds
for an outcome to be non-zero) of zero-inflated models were calculated for easier interpretation.
Note: BMI = Body Mass Index. GPS = Global Positioning System.

44
Table 2.5. Zero-inflated Generalized Mixed-Effects Model (GLMM) results for time spent in
home and non-home contexts and microenvironments adjusted for covariates.
Predictors
Home
Residence
(h/day)
Home Residence
Excluding Sleep
Hours (min/day)
Home
Residence
Outdoor
(min/day)
All Non-Home
Contexts
(min/day)
All Non-Home
Contexts
Outdoor
(min/day)
Count Model Incidence Rate Ratio (95%CI)
Period: 1st
Trimester
.98
(.93 – 1.05)
.95
(.85 – 1.06)
.87
(.74 – 1.02)
1.18
(.94 – 1.47)
1.15
(.84 – 1.57)
Period: 4-6
Months
Postpartum
1.00
(.93 – 1.07)
1.01
(.90 – 1.13)
.90
(.75 – 1.10)
1.08
(.86 – 1.36)
1.07
(.78 – 1.47)
Valid GPS
observation
(h/day)
1.06 ***
(1.05 – 1.07)
1.15 ***
(1.13 – 1.17)
1.06 ***
(1.04 – 1.08)
1.05 **
(1.01 – 1.08)
1.03
(.98 – 1.07)
Employment
status:
Employed
.92
(.84 – 1.01)

1.48 **
(1.10 – 1.99)

Type of day:
Weekend

1.06
(.89 – 1.27)

Maternal age at
consent

1.00
(.97 – 1.02)
.99
(.95 – 1.03)
Zero-Inflated
Model
Odds Ratio (95%CI)
Period: 1st
Trimester

1.14
(.4 – 3.23)
.56
(.15 – 2.13)
.71
(.42 – 1.22)
1.05
(.62 – 1.82)
Period: 4-6
Months
Postpartum

1.49
(.53 – 4.35)
1.02
(.21 – 5.00)
.83
(.45 – 1.54)
.72
(.42 – 1.27)
Valid GPS
observation
(h/day)

1.92 ***
(1.67 – 2.38)
.20
(.02 – 1.85)
1.19 ***
(1.15 – 1.25)
.23 *
(.07 – .79)
Employment
status:
Employed

.59 *
(.38 – .93)

Type of day:
Weekend

1.89
(1.00 – 3.70)

Maternal age at
consent

1.10 **
(1.11 – 1.18)
1.10 **
(1.11 – 1.16)
*p<.05. **p<.01. ***p<.001. Exponentiated parameter estimates are shown. Reversed odds ratio (i.e., odds
for an outcome to be non-zero) of zero-inflated models were calculated for easier interpretation. Zero-inflated
model was not applied to home residence related outcomes given that extremely rare cases of having zero
min/day spent at home residence.
Note: GPS = Global Positioning System.

45
Table 2.6. Zero-inflated Generalized Mixed-Effects Model (GLMM) results for time spent in
pedestrian and vehicular trips and number of pedestrian and vehicular trips performed adjusted
for covariates.
Predictors
All Trip
(min/d)
Pedestrian-
based Trip
(min/d)
Vehicular-
based Trip
(min/d)
All Trip
(N/d)
Pedestrian-
based Trip
(N/d)
Vehicular-
based Trip
(N/d)
Count Model Incidence Rate Ratio (95%CI)
Period: 1st
Trimester
1.03
(.86 – 1.24)
1.03
(.75 – 1.41)
1.05
(.86 – 1.28)
1.00
(.85 – 1.18)
1.04
(.73 – 1.47)
1.04
(.88 – 1.23)
Period: 4-6 Months
Postpartum
1.04
(.87 – 1.25)
.96
(.69 – 1.32)
1.17
(.95 – 1.43)
.90
(.76 – 1.06)
.86
(.59 – 1.25)
1.00
(.84 – 1.20)
Valid GPS
observation (h/day)
1.05 ***
(1.02 – 1.08)
1.02
(.97 – 1.08)
1.05 **
(1.02 – 1.09)
1.04 **
(1.01 – 1.07)
.97
(.92 – 1.03)
1.05 **
(1.02 – 1.08)
Type of day:
Weekend
.96
(.83 – 1.11)

.93
(.82 – 1.06)

Maternal age at
consent
1.01
(.99 – 1.03)
.99
(.96 – 1.02)
1.01
(.99 – 1.03)
1.03 *
(1.00 – 1.05)

1.02
(1.00 – 1.04)
Education: Some
college/Graduate
1.08
(.86 – 1.36)

1.05
(.82 – 1.35)

1.13
(.87 – 1.48)
Neighborhood
deprivation
score

1.12 *
(1.01 – 1.25)

Neighborhood
safety and
cohesion score

.86 *
(.76 – .97)
Zero-Inflated
Model
Odds Ratio (95%CI)
Period: 1st
Trimester
.76
(.45 – 1.27)
.74
(.43 – 1.25)
.79
(.45 – 1.41)
.75
(.42 – 1.37)
.51
(.21 – 1.27)
.71
(.37 – 1.43)
Period: 4-6 Months
Postpartum
.78
(.45 – 1.35)
.6
(.34 – 1.03)
.5 *
(.28 – .92)
.79
(.42 – 1.49)
.48
(.18 – 1.28)
.53
(.24 – 1.18)
Valid GPS
observation (h/day)
1.18 ***
(1.11 – 1.23)
.2 *
(.05 – .86)
.36 *
(.14 – .94)
1.18 ***
(1.11 – 1.23)
.21
(.03 – 1.82)
.31
(.08 – 1.18)
Type of day:
Weekend
.54 **
(.36 – .82)

.53 *
(.33 – .87)

Maternal age at
consent
1.08 **
(1.01 – 1.14)
1.06 *
(1.01 –
1.11)
1.10 **
(1.11 – 1.18)
1.06 *
(1.01 – 1.14)

1.09 *
(1.01 – 1.16)
Education: Some
college/Graduate
2.13 *
(1.11 – 4.35)

3.33 **
(1.43 – 7.69)

3.33 *
(1.25 – 9.09)
Neighborhood
deprivation
score

.74
(.56 – 1.01)

Neighborhood
cohesion and safety
score

1.3
(.83 – 2.13)
*p<.05. **p<.01. ***p<.001. Exponentiated parameter estimates are shown. Reversed odds ratio (i.e., odds
for an outcome to be non-zero) of zero-inflated models were calculated for easier interpretation.
Note: GPS = Global Positioning System.

46
residence (when participants were there). Results remained unchanged in sensitivity analyses
excluding days with <10 h or <16 h of GPS observations, or self-reported sleeping hours. As for
non-home contexts, when women visited non-home residential locations or parks and open
spaces during weekend days, they spent 64% (IRR=1.64, 95%CI: 1.05-2.56) and 202%
(IRR=3.02, 95%CI: 1.32-6.92) more min/day at each context, respectively, as compared to
weekdays. Additionally, during weekend days, the odds (OR=.48; 95%CI: .35–.82) of
accumulating any minutes in trips decreased by 52%.
2.3.5.3. Individual Sociodemographic and Residential Neighborhood Characteristics
Other than weekdays vs weekend days, individual sociodemographic and residential
neighborhood characteristics, including employment status, maternal education, and self-
reported neighborhood cohesion and safety scores, were also significantly associated with time-
activity (Table 2.4) and mobility (Table 2.6) patterns. Specifically, those employed spent on
average 48% more min/d (IRR=1.48, 95%CI: 1.10-1.99) when they visited non-home contexts
and had 133% higher odds (OR=2.33, 95%CI: 1.10-5.00) of visiting industrial and office spaces
compared to non-employed counterparts. In addition, women who already had at least one child
(IRR=.65; 95%CI: .45-.93) spent 35% fewer min/day visiting commercial and services locations
compared to women experiencing their first pregnancy.
In terms of mobility patterns, maternal education was significantly associated with longer
duration of time spent in trips when they were taken. Specifically, women with post high school
education had 223% greater odds (OR=3.33, 95%CI: 1.41-7.69) of accumulating minutes on
vehicle-based trips and 113% greater odds (OR=2.13, 95%CI: 1.05-4.35) of accumulating
minutes on all trips regardless of mode (Table 2.6) compared to women with high school
diploma and below. Moreover, women living in safer neighborhoods (based on reported safety

47
and cohesion score) took 14% fewer vehicle-based trips per day (IRR=.86; 95%CI: .76-.97)
overall.
2.4. Discussion
The overarching goal of this analysis was to examine how dynamic time-activity and
mobility patterns vary for both the pregnant woman and the fetus, and how these might differ
across levels of personal, socioeconomic, or neighborhood level disadvantage. In this work, a
data processing and analysis pipeline was developed for highly resolved GPS data in a panel
study of Hispanic pregnant women who were continuously monitored for 4 days during each of
the 1st and 3rd trimesters of pregnancy and at 4-6 months postpartum. This study identified stays
and trips and classified their spatial and indoor/outdoor microenvironmental contexts (for stays)
and modes (for trips). Then whether time-activity and mobility patterns varied over time during
pregnancy and the early postpartum period, and by individual sociodemographic, residential
neighborhood, and other temporal factors were tested. This work also highlights the inadequacy
of assuming individuals are stationary when assessing environmental exposures during
pregnancy and their effects on maternal and child health.
2.4.1. Time-Activity and Mobility Patterns of Pregnant Women
To start, this study found that participants on average spent nearly 70% (17.3 h/day) of
their time at their home residences during pregnancy and the early postpartum period, which
matches several studies examining the time-activity and mobility patterns of pregnant women
(Nethery, Brauer, and Janssen 2009; Ouidir et al. 2015; Wu et al. 2013). For instance, Nethery,
Brauer, and Janssen (2009) reported a cohort of Canadian pregnant women spent 16.2 h/day
at/near home during pregnancy while Zhu et al. (2019) reported a cohort of Chinese pregnant
women spent 15 h/day at/near home. Moreover, although the finding of this study - this group of

48
Hispanic women rarely visited parks and open space could not be directly compared to other
pregnancy studies among Hispanic women in the US, they indicate a potential public health
concern since multiple studies have shown that exposure to greenness is associated with lower
exposure to environmental hazards and decreased risk of adverse pregnancy outcomes (Dadvand
et al. 2012; McEachan et al. 2016; Zhan et al. 2020). Past studies have indicated that minority
and low SES populations have lower parks and open space availability (e.g., no park within
walking distance) and quality (e.g., crime, lack of maintenance), which might help explain the
low utilization of parks and open space in the cohort of this study. Consistently, because urban
Los Angeles has limited parks and green infrastructure in general and higher quality parks and
open space occur in the more expensive areas of the city, the low-income participants in this
study may have had less access to parks and open space (Sister, Wolch, and Wilson 2010;
Wolch, Wilson, and Fehrenbach 2005). The next study (Chapter 3) used GIS to measure
greenness exposure and parks and open space access in participants’ residential neighborhoods,
as well as interactions with individual health characteristics, to further understand the
consequences that flow from this state-of-affairs.
The daily mobility patterns of participants in this study differed from results reported by
Wu et al. (2013) in the other GPS-based Southern California study that also examined mobility
patterns of pregnant women (Wu et al. 2013). Specifically, participants of this study spent 1.7
times more min/day on average in vehicle-based trips compared to the prior study. However, Wu
et al. (2013) study participants were from different counties with a more diverse racial and ethnic
composition and wider SES range compared to this study that focused on predominantly low
income, Hispanic participants from Central, East, and South Los Angeles. A study by MacLeod
et al. (2018) found low-income, pregnant women in another urban cohort in Los Angeles

49
reported significantly more time in vehicle-based trips in a cross-sectional survey, which might
explain this discrepancy since low SES groups may have longer commuting times and make
more frequent use of public transit than higher SES groups.
2.4.2. Changes in Time-Activity and Mobility Patterns during Pregnancy and Postpartum
Longitudinally, this study did not find women’s time spent at home differed significantly
across pregnancy and early postpartum. This finding differs from the results of other studies
examining time-activity patterns of women across pregnancy (Nethery, Brauer, and Janssen
2009; Zhu et al. 2019). Nethery et al. (2009) reported that increasing weeks of pregnancy until
the 3
rd
trimester were associated with increased time spent at home in a sample of 62 pregnant
women living in Vancouver, BC, Canada. The authors hypothesized this might be due to the
decrease in PA in later trimesters of pregnancy. However, this study focused on a group of
Hispanic women that were primarily low SES. Consequently, they might not be able to afford or
have time to engage in leisure activities due to increased home or work responsibilities
(Kakinami et al. 2018).
In terms of time spent in non-home contexts, this study found women’s odds of visiting
commercial and services locations decreased at 4-6 months postpartum compared to the 3
rd

trimester of pregnancy. This change may be explained by increasing stays at home residence due
to childcare responsibilities or the fact that women permanently or temporarily left their jobs at
these times since the employment rate dropped from 39.6 to 19.6% between the 3
rd
trimester and
the 4-6 months postpartum. This study did not find any difference between women’s time spent
in commercial and services locations between the 1
st
and 3
rd
trimesters. However, a similar study
in Shanghai, China reported women’s time spent working decreased by two hours in the 3
rd

trimester compared to the 1
st
trimester (Zhu et al. 2019). This study was able to disentangle

50
whether the purpose of visiting commercial and service locations was for work or fulfilling daily
life needs such as visiting hospitals, schools, and supermarkets, which might explain why results
of this study differed from the Shanghai study above.
Regarding daily mobility patterns, my finding of no meaningful changes in time spent in
vehicle- or pedestrian-based trips between the 1
st
and 3
rd
trimesters of pregnancy adds to the
mixed results reported in the literature. Results of this study were consistent with the study by
Nethery et al. (2009) that reported no longitudinal changes in time spent in transit across the 1
st
,
2
nd,
and 3
rd
trimesters of pregnancy, but inconsistent with Zhu et al. (2019) who reported
pregnant women’s time spent in vehicles increased between the 1
st
and 3
rd
trimester. However,
the latter study was located in Shanghai, China, a city with an urban planning system that heavily
incorporates pedestrian-oriented street networks and public transit systems in contrast to the Los
Angeles metropolitan area, which may result in different travel behaviors. Until now, there are
few studies that have examined daily mobility patterns of pregnant women, and more are needed
to understand how mobility patterns change across pregnancy and postpartum periods.
2.4.3. Additional Predictors of Time-Activity and Mobility Patterns
Findings of this study that pregnant women’s time-activity and daily mobility patterns
vary with additional temporal, individual sociodemographic, and residential neighborhood
factors suggest that there may be highly variable patterns even among a primarily low-income,
Hispanic population. This study found that those who were employed spent more time at
industrial and office spaces during the week and more time at parks and open spaces during
weekends. Participants with higher educational attainment were more likely to take vehicle-
based trips, a fact that was consistent with study results of Wu et al. (2013). This study also
found that women living in safer neighborhoods performed fewer vehicle-based trips daily,

51
which might be explained by their inclination to take more walking trips given safer streets,
although it did not find a statistically significant relationship between neighborhood safety and
numbers and durations of pedestrian-based trips.
2.4.4. Implications for Future Studies
This study found pregnant and early postpartum women spent a substantial portion of
their time at indoor locations and took several trips per day – approximately a quarter of which
were pedestrian trips to visit a variety of destinations. These patterns also differed over the
course of the pregnancy and the postpartum period. Findings of this study have important
implications for future studies that aim to investigate the association between environmental
exposures of pregnant women and maternal or child health outcomes. The residential-based
approach used by most studies in the past may under- or over-estimate physical, built, and social
environment exposures of interest (e.g., PM2.5, green space, crime). Consequently, the true
relationships between environmental exposures and targeted health behaviors (i.e., PA) and
outcomes (i.e., respiratory diseases) may be masked, especially when investigating acute or
short-term dose-response relationships (e.g., daily, weekly, monthly). In addition, the variations
in time-activity patterns across pregnancy and postpartum periods suggest the need for more
longitudinal studies to complement cross-sectional studies.
Kwan (2018) argues that spatial and temporal mismatches and uncertainties make it
difficult to clarify the influence of contextual variables on health behaviors or outcomes. Given
the need to prepare for childbirth, infant care, and other responsibilities during pregnancy and
early postpartum, women’s day-to-day time-activity and daily mobility patterns may vary more
than those of the general population (Varshavsky et al. 2020). As a result, future studies should
move from “snapshot” to activity space-based approaches to assess the environmental exposures

52
of pregnant women (Yi et al. 2019). Mobile sensing technologies, such as GPS, can provide fine-
grained mobility trajectories that can be used to assess environmental exposures that reflect time-
activity patterns. As a result, these technologies can reduce the uncertainties in contextual
exposures (i.e., the disparities between the true contextual and measured contextual units) (Chaix
2018; Matthews and Yang 2013; Robertson and Feick 2018). Lastly, this study found that
pregnant and early postpartum women’s time-activity and mobility patterns varied across
weekend days versus weekdays, employment status, education attainment, and neighborhood
cohesion and safety, which suggest that these might be important modifiers to account for in
future exposure studies.
2.4.5. Study Strengths and Limitations
To the best of my knowledge, this is the first study that examines time-activity and daily
mobility patterns of pregnant women across the pregnancy and early postpartum periods. A
major strength is the application of GPS to repeatedly collect highly resolved geospatial location
data across the 1
st
and 3
rd
trimesters of pregnancy and at 4-6 months postpartum. As a result, this
study overcame recall biases inherent in self-reported time-activity or mobility surveys and
provided insights into longitudinal changes in these patterns. Additionally, the study applies a
kernel density-based algorithm to classify stay contexts and trip modes, achieving higher
accuracy and better sensitivity than the point-by-point classification approach. Compared to
computationally intensive methods, this study’s GPS processing and stay/trip detection workflow
may offer a lower technical difficulty threshold for future studies that aim at utilizing mobile-
phone collected location-tracking data to generate time-activity patterns. Furthermore, this study
collected and used highly time-resolved (10-s epoch) GPS data to detect stays and trips and
classified spatial context, indoor/outdoor microenvironments, and trip modes in GIS. These fine-

53
grained data and advanced GIS analytical tools helped us to examine the time-activity and
mobility patterns during pregnancy and early postpartum at various temporal spacings. The
longitudinal design used for this study allowed us to examine both the variations in time-activity
patterns between women and the day-to-day variations for each woman.
This study also has a few limitations. First, the GPS data collected by this study had some
missingness. To mitigate its impacts on analyses, this study made efforts to impute GPS data
using existing information and re-run the analysis with stricter thresholds of daily observation
hours or excluding data collected during sleep hours. Results of this study remained largely
unchanged. Additionally, missing data did not demonstrate diurnal patterns (i.e., it was roughly
invariant throughout the day). However, there are other factors that may still potentially affect
outcomes of this study. For instance, missingness patterns of GPS data may be correlated with
spatial contexts (e.g., tall buildings, trees) that could obstruct receiver signals. Second, although
this study tackled signal loss issue by flagging signal loss scenarios with confidence levels and
excluded those with extremely low confidence, it still could not be sure that the locations
recorded by the device during signal loss matched the true location. Third, this study could not
distinguish a trip to and from work locations from other trips, which might inform the
interpretation of results of time-activity patterns for certain respondents and the types of contexts
in play. Fourth, this study had a relatively small sample size and only collected 4-day GPS data
in two weekdays and two weekend days during each study period. Thus, the time-activity and
mobility patterns detected from samples of this study may not capture some less frequent
activities that occur on a weekly basis or on other days of the week, such as grocery shopping.
Lastly, this study focused on a health disparity group of low-income, Hispanic women, a
population that has been understudied and disproportionally exposed to various environmental

54
hazards. Thus, this study’s results may not generalize to pregnant women in other regions or SES
or racial/ethnic groups; nevertheless, they shed light on an important population, and they may
pave the way for future studies to examine pregnant women’s environmental exposures within
their everyday activity spaces.
2.5. Conclusions
Pregnancy and early postpartum are critical periods for women’s health, and this study
have shown that time-activity and mobility patterns of women will likely vary over this journey
for many women. Time-activity and mobility patterns can also be used to directly determine
environmental exposures that may affect both short- and long-term maternal and infant health
outcomes. Therefore, future studies examining the impacts of environmental or contextual
exposures on maternal or fetal health should consider the dynamics of these patterns because
they may directly influence exposure measurement error and the ability to detect meaningful
relationships.

55
Chapter 3 Assessing Dynamic Daily Exposures to Built-Environment
Characteristics during Pregnancy and Early Postpartum using Smartphone
Location Data
Emerging research has associated BE characteristics (e.g., park access, walkability) with
maternal and infant health during and after pregnancy (Porter et al. 2019; Thomson, Goodman,
and Landry 2019; McEachan et al. 2016; Boll et al. 2020; Liao et al. 2019; Anabitarte et al.
2020; Torres Toda et al. 2020). However, most studies have measured the BE exposure via static
methods (i.e., residential location at one timepoint during pregnancy). These static methods may
not capture pregnant women’s dynamic daily exposure encountered in activity spaces across the
pregnancy and postpartum periods. This chapter evaluates BE exposure measurement error using
static versus dynamic daily exposures using 552-days of smartphone location data from 62
Hispanic pregnant women during the 1
st
and 3
rd
trimesters, and at 4-6 months postpartum.
3.1. Related Work
The impact of the BE on pregnant women’s health has attracted considerable attention
over the past decade. Past research has associated exposure to BE characteristics such as shorter
distance to parks and public transit stops and higher neighborhood walkability with greater PA
outcomes in pregnant women (Porter et al. 2019; Thomson, Goodman, and Landry 2019).
Additionally, studies have shown that greater exposure to urban nature (e.g., visual access,
proximity, coverage) decreases pregnant women’s stress levels and depressive symptoms
(McEachan et al. 2016; Boll et al. 2020). Furthermore, studies have linked higher levels of
greenness exposure and park coverage to decreased maternal glucose levels, attenuated risks of
gestational diabetes mellitus and weight gain, and higher infant birthweight (Liao et al. 2019;
Anabitarte et al. 2020; Torres Toda et al. 2020).

56
However, most studies assess prenatal BE exposure at a single point in time and at the
residential neighborhood level (i.e., in a circular buffer around the home location as determined
by an address obtained at time of birth) (Banay et al. 2017). This “static” approach has several
major limitations. Spatially, relying on residential address at a single point late in the pregnancy
ignores residential mobility (or moving) across pregnancy and postpartum (Bell and Belanger
2012; Chen et al. 2010; Hodgson et al. 2015). Also, static methods do not capture women’s
dynamic BE exposures occurring outside the home in their activity spaces (e.g., walking trips,
errands, other visited locations) which can also influence their health outcomes (Matthews and
Yang 2013; Perez, Ruiz, and Berrigan 2019). Moreover, it does not consider changes in activities
and behaviors as pregnancy progresses over time and into the postpartum period. These could
dramatically vary due to preparations for childbirth, difficulty of physically moving around and
increased fatigue later in pregnancy, and childcare responsibilities after birth (Varshavsky et al.
2020). Further, within a particular period (e.g., 3
rd
trimester), day-to-day changes in BE exposure
could result from household, occupational, and recreational activities (e.g., grocery shopping,
commuting to work) which differ by days of the week. As a result, assessing exposures using
static approaches may fail to capture the “true causally relevant” BE characteristics that exert
contextual influence on pregnant women’s and infants’ health and could introduce exposure
measurement error and potential bias (Robertson and Feick 2018; Yi et al. 2019).
Increasingly, studies are starting to apply “dynamic” exposure assessment approaches
(i.e., matching dynamic human movement with environmental features) to derive BE exposures
(Zenk et al. 2011, 2018; Chaix et al. 2012, Jankowska et al. 2021). In these studies, highly
resolved (e.g., every few seconds) GPS monitoring data are usually collected, based upon which
individual activity spaces are constructed (e.g., activity locations one visits and paths one travels)

57
and integrated with BE layers (e.g., parks polygons, street centerlines) in GIS software to
measure individual’s dynamic exposures at high spatiotemporal resolutions (e.g., total areas of
parks in daily path areas) (Zenk et al. 2011; Yi et al. 2019). Consequently, these dynamic
approaches may improve exposure assessment in pregnancy studies compared to residential-
based static approaches since they incorporate information about where and when individuals
spend their time. However, very few studies have been able to assess dynamic BE exposures
during pregnancy and postpartum on a large number of participants given the potentially higher
burden of collecting personal, highly resolved geolocation data. Therefore, to understand
whether quantified BE exposures and observed health relationships are sensitive to the choice of
measurement method, it is important to evaluate the similarity (i.e., correlation) between static
and dynamic exposure estimates. To date, few studies have tackled this topic, and none have
focused on pregnant women to our knowledge. (Jankowska et al. 2017; Zhao, Kwan, and Zhou
2018).
In addition, there are several examples of health studies where static residential estimates
are used to classify individuals into BE exposure groups (e.g., representing high- and low-
walkable neighborhoods using top and bottom quartiles of the residential walkability index
score) (Van Dyck et al. 2010; Carlson et al. 2015). Especially for pregnancy studies, comparing
dynamic exposure estimates to the typically used static residential approach might shed light on
the potential for exposure measurement error and how this varies over time (e.g., pregnancy vs.
postpartum) and human mobility patterns (i.e., spatial extent of human movement footprints).
To fill these gaps, this study collected 4-day smartphone location data from a group of
Hispanic, predominantly low-income women during the 1
st
and 3
rd
trimesters of pregnancy and at
4-6 months postpartum, and characterized their daily dynamic and static BE exposures in this

58
critical life stage. The potential for exposure measurement error was then examined in relying on
daily, static BE exposures (using the 3
rd
trimester residential location) compared to dynamic
exposures within activity spaces in this environmental health disparities subpopulation in Los
Angeles, CA. This study has the three following aims:
1) To describe pregnant women’s dynamic daily BE exposure patterns during the 1
st
and 3
rd

trimesters of pregnancy and at 4-6 months postpartum;
2) To assess correlations between various static and dynamic BE exposure measures; and
3) To evaluate the potential extent and drivers of exposure measurement error in static vs.
dynamic BE exposure measures.
3.2. Methods
3.2.1. Design and Overview
Data for this study comes from the Real-Time and Personal Sampling sub-study of the
MADRES Center. This study uses an intensive longitudinal, observational panel study design
and examines the daily effects of environmental exposures and social stressors on maternal pre-
and post-partum obesity-related biobehavioral responses. A total of 65 Hispanic, predominantly
lower income mothers were drawn from the larger MADRES prospective cohort study between
2016-2018. Participants were recruited from prenatal care providers serving predominantly
medically-underserved populations in Los Angeles, California, including two non-profit
community health clinics, one county hospital prenatal clinic, one private obstetrics and
gynecology practice, and through self-referral from community meetings and local
advertisements (O’Connor et al. 2019). To be eligible for this sub-study, a participant needed to
be >18 years old with a singleton pregnancy and be at less than 30 weeks’ gestation at time of
recruitment. In addition, participants who were HIV positive, had physical, mental, or cognitive

59
disabilities that prevented participation, or were currently incarcerated were excluded from the
study (O’Connor et al. 2019). The USC Institutional Review Board approved all study
procedures, and participants signed an informed consent before enrolling into the study.
3.2.2. Data Collection
3.2.2.1. Geolocation using GPS
This study continuously collected GPS data from 65 study participants at 10-s intervals
for four days (two weekdays and two weekend days) during the 1
st
and 3
rd
trimester and at 4-6
months postpartum. To enable collecting highly resolved and encrypted GPS data collection,
MADRES researchers designed a custom smartphone application (madresGPS app) for Android
operating systems. The application on dedicated study smartphones (Samsung MotoG phone)
were configured by study coordinators to record geographic coordinates and geolocation/motion
metadata, which logged instantaneous GPS location and sensor data every 10 s from the
smartphone’s multiple built-in location finding features (cell tower triangulation, Wi-Fi
networks, and GPS) and motion sensors (O’Connor et al. 2019). Along with the timestamp, the
application recorded metadata such as the number of satellites in use/view, geolocation accuracy,
source of GPS, velocity (if GPS source), and network connection status (if network source)
(O’Connor et al. 2019).
3.2.2.2. Ecological Momentary Assessment (EMA)
Participants self-reported physical context information in EMA surveys delivered through
the MovisensXS application (Android), which was pre-installed on the same study phone used to
collect GPS data. EMA surveys were designed to prompt at random times during each five pre-
specified sampling windows (i.e., wake-up to 10 a.m.; 11 a.m.to 1 p.m.; 2 p.m.to 4 p.m.; 5 p.m.to

60
7 p.m., and 8 p.m.to bedtime) within the same four-day GPS data collection windows during the
three study periods. O’Connor et al. (2019) describes the EMA survey questions in more detail.
3.2.3. Data Processing
3.2.3.1. GPS Data Analysis
The 10-s epoch GPS geolocation data was processed using a custom algorithm I
developed and described in detail in Chapter 2 using SAS version 9.4 (SAS Institute Inc).
Missing geolocation data was then imputed when participants were very likely to be at their
home location during the day- or night-time. At night, typical sleep and wake time windows
were used along with location before and during to determine whether participants were likely to
be home and to fill in any gaps in the GPS data. During the day, confirmed EMA reports of
being at “Home-Indoor” or “Home-Outdoor” all day were used to fill in gaps with the home
location. Flags were then created for days with <6 hours of GPS data (post-imputation) as invalid
days, after considering missing data patterns.
3.2.3.2. Exposure Assessment
Constructing Activity Spaces
Daily activity spaces were constructed for each participant based on their geolocation
trajectories via the route buffer (RB) and kernel density estimation (KDE) methods (Figure 3.1)
using ArcGIS Pro 10.7.1 (Esri, 2021). The two methods complement one another and capture
pregnant women’s daily exposure to BE characteristics, including those along women’s daily
travel paths (via RB) and around activity locations where they spent the most time (via KDE).

61

Figure 3.1. An illustration of two activity space metrics – route buffer (RB) and kernel density
estimation (KDE) applied in this study

To construct buffers along routes, consecutive GPS points were joined into lines and
buffered using a 250 m radius. This distance corresponds to the depth of several blocks in urban
Los Angeles, CA. It was selected to approximate two hypothetical mechanisms on how BE
characteristics could potentially influence health behaviors and outcomes, including viewshed
(i.e., seeing a park when walking on a street) and awareness (i.e., knowing a recreational facility
is located nearby) (Yi et al. 2019). Additionally, the geographical area of these route buffers was
calculated to represent spatial extent of daily mobility and classified into low, medium, and high
mobility days based on tertiles.
Additionally, the KDE approach was applied to generate time-weighted activity grids
using 50 by 50 m cells with a bandwidth of 250 m (to match the RB radius). Each grid was

62
assigned a normalized time weight (range from 0 to 1) based on percentage of stay duration
within a given day, with total weights adding up to 1.
Defining Residential Neighborhoods
To assess residential neighborhood exposures using the “static” approach, the home
location was first determined based on where participants spent the most time in their 3
rd

trimester. This location was selected as most similar to the typically used approach in pregnancy
studies (i.e., assessing exposure based on residential address provided on birth certificates or
questionnaires after delivery and assuming it represents the entire prenatal period). Briefly, KDE
was applied to identify the longest duration stay during the 3rd trimester as the home location,
which was then used to derive 800 m residential network buffers (RNB) via the Service Area
Analysis tool and StreetMap Premium product in ArcGIS Pro 10.7.1 (Esri, 2021). The network
buffer better captures the area to which a pregnant woman could realistically travel from the
home residence as compared to a circular buffer (James et al. 2014). An 800 m radius, which
corresponds to a 5-10-minute walk, is recommended for populations with relatively restricted
mobility (James et al. 2014; Frank et al. 2017). Additionally, a sensitivity analysis was
performed using a larger radius (e.g., 1,600 m) and the resulting BE exposure estimates differed
only slightly, so we chose to report 800m results only.
Assessing Built Environment Exposures
Eight BE measures were derived using the RNB, RB, and KDE methods, respectively.
These included greenness (3 measures), park and public transit access (3 measures), street
connectivity (1 measure), and walkability (1 measure) as follows: mean Normalized Difference
Vegetation Index (NDVI; ranges from -1 to 1 with higher value represents higher greenness; the
NDVI dataset was date-matched to corresponding residential buffers and activity spaces when

63
calculating mean NDVI), percent green space along walkable roads, and percent tree cover along
walkable roads, distance to the nearest park entrance, distance to the nearest public transit stop,
total area of parks and open space, pedestrian-oriented intersection density, and walkability index
score (ranges from 1 to 20 where higher scores represent more walkable), all of which have been
reported by previous studies to be associated with various human activities and health outcomes,
particularly among pregnant women (Banay et al. 2017; Besser and Dannenberg 2005, Jiang et
al. 2016; Pickard et al. 2015; Porter et al. 2019; Pretty et al. 2005; Tsai, Davis, and Jackson 2019;
Saelens et al. 2014). All BE measures, corresponding data sources, resolution, and processing
steps and interpretation are shown in Table 3.1.
RNB and RB exposures were calculated by summarizing values of the respective BE
measures (e.g., mean NDVI, total areas of parks and open space) within boundaries of
corresponding spatial units (i.e., 800 m residential network buffer, 250 m route buffer). KDE
exposures were calculated by weighting grid-based values of the respective BE measures with
time spent (e.g., multiplying walkability index score value of the grid by percentage of time
spent to produce weighted walkability scores per grid, then taking the sum of weighted scores
across all grids). Exposures were calculated using Eq 3.1 for the RNB or RB methods and Eq 3.2
for the KDE methods as follows:
!"#$%&'(
!"#/!#
= *(%-'(.
%
∗0!1
%
)
%
%&'
Eq 3.1
!"#$%&'(
()*
= *(%-'(.
%
∗0!1
%
∗34
%
)
%
%&'
Eq 3.2

64
Table 3.1. Built-environment (BE) measures, corresponding data sources, resolutions, and
processing steps.
Measures Data Sources Data Resolutions Processing Steps and Interpretation
Mean
NDVI
(range
from -1 to
1)
NASA MODIS 16-
Day L3 Global
product
1
250 x 250 m raster
grids
The mean Normalized Difference Vegetation Index
(NDVI) was used to represent overall vegetation levels
within respective spatial units (i.e., residential
neighborhood, activity space). NDVI values range from -
1 to 1, with negative values corresponding to areas with
water surfaces, 0-.2 representing barren surfaces, .2-.5
representing sparse vegetation, and >.5 representing
dense vegetation.
Percent
green space
along
walkable
roads (%)
EPA EnviroAtlas
Community Data
Census block
groups
Green space areas were derived by combining areas of
multiple land cover classes including water, trees and
forest, grass and herbaceous cover, shrubs, agriculture,
orchards, and woody and emergent wetlands. Sidewalk
areas were derived by buffering NAVTEQ roads with a
speed limit less than 55 miles per h (potentially walkable
roads) by a width of 25 m on each side. The measure then
was calculated by intersecting tree cover and sidewalk
areas per city block.
Percent
tree cover
along
walkable
roads (%)
EPA EnviroAtlas
Community Data
Census block
groups
Tree cover areas were derived by combining areas of
three land cover classes - trees, forests, and woody
wetlands. Sidewalk areas were derived by buffering
NAVTEQ roads with a speed limit less than 55 miles per
h (potentially walkable roads) by a width of 8.5 m on
each side. This measure then was calculated by
intersecting tree cover and sidewalk areas per city block.
Distance to
the nearest
park
entrance
(m)
EPA EnviroAtlas
Community Data
Buffer zones
This measure was derived by delineating approximate
walking areas from a park entrance at any given location
within the EnviroAtlas community boundary (i.e., Los
Angeles County).
Total parks
and open
space
coverage
(km
2
)
California Protected
Areas Database
Parks and open
space polygons
This measure included 1) National/state/regional parks,
forests, preserves, and wildlife areas; 2) large and small
urban parks that are mainly open space (as opposed to
recreational facility structures); 3) land trust preserves;
and 4) Special district open space lands (watersheds,
recreational areas, etc.) and other types of open space.
Distance to
the nearest
public
transit stop
(m)
EPA Smart Location
Database
Census block
groups
This measure was derived by measuring the minimum
walking distance in meters between the 2010 population-
weighted census block groups (CBG) centroid (as used
by SLD version 2.0) to the nearest transit stop of any
route type.

65
Table 3.1 (Cont.)
Measures Data Sources Data Resolutions Processing Steps and Interpretation
Pedestrian-
oriented
intersection
density (m)
EPA Smart Location
Database
Census block
groups
This measure was derived based on an analysis of
NAVTEQ 2011 Streets data. All 3-way intersections
were weighted by .6667 since they reduced street
connectivity compared to intersections with 4 or more
legs.
Walkability
index score
(range
from 1 to
20)
EPA Smart Location
Database
Census block
groups
A composite index score combining household and
employment density, street intersection density, and
distance to nearest transit stops. This measure represent
different BE characteristics that are known to be
supportive of walking. The index scores range from 1 to
20, with higher values representing better walkability.
1
The MODIS NDVI dataset was date-matched to corresponding activity spaces when calculating mean NDVI.
Note: NASA = National Aeronautics and Space Administration. MODIS = Moderate Resolution Imaging
Spectroradiometer. EPA = Environmental Protection Agency.

where BECn is the exposure estimate for a BE measure in the n
th
spatial unit (e.g., census block
group, grid) that intersects the RNB, RB, or KDE; Arean is the percentage of the area in the n
th

spatial unit that falls within the RNB, RN, or KDE; and TWn is the normalized time weights
(range from 0 to 1) of the n
th
grid.
3.2.4. Statistical Analysis
Descriptive statistics were calculated for static (i.e., RNB) and dynamic (i.e., RB and
KDE) BE exposures, respectively. Additionally, intraclass correlation coefficients (ICC) were
calculated to estimate the proportion of day-to-day variability in the dynamic exposures that was
between-participant (i.e., ICC%) compared to within-participant (i.e., 1-ICC%). Two-level (days
nested within participants) linear mixed-effects random intercept only models were used to
calculate the ICCs for the 8 dynamic BE measures. ICC value cut-offs of .40 and .75 were
selected to indicate weak (ICC<.40), moderate (.40≤ICC<.75), or strong (ICC≥.75) within-

66
person correlations (Zenk et al. 2018). Analyses were conducted in R 4.0.2 (R Foundation for
Statistical Computing, 2021) using the lme4 package (version 27.1) (Bates et al. 2015).
Pearson’s correlation coefficients were used to examine similarity of BE exposures
assessed using static versus dynamic methods (e.g., correlation between RNB and KDE mean
NDVI), given the approximately normal distributions of BE measures. Moreover, Pearson
correlation coefficients were calculated by study period (i.e., 1
st
trimester, 3
rd
trimester, and 4-6
months postpartum) to investigate how well dynamic estimates correlate with static 3
rd
trimester
estimates as an indication of potential for measurement error in relying solely on the latter.
Furthermore, daily BE exposures were classified into four groups (quartiles) using KDE
as the most dynamic method that is spatiotemporally matched to human movement and the
percent of “misclassified” days that were assigned to a different group using the static RNB
method was calculated as an indication of potential for exposure measurement error in the latter
approach. For example, if the KDE method classifies a day into the highest exposure (4
th

quartile) group but the RNB method classifies it into lower exposure groups (1
st
, 2
nd,
or 3
rd
quartile), it was labelled as a misclassified day. Sankey diagrams were then used to illustrate
these relationships, where arrows represent flows from quartiles of KDE exposures (left,
assumed “true” daily varying exposure) to quartiles of RNB exposure (right, static exposure),
and the thickness of the lines flowing from left to right (representing percent of days) indicates
the potential for exposure misclassification. In this study, whether potential exposure
misclassification was sensitive to daily mobility was investigated by creating the same Sankey
diagrams for low, medium, and high mobility days. Analyses were conducted in R 4.0.2 (R
Foundation for Statistical Computing, 2021) and Sankey diagrams were created using the
ggalluvial package (version 0.12.3) (Brunson 2020).

67
3.3. Results
3.3.1. Data Completeness
Across three study periods, 62 out of 65 participants who were initially enrolled in the
study provided at least one valid GPS observation day (≥6 hours of data). Thirty-five out of 62
participants had at least one valid day in all three study periods; 17 out of 62 in two of the three
periods; and 10 out of 62 in just one of the three periods. The final analytical sample of this study
comprised a total of 552 valid person-days of GPS data from 62 participants, with 205 person-
days during the 1
st
trimester, 180 person-days during the 3
rd
trimester, and 167 person-days at 4–
6 months postpartum. An average of 8.9 valid GPS days (SD=3.00; Range: 3.00-12.00) were
provided by participants across the three periods. On average, 21.7 hours (SD= 5.00; Range: 6.2-
24.00) of GPS observations were collected on valid days. The number of hours was highest in
the 3
rd
trimester (Mean=22.3 hours; SD=4.4; Range: 6.5-24.00) followed by the 4-6 months
postpartum period (Mean=21.8 hours; SD=4.7; Range: 7.00-24.00) and then the 1
st
trimester
(Mean=21.1 hours; SD=5.6; Range: 6.2-24.00). Almost half of the valid person-days (49.3%)
were weekend days across the three periods.
3.3.2. Daily BE Exposure Estimates during Pregnancy and Early Postpartum
Descriptive statistics for daily BE exposure estimates derived by the RNB, RB, and KDE
methods are shown in Table 3.2. The median geographic extent was 2.49 km
2
(IQR=11.20),
which is equivalent in coverage to several city blocks (.5 to 1 mile long) in urban Los Angeles,
CA. Depending upon the BE measure, exposure estimates can greatly vary across the three
methods. In addition, the exposure estimates derived by the KDE method had the largest
variability (i.e., IQR values) among the three methods.

68
Table 3.2. Descriptive statistics of daily built-environment (BE) exposure estimates for
participants using the route buffer (RB), kernel density estimation (KDE), and residential
network buffer (RNB) methods.
N=552 person-days from 62 participants
Variables Mean SD Median IQR Min Max
Geographic extent (km
2
)
RB 250 m 7.96 11.94 2.49 11.20 .20 91.91
Normalized Difference Vegetation Index (NDVI) (range from -1 to 1)
RB 250 m .18 .05 .18 .06 .06 .41
KDE 250 m .18 .05 .17 .06 .08 .41
RNB 800 m .18 .05 .18 .07 .10 .38
% Green space along walkable routes
RB 250 m 23.30 5.87 22.76 6.77 6.61 46.76
KDE 250 m 23.28 8.30 21.87 10.32 1.70 46.27
RNB 800 m 23.04 5.97 22.40 7.10 12.14 45.70
% Tree cover along walkable routes
RB 250 m 21.45 5.51 20.75 6.03 6.32 44.78
KDE 250 m 22.58 8.16 21.99 10.77 1.74 50.37
RNB 800 m 21.72 5.75 20.57 6.06 10.68 45.63
Distance to the nearest park entrance (m)
RB 250 m 877.97 397.02 791.82 421.93 294.97 2409.52
KDE 250 m 812.31 455.44 668.94 411.95 51.67 2342.11
RNB 800 m 789.29 414.09 660.04 302.91 400.38 2387.07
Total parks and open space area (km
2
)
RB 250 m 243.53 608.87 15.50 146.21 .00 4817.38
KDE 250 m .03 .10 .00 .01 .00 .93
RNB 800 m 9.67 15.65 2.84 11.44 .00 77.68
Distance to the nearest public transit (m)
RB 250 m 280.02 103.27 275.38 144.08 1.94 532.25
KDE 250 m 284.95 123.13 282.97 177.60 .03 1041.17
RNB 800 m 268.94 105.79 265.49 135.86 81.28 645.76
Pedestrian-oriented intersection density (# per mi
2
)
RB 250 m 47.90 33.35 42.27 26.55 2.20 190.79
KDE 250 m 51.81 40.78 41.73 38.92 .24 240.01
RNB 800 m 57.60 44.70 46.76 32.24 3.00 222.66
Walkability index score (range from 1 to 20)
RB 250 m 14.93 1.50 14.96 2.02 10.29 18.43
KDE 250 m 14.84 1.72 14.92 2.59 10.09 18.08
RNB 800 m 14.94 1.43 15.17 1.89 11.59 18.11

69
In terms of KDE-based activity space neighborhood greenness, women on average were
exposed to an NDVI value (range from -1 to 1) of .18 (SD=.05). This indicates their daily
activity locations, on average, had barren surfaces or very sparse vegetation. In addition, the
KDE-measured percent of green space and tree cover along walkable roads at their daily activity
locations on average was 23.3% (SD=8.3) and 22.6% (SD=8.2), respectively. Both numbers were
lower than the Los Angeles County average of 32.4% (% of green space along all walkable roads
of Los Angeles County) and 28.1% (% of tree cover along all walkable roads of Los Angeles
County). Mean exposure estimates of all greenness measures varied slightly for the same
measure across the RNB, RB, and KDE methods.
Turning next to park access, women’s KDE-measured mean daily distance from their
daily activity locations to the nearest park entrance was 812 m (SD=455; corresponding to a 15-
20-minute walk). The estimates differed slightly with the RB or RNB approaches. The total
estimated exposures to parks and open space area; however, varied tremendously. Women were
exposed to an average of only 0.03 km
2
parks and open space per day (about the size of a tiny
neighborhood park in urban Los Angeles) using KDE, which was on average < 0.5% of exposure
to parks and open space areas using RB and RNB. Women’s KDE-measured mean distance to
the nearest public transit stop at daily activity locations was 285 m (SD=123; corresponding to
<5-min walk).
Lastly, regarding street connectivity and walkability, the KDE-measured mean number of
pedestrian-oriented street intersections at daily activity locations women visited was about 52
intersections per square mile (SD=41), and the mean walkability index score was 14.8 (SD=1.72;
range from 1 to 20). These numbers were 34.7% and 1.5 scores higher than the LA County

70
average, respectively. Exposure estimates of these two measures varied slightly across the three
methods.
3.3.3. Within-person Correlations in Daily BE Exposures across Time
ICCs for KDE- and RB-derived day-level exposure estimates are presented in Table 3.3.
ICC represents the within-person correlation between (day-level) exposure estimates for each BE
variable. In other words, the higher the ICC value is, the larger portion of the variance in day-
level BE exposure would be explained by between-person (person-level) differences.
According to Table 3, most BE exposures demonstrated strong within-person correlations
(ICC>.75), which can be interpreted as >75% of differences in day-level exposures for a specific
BE variable were driven by between-person differences. However, an exception was the Total
Park and Open Space Area variable, which showed weak (ICCRB=.13) to moderate (ICCRB=.49)
within-person correlations. Lastly, for each BE variable, its exposures derived using the RB
method had overall lower ICC values than the KDE method (e.g., ICCKDE-NDVI > ICCRB-NDVI),
suggesting weaker within-person correlations for RB-based day-level estimates.
3.3.4. Correlations of BE Exposures across methods (Static versus Dynamic)
Pearson correlation coefficients between exposure estimates of the same BE measure
derived by static versus dynamic methods (e.g., RNB-based mean NDVI versus KDE-based
mean NDVI) are summarized in Table 3.4. In general, stronger positive correlations were found
between RNB and KDE exposure estimate pairs than between RNB and RB pairs. Moreover, the
strength of correlations varied substantially by BE measure.
Specifically, RNB estimates of total area of parks and open space were weakly positively
correlated (r=.31, p<.01) with KDE estimates. The RNB and KDE exposure estimates for %
green space and % tree cover along walkable roads were moderately positively correlated (r=.52,

71
Table 3.3. Intraclass-Correlation Coefficients (ICCs) of daily exposure estimates derived using
the kernel density estimation (KDE) and route buffer (RB) methods.
Variables ICC (KDE) ICC (RB)
Normalized Difference Vegetation Index
(NDVI; range from -1 to 1)
.79 .62
% Green space along walkable roads .82 .56
% Tree cover along walkable roads .87 .60
Total parks and open space area (km
2
) .49 .13
Distance to the nearest park entrance (m) .91 .63
Distance to the nearest public transit stop (m) .76 .61
Pedestrian-oriented intersection density (# per mi
2
) .90 .74
Walkability index score (range from 1 to 20) .88 .64
Note: ICC value cut-offs of .40 and .75 were selected to indicate weak (ICC<.40), moderate (.40≤ICC<.75), or
strong (ICC≥0.75) within-person correlations.

P<.01 and r=.55, p<.01, respectively). Lastly, the exposure estimates derived by RNB and KDE
for the remainder of BE measures were strongly positively correlated (r>.7, p<.01). The
correlation coefficients between RNB and KDE exposure estimates decreased slightly at 4-6
months postpartum and decreased substantially during the 1
st
trimester compared to the 3
rd

trimester (Table 3.4). This same pattern is manifested in the distance to the nearest public transit
stop measure (r=.61 between RNB versus KDE during the 1
st
trimester compared to r=.82
between RNB versus KDE during the 3
rd
trimester).
3.3.5. Impact of Daily Mobility on Potential for Exposure Misclassification using the Static
Method
Impact of daily mobility (i.e., low, medium, and high mobility days) on potential
exposure misclassification using the “static” RNB method (at single 3
rd
trimester residential
location) as compared to “dynamic” KDE method (at a daily level based on GPS tracks) is
shown in Figures 3.2a-h.

72
Table 3.4. Pearson correlation coefficients between day-level static versus dynamic exposure
estimates for the same built-environment (BE) variable during the 1
st
and 3
rd
trimesters and 4-6
months postpartum periods.
BE Variable Pairs Pearson r
Overall
(N=552 person-
days)
The 1
st
trimester
(N=205 person-days)
The 3
rd
trimester
(N=180 person-
days)
4-6 months
postpartum
(N=167 person-days)
Normalized Difference Vegetation Index (NDVI; range from -1 to 1)
RNB ~ KDE .82* .84* .71* .88*
RNB ~ RB .73* .78* .58* .78*
% Green space along walkable routes
RNB ~ KDE .52* .45* .57* .56*
RNB ~ RB .61* .56* .63* .64*
% Tree cover along walkable routes
RNB ~ KDE .55* .48* .62* .59*
RNB ~ RB .67* .66* .69* .67*
Distance to the nearest park entrance (m)
RNB ~ KDE .89* .9* .9* .88*
RNB ~ RB .75* .81* .74* .71*
Total parks and open spaces areas (km
2
)
RNB ~ KDE .31* .29* .4* .3*
RNB ~ RB -.04 -.02 .01 -.11
Distance to the nearest public transit stop (m)
RNB ~ KDE .74* .61* .82* .85*
RNB ~ RB .65* .66* .58* .72*
Pedestrian-oriented intersection density (# per mi
2
)
RNB ~ KDE .79* .81* .79* .76*
RNB ~ RB .69* .63* .7* .73*
Walkability index score (range from 1-20)
RNB ~ KDE .84* .83* .86* .83*
RNB ~ RB .73* .72* .72* .75*
*p<.01.
Note: KDE = Kernel Density Estimation. RB = Route Buffer. RNB = Residential Network Buffer.

73

(a) % green space along walkable routes
Figure 3.2. The extent and direction of daily exposure misclassification using third trimester residential network buffer (RNB)
exposures for built-environment (BE) measures compared to trimester-specific kernel density estimation (KDE) estimates.
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

74

Figure 3.2. (b) % tree cover along walkable routes
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

75

Figure 3.2. (c) Normalized Difference Vegetation Index (NDVI; ranges from -1 to 1)
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

76

Figure 3.2. (d) Distance to the nearest park entrance (in m)
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

77

Figure 3.2. (e) Distance to the nearest public transit stop (in m)
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

78

Figure 3.2. (f) Total area of parks and open space (in km
2
)
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

79

Figure 3.2. (g) Pedestrian-oriented intersection density (per mi
2
)
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

80

Figure 3.2. (h) Walkability index score (ranges from 1 to 20)
Note: The geographical area of route buffers was calculated to represent spatial extent of daily mobility and classified into low, medium, and high mobility days
based on tertiles.

81
According to these figures, impact of daily mobility on potential exposure
misclassification was observed across all BE measures when relying on the 3
rd
trimester RNB
method vs the dynamic KDE method, so that misclassification was largest in high mobility days.
This was manifested for measures such as NDVI, distance to nearest park entrance, distance to
public transit stop, and total area of parks and open space (Figures 3.2c, 3.2d, 3.2e, and 3.2f).
However, this result was less consistent for percent green space along walkable roads,
pedestrian-oriented intersection density, and walkability index score measures (Figures 3.2a,
3.2g, and 3.2h) and reversed for mean percent tree cover along walkable roads measure (Figure
3.2b).
Notably, in the highest exposure group (4
th
quartile) of high mobility days, 49.6% and
57.9% of person-days were misclassified into different groups by the RNB method for % green
space along walkable roads and % tree cover along walkable roads measures (Figures 3.2a and
3.2b). The percent of misclassification ranged from 25-35% for the highest exposure group in the
remainder of BE measures.
3.4. Discussion
In this study, daily dynamic exposure to BE characteristics during the 1
st
and 3
rd

trimesters and at 4-6 months postpartum was described by analyzing highly resolved smartphone
location data collected from a sample of 62 Hispanic pregnant women (across 552 observation
days) in Los Angeles, CA. Additionally, a critical yet unaddressed research question was
answered – how similar were pregnant women’s BE exposure derived by the residential-based
static method applied by previous studies, to GPS-based dynamic exposure in their daily activity
spaces? This study’s results have important implications for future studies interested in the

82
association between the BE exposure and maternal and infant health, especially in lower SES
and racial/ethnic minority groups. The implications of findings are discussed below.
3.4.1. Women’s Daily BE Exposures during Pregnancy and Early Postpartum
Women in our sample were found to have daily activity spaces that were equivalent to
several city blocks in Los Angeles (LA). Past studies have found lower SES groups have lower
levels of mobility in LA compared to higher SES groups, which may explain the overall small
daily spatial footprints in our predominantly low-income sample in this largely sprawling city
(Kim and Kwan 2021; Giuliano 2005). Kim and Kwan (2021) found low income and female
groups in Los Angeles had lower levels of daily mobility and were often “trapped” in their
residential neighborhoods which have disproportionally high exposures to environmental hazards
(e.g., traffic, air pollution). Therefore, the low mobility of the MADRES sample indicates
residential BE exposure may have greater influences on their activities and health during
pregnancy and early postpartum than their high SES counterparts, contrary to my original
hypothesis.
In this study, multiple measures were applied to capture different aspects of the greenness
exposure, including park coverage and proximity, vegetations levels (i.e., NDVI value), and %
green space and tree cover along walkable roads. Results showed women in this sample were
found to be typically exposed to daily activity spaces featuring very low to low vegetation levels
and minimal parks and open space. Wolch, Wilson, and Fehrenbach (2005) report low-SES areas
and neighborhoods dominated by Latinos in Los Angeles have dramatically lower levels of park
coverage compared to White-dominated areas, which may explain the overall low greenness
exposure in this sample given the greater role of residential exposures in women’s total
exposures. Interestingly, the women in sample of this study had convenient access to parks in

83
their daily activity spaces (i.e., the nearest park was on average only a 5-10 min walk), which did
not lead to actual park use (Chapter 2 found they had very low visits to parks and open space). It
is possible that women in the study sample have individual preferences to visit certain parks,
which may or may not be the nearest park (Kaczynski and Mowen 2011). Moreover, park
quality, amenities, and safety may be more important factors than distance to the closest park to
determine park use behaviors, especially for low SES groups (Kaczynski, Potwarka, and Saelens
2008).
The women were exposed to activity spaces with very high proximity to public transit
(typically <5-min walk to the nearest public transit stop) and higher than Los Angeles County
average street connectivity and walkability. These findings indicate women overall may be
exposed to a BE which better facilitates utilitarian walking (e.g., visits to corner groceries) and
active transport (e.g., walking to transit stops) than recreational activities (e.g., exercise in a
park). Nevertheless, the health impacts of daily exposures to high density areas as identified in
this sample should be further examined since previous research has also reported that urban areas
with concentrated activities may yield both health benefits and costs such as exposure to high
levels of air pollution (Marshall, Brauer, and Frank 2009).
3.4.2. Similarities of Exposure Estimates between Static and Dynamic Methods
Overall, the results showed that the correlations between the static and KDE estimates
were higher than those between the static and RB estimates. This is not surprising given the KDE
method weights exposure based on the duration of the stay (i.e., time-weighted spatial averaging)
and women on average spent a significant amount of time at home locations according to the
results reported in Chapter 2, whereas the RB method is time-insensitive (i.e., spatial averaging)
and weights the home location similarly to any other visited location. Despite the difference in

84
correlations between the static and the two dynamic methods, one method may be better at
measuring some BE characteristics than the other. For instance, KDE may be better at capturing
exposure that is time-sensitive in terms of the related health effects. For example, driving
through a park may have minimal impact on women’s activities, compared to an encounter with
an urban park during a leisure walk.
However, among all BE characteristics measured, this study found stronger correlations
between the static RNB and dynamic KDE estimates for those measures that required summaries
within administrative boundaries (e.g., census block group level walkability) or raster surfaces
(e.g., 250 m x 250 m NDVI grids) types of data inputs. For measures that relied on distinct
features in space like lines (e.g., % tree cover along a street segment) or polygons (e.g., area of
parks and open space), correlations were only low to moderate. This finding is in line with a
previous study which also reported similarities in exposure estimates across different methods
differed by data input types, with a focus on dynamic methods (Jankowska et al. 2021).
Across the two pregnancy and the early postpartum periods, the results showed that the
correlations of exposure estimates further decreased during the 1
st
trimester and at 4-6 months
postpartum compared to the 3
rd
trimester. This may be partially caused by the residential
mobility (10 of 62 women changed their place of residence during these periods). Additionally,
this decrease may also be explained by changes in time-activity and mobility patterns across the
three study periods, since women in this study spent less time at commercial and service
locations and performed more vehicular trips during the 3
rd
trimester, according to the results
presented in Chapter 2.
Lastly, the ICC analysis results indicated most of variance in exposure estimates for a BE
measure were due to between-participant differences (i.e., differences in person-mean exposure

85
estimates). This finding suggests that the static method may introduce larger exposure
measurement errors when the research goal is to examine the day or within-day level association
between the BE and maternal and infant health outcomes, compared to the person level.
3.4.3. Exposure Misclassification Introduced by the Static Method to Due to Daily Mobility
This study’s results showed exposure misclassification could potentially occur if the
static methods were relied on to estimate exposure levels, especially for days when individuals
are highly mobile. Exposure measurement error or misclassification can weaken statistical power
to detect associations and potentially bias observed risk estimates in health studies (Zeger et al.
2000). This concern is warranted by recent studies which have increasingly reported mixed
results on associations between neighborhood green space and pregnant women’s activities and
health outcomes (Anabitarte et al. 2020; Banay et al. 2017; Nichani et al. 2016; Porter et al.
2019).
Despite the overall dependence of exposure misclassification on daily mobility, the
evidence for measures of street-level greenness exposure, street intersection density, and
neighborhood walkability was less consistent. It may be that these measures have very low
variability in women’s usual daily activity spaces. As a result, even if more activity locations
were visited during high mobility days, the exposure estimates around these locations would only
slightly differ from static estimates. This lack of variability may be further amplified by the fact
that women in our sample had relatively small daily spatial footprints (a few urban blocks in LA,
CA) and as a result, the differences in values between BE data in adjacent spatial units (e.g.,
walkability index score of US census block groups) would be further reduced.

86
3.4.4. Study Strengths and Limitations
To the best of my knowledge, this is the first study that examines GPS-based dynamic
exposures to BE characteristics of pregnant women across the pregnancy and early postpartum
periods. A major strength is the estimation of daily exposure to neighborhood BE characteristics
by repeatedly collecting highly resolved smartphone location data across the 1
st
and 3
rd
trimesters
of pregnancy and at 4-6 months postpartum. Consequently, this work overcame the recall biases
inherent in self-reported BE exposures and provides insights into longitudinal changes in these
exposures. Additionally, the longitudinal design used for this study allows the examination of
both the variations in BE exposures between the pregnant women and the day-to-day variations
for each woman, which previous studies applying the static method could not do. Moreover, the
study applies both spatial averaging (RB) and time-weighted spatial averaging (KDE) methods to
capture women’s actual exposures to BE characteristics in daily activity spaces. As a result, the
exposure estimates reported in this chapter may have less error than those derived by the static
approach. Lastly, this study is among the first to examine similarities and differences between
BE exposure derived by static and dynamic methods and, to the best of my knowledge, the first
one that focuses on pregnant women (Zhao, Kwan, and Zhou 2018; Jankowska et al. 2021). The
findings of low correlations between static and KDE neighborhood greenness measures have
important implications for future studies examining the association between prenatal
neighborhood greenness and maternal and infant health outcomes.
This study also has a few limitations. First, the GPS data that was collected has some
missingness. To mitigate its impacts on analyses, an effort was made to impute GPS data using
existing information and ruled out the existence of diurnal patterns for the remaining missing
segments (it was roughly invariant throughout the day). Despite these efforts, there are other
factors that may still potentially bias this study’s exposure measurements. For instance,

87
missingness patterns of GPS data may be correlated with spatial context (e.g., tall buildings,
trees) that could obstruct receiver signals. As a result, BE characteristics (e.g., proximity to
public transit stops) in these spatial contexts may not be captured.
Second, this study was subject to the uncertain geographic context problem (UGCoP),
which refers to the uncertainties of environmental contexts that influence health behaviors and
outcomes (Kwan 2012). Although the GPS-based dynamic method as applied in this study
tackles UGCoP better than the static method by reducing the spatial mismatch between exposure
and outcomes, it is limited in addressing other potential methodological issues such as the choice
of spatial parameters (e.g., buffer sizes). In this study, a 250 m buffer was used to construct
activity spaces given this distance considers multiple pathways on how the BE influences
maternal activities and health. In addition, the correlation analyses were run a second time using
exposure estimates derived with a 100 m buffer. Study results were largely unchanged and thus
not reported.
Third, exposure estimates derived by GPS-based methods were aggregated to the day
level and daily exposures were computed as the averaged time-weighted exposure value within
activity spaces. Other temporal unit (e.g., trip-level, minute-level) can be chosen if future studies
are interested in finer grain within-day relationships between BE exposures and health behaviors.
For example, one can examine the greenness exposures within a time window (e.g., 30-minute)
preceding a walking episode to better understand the time-lagged effects of the greenness
exposure on physical activity behaviors. Also, the BE datasets applied in this study to derive
exposure estimates may be limited in variability, given the used aggregation method to produce
them (e.g., distance to the nearest public transit stop was calculated as the minimum walking
distance in meters between the 2010 population-weighted CBG centroid). Consequently, this

88
may in turn decrease the variability of BE exposure estimates derived and thereby reduce their
statistical power to detect meaningful associations with health behaviors and outcomes of
interest.
Lastly, this work focused on a health disparity group of low-income, Hispanic women, a
population that has been understudied and disproportionally exposed to various environmental
hazards. Thus, the results may not be generalized to pregnant women in other regions or SES
groups; nevertheless, they shed light on an important population, and they may pave the way for
future studies to examine women’s BE exposures and health outcomes during pregnancy and
postpartum.
3.5. Conclusions
Pregnancy and early postpartum are critical periods of exposure, and this study has
shown that BE exposures will likely vary over this journey for many days, with potential
implications on both short- and long-term maternal and child health. More importantly, this
study have demonstrated that the residential-based static methods commonly used in prior
studies can introduce exposure measurement error, the extent of which differs by type of BE
measure studied. Therefore, future studies examining the impacts of the BE on maternal and
infant health should consider the spatiotemporal movement patterns of the pregnant women in
exposure measurements as they will directly influence the ability to detect meaningful
relationships.

89
Chapter 4 The Association between Built Environment Characteristics and
Physical Activity during Pregnancy and Early Postpartum
An increasing number of studies has examined the relationship of residential BE characteristics
to pregnant women’s PA (Kershaw et al. 2021; Nichani et al. 2016; Porter et al. 2019;
Richardsen et al. 2016). Yet, very little work has been done to understand this relationship at
day-level using GPS-based dynamic BE exposures. Applying 4-day smartphone location and
accelerometry-assessed movement data collected from a group of Hispanic, predominantly low-
income pregnant women during the 1
st
and 3
rd
trimesters, and at 4-6 months postpartum. This
chapter examines the effects of women’s daily GPS-based greenness, parks and open space, and
walkability exposures on their day-level MVPA outcomes.
4.1. Related Work
Pregnancy PA is an important risk factor for short- and long-term maternal and infant
health outcomes. Past studies have associated physical inactivity with increased gestational
weight gain and health conditions including diabetes, heart diseases, and stress and anxiety
(Currie et al. 2014; Daley et al. 2007; Evenson et al. 2014). Evidence indicates Hispanic women
had disproportionally high rates in pregnancy-related obesity risk and health outcomes and were
overall less likely to meet the PA guidelines during pregnancy than non-Hispanic white
population (Brawarsky et al. 2005; Chasan-Taber et al. 2008; Evenson and Wen 2010; Headen et
al. 2012). Understanding the drivers of PA behaviors among Hispanic women is a critical step
towards reducing disproportionate obesity risks and health consequences borne by this group.
Increasingly, studies have begun to examine the impact of the BE on pregnant women’s
PA. In this realm, studies have reported positive relationships of neighborhood greenness, parks
and open space, street connectivity, and walkability on women’s PA outcomes during pregnancy

90
and postpartum (Kershaw et al. 2021; Nichani et al. 2016; Porter et al. 2019; Richardsen et al.
2016). In addition, the evidence indicates that presence of sidewalks and access to parks and
recreational facilities are associated with higher PA outcomes among Hispanic adults in the US
(Cronan et al. 2008; Fields et al. 2013; Larsen et al. 2013; Mama et al. 2015). Nevertheless, to
the best of my knowledge, no studies have focuses on the relationship between BE and PA
during pregnancy in Hispanic and low-income women groups.
Moreover, most aforementioned studies relied on one residential location and time point
to estimate exposures to BE characteristics (most inferred from addresses provided on birth
certificates or questionnaires after delivery). This “static” method does not capture women’s
dynamic exposure to BE characteristics in non-home activity spaces (non-home activity
locations, walking trips) and day-to-day changes in BE exposure resulting from household,
occupational, and recreational activities (e.g., grocery shopping, commuting to work, visiting a
park), and ignores residential mobility and dramatic variations in activities across the pregnancy
and postpartum periods (Matthews and Yang 2013; Perez, Ruiz, and Berrigan 2019; Varshavsky
et al. 2020). As a result of these limits, studies applying the static method may fail to capture the
“true causally relevant” (Robertson and Feick 2018) BE characteristics that exert contextual
influences on pregnant women’s PA and bias study results. Few studies have examined the day-
level BE-PA associations in pregnant women.
Furthermore, temporal, individual, and neighborhood factors may modify the associations
between day-level BE-PA associations in pregnant women. Some studies have reported positive
associations between individual sociodemographic measures such as being employed, single
parity, and low body mass index (BMI) and women’s PA during pregnancy and postpartum
(Borodulin, Evenson, and Herring 2009; Hausenblas et al. 2011; Redmond, Dong, and Frazier

91
2015; Wit et al. 2015). Evidence provided by two studies also suggests that neighborhood social
environment characteristics such as perceived safety and cohesion have positive effects on PA
outcomes during pregnancy (Evenson et al. 2009; Perez et al. 2016). In addition, many studies
have reported pregnant women’s PA behaviors differed by weekdays versus weekends, and
pregnancy and postpartum periods (Berntsen et al. 2014; Borodulin, Evenson, and Herring 2009;
da Silva et al. 2019; Jenum et al. 2013; Schmidt et al. 2006; Sinclair et al. 2019). As a result,
examining the interactive effects of dynamic BE exposures and these potential moderators may
help to understand the important nuances (i.e., at when, for pregnant women of what
sociodemographic characteristics) in its associations with day-to-day PA behaviors. Yet, in this
arena, studies remain very limited.
To fill the three aforementioned gaps, the study described in this chapter applied
smartphone GPS and accelerometry data collected from a group of Hispanic, predominantly low-
income women from MADRES personal and real-time data sampling study to answer the
following research questions:
1. Are women’s daily exposure to BE characteristics associated with their day-level PA
outcomes during the 1
st
and 3
rd
trimesters of pregnancy and at 4-6 months postpartum?
2. Do the BE exposure measurement methods applied (i.e., static and dynamic) influence the
detection of the day-level associations between BE exposures and PA outcomes?
3. Do the day-level associations between BE exposures and PA outcomes, if any, differ by
temporal factors, individual sociodemographic, and neighborhood characteristics?
This study hypothesized that the study population would have higher day-level PA
outcomes during days when the women were exposed to areas with higher greenness, better park
access, and higher walkability. Additionally, this study explored whether the detection of the

92
day-level associations between PA outcomes and BE exposures are influenced by the residential
versus activity space methods applied to derive these exposures. Lastly, this study explored
whether the day-level associations between BE exposures and PA outcomes are moderated by
the pregnancy and postpartum periods, weekdays versus weekend days, maternal parity, pre-
pregnancy BMI categories, women’s employment status, and neighborhood deprivation,
cohesion, and safety.
4.2. Methods
4.2.1. Study Design and Participants
The data used for this study comes from the Real-Time and Personal Sampling sub-study
conducted in the USC MADRES Center. This study uses an intensive longitudinal, observational
panel study design and examines the daily effects of environmental exposures and social
stressors on maternal pre- and post-partum obesity-related biobehavioral responses. A total of 65
Hispanic, predominantly lower income mothers were drawn from the larger MADRES
prospective cohort study, which recruited participants from prenatal care providers serving
predominantly medically-underserved populations in Los Angeles, California, including two
non-profit community health clinics, one county hospital prenatal clinic, one private obstetrics
and gynecology practice, and through self-referral from community meetings and local
advertisements (O’Connor et al. 2019). To be eligible for this sub-study, a participant needed to
be >18 years old with a singleton pregnancy and be at less than 30 weeks’ gestation at time of
recruitment. In addition, participants who were HIV positive, had physical, mental, or cognitive
disabilities that prevented participation, or were currently incarcerated were excluded from the
study (O’Connor et al. 2019). Recruitment of 65 women occurred on a rolling basis between
2016 and 2018 and is described in further detail in O’Connor et al. (2019). The USC Institutional

93
Review Board approved all study procedures and participants signed an informed consent before
enrolling into the study.
4.2.2. BE Measures
4.2.2.1. Location Information using GPS
GPS data were continuously collected from 65 study participants at 10-s intervals for four
days (two weekdays and two weekend days) during the 1
st
and 3
rd
trimester and at 4-6 months
postpartum. To enable collecting highly resolved and encrypted GPS data collection, MADRES
researchers designed a custom smartphone application (madresGPS app) for Android operating
systems. Study coordinators configured the application on dedicated study smartphones
(Samsung MotoG phone) to record geographic coordinates and geolocation/motion metadata
(O’Connor et al. 2019). The application logged instantaneous GPS location and sensor data
every 10 s from the smartphone’s multiple built-in location finding features (cell tower
triangulation, Wi-Fi networks, and GPS) and motion sensors. Along with the timestamp, the
application recorded metadata such as the number of satellites in use/view, geolocation accuracy,
source of GPS, velocity (if GPS source), and network connection status (if network source).
The 10-s epoch GPS location data was processed using a customized algorithm in SAS
version 9.4 (SAS Institute Inc). The algorithm subsets GPS data to the 4-day designated
monitoring period (during device set up and return), selects better quality data from one of the
two GPS data sources (GPS/network), smooths extreme outliers, and aggregates data at the 10-s
level. Missing GPS data was imputed using a customized imputation algorithm that identified
participants’ time periods at home during nighttime using sleep and waking times they self-
reported prior to each study period to help the technology team configure suitable timing and
frequencies for the EMA survey (O’Connor et al. 2019). The time periods spent at home were

94
identified when participants reported a home day (i.e., all EMA survey prompts within the day
reported physical context as either “Home-Indoor” or “Home-outdoor”). The entire workflow of
the missingness imputation process is discussed in Chapter 2. After GPS processing and
imputation, flags were created which labelled days with <6 hours of GPS data (post-imputation)
as invalid days, after considering missing data patterns.
4.2.2.2. Delineation of Contextual Units
Activity Space Metrics
The processed and imputed GPS data was separated into person-day levels and daily
activity spaces were constructed via both the RB and KDE methods (see Figure 3.1 in Chapter 3)
using ArcGIS Pro 10.7.1 (Esri, 2021). The two methods complement one another and capture
pregnant women’s daily exposure to BE characteristics along women’s daily travel paths (via
RB) and around activity locations where they spent most of their time (via KDE).
To construct RBs, consecutive GPS points were joined into lines and buffered using a
250 m radius. This distance corresponds to several blocks in urban Los Angeles and was selected
to approximate two hypothetical mechanisms on how BE characteristics could potentially
influence health behaviors and outcomes (Yi et al. 2019): the viewshed (e.g., seeing a park when
walking on a street) and their awareness (e.g., knowing a recreational facility is located nearby).
Additionally, 250 m RBs were calculated and classified into three categories (low mobility:
<33%, medium mobility: 33-66%, high mobility: >66% of sizes) to represent participants’
mobility during a given day.
Additionally, a KDE approach was applied to generate time-weighted activity grids using
50 by 50 m cells with a bandwidth of 250 m (to match the RB radius). Each grid was assigned a

95
normalized time weight (range from 0 to 1) based on the percentage of time staying put on a
given day, with total weights adding up to 1.
Home Residential Neighborhood
To identify residential locations, a kernel-based stay-trip detection algorithm was applied
to 4-day GPS data (per combinations of participants and three study periods) to identify activity
locations visited and total time spent at each location (Thierry, Chaix, and Kestens 2013). For
this study, the women’s activity location with the longest stay duration during the 3
rd
trimester
was assigned as the residential location. This location was selected as most similar to one that
was used in a typical approach (i.e., assessed exposures based on addresses provided on birth
certificates or questionnaires after delivery and assumed these exposures represent the entire
prenatal period). Details of the stay-trip detection and home identification were documented in
Chapters 2 and 3, respectively.
The identified 3
rd
trimester residential locations were then used to derive 800 m RNBs via
the Service Area Analysis tool in ArcGIS Pro 10.7.1 (Esri, 2021). The network buffer better
captures the area to which a pregnant woman could realistically travel from the home residence.
A 800 m radius, which corresponds to a 5-10-min walk, is recommended for populations with
relatively restricted mobility (James et al. 2014; Frank et al. 2017). Additionally, a sensitivity
analysis was performed using a larger radius (e.g., 1,600 m) and the resulting BE exposure
estimates differed only slightly.
4.2.2.3. BE Measures
Four BE measures capturing daily exposures to neighborhood greenness, parks and open
space, and walkability were derived using the KDE, RB, and RNB methods, respectively. These
measures included percent green space along walkable roads, distance to the nearest park

96
entrance, parks and open space exposure, and walkability index score, all of which have been
reported by previous studies to be associated with various human activities and health outcomes,
particularly among pregnant women (Banay et al. 2017; Besser and Dannenberg 2005; Jiang et
al. 2016; Pickard et al. 2015; Porter et al. 2019; Pretty et al. 2005; Tsai, Davis, and Jackson 2019;
Saelens et al. 2014;). All aforementioned BE measures along with data source, resolution, and
processing steps and interpretation are summarized in Table 3.1 in the previous chapter.
RNB and RB exposures were calculated by summarizing values of the respective BE
measures (e.g., mean NDVI, total areas of parks and open space) within the boundaries of the
corresponding spatial units (i.e., 800 m residential network buffer, 250 m route buffer). KDE
exposures were calculated by weighting grid-based values of the respective BE measures with
time spent (e.g., multiplying walkability index score value of the grid by percentage of time
spent). Exposures were calculated using Eq. 3.1 for the RNB or RB methods and using Eq. 3.2
for the KDE methods, as illustrated in Chapter 3.
4.2.3. PA Outcomes
A wGT3X-BT ActiGraph accelerometer was used to measure women’s PA during
pregnancy and early postpartum. Women were instructed to wear the accelerometer for four
consecutive days (two weekdays and two weekend days) during each study period. The
accelerometer was attached with an adjustable belt and worn all the time on the right hip except
when sleeping, bathing/showering, or swimming (O’Connor et al. 2019). Body movement data
were collected by the accelerometer in activity count units for each 10-s epoch. Non-wear times
were defined as >60 continuous min of zero activity counts and non-valid days were defined as
<10 h of wearing time (Troiano et al. 2008), both of which were removed from the analyses. To
be consistent with national surveillance data, MVPA, an intensity of PA that is sufficient to

97
reduce the risk of many adverse health outcomes (Piercy et al. 2018), was identified using the
Freedson prediction equation above four metabolic equivalents of task (i.e., activities that
accumulate metabolic rate ≥4 times of the resting metabolism, such as brisk walking, based on
the Freedson prediction equation; Troiano et al. 2008; Bell et al. 2013; Harrison et al. 2011).
Day-level PA outcome variable was created by summing total number of 10-s MVPA epochs
occurred within an observation day and converted to a unit of minutes.
4.2.4. Covariates
Individual sociodemographics included enrollment age, education (recoded as some
college or graduate degrees versus high school diploma or less), and maternal parity (recoded as
first born versus second or greater birth), which were assessed in prenatal interviewer-
administered questionaries with the women (Bastain et al. 2019). Body Mass Index (BMI)
categories (recoded as normal versus overweight versus obese) were also calculated based on
height and weight measured during pre-natal visits, according to Center for Disease Control and
Prevention guidelines (Bastain et al. 2019). Working status was collected via questionnaire
during the 1
st
and 3
rd
trimester, and at 6 months postpartum (Bastain et al. 2019). Participants
self-reported neighborhood cohesion and safety scores in questionnaires at the 2
nd
trimester
(chosen to represent pregnancy) and for 6 months postpartum (Sampson, Raudenbush, and Earls
1997), and the Areal Deprivation Index Score from the Neighborhood Atlas dataset produced by
the University of Wisconsin-Madison was used to represent neighborhood-level SES of
participants (Kind and Buckingham 2018). The score was assigned to participants’ residences
based on the 2010 census block group in which their home residences were situated. Temporal
factors included weekend versus weekday (weekend=1), daily average temperature in degrees
Celsius, and study period (the 1
st
and 3
rd
trimester and 4-6 months postpartum). Daily

98
accelerometer wear time was also included to adjust for individual device-wearing behaviors. A
full list of covariates measures and corresponding data sources is provided in Appendix D.
4.2.5. Statistical Analysis
4.2.5.1. Analytical Samples
A total of 651 accelerometer-measurement days of PA were recorded by the 62 women.
Among them, 210 non-valid days with <10 h of wear time were removed. Examination of
predictors of non-valid day analyses showed that weekend days were more likely to be non-valid
days; all other covariates did not predict valid versus non-valid days. Moreover, 94 data
collection days were removed due to lack of daily BE measures data. These efforts resulted in a
final data analysis sample of 350 days (NParticipant=55).
4.2.5.2. Linear Mixed-Effects Models (LMMs)
To account for the interdependency of the nested data structure in the current study
(Level 1-days nested within Level 2-participants), LMMs with participant-level random
intercepts were applied. To test the necessity of a multi-level model (i.e., within-participant
clustering of day-level MVPA outcomes), ICCs was estimated using a LMM with only random
intercepts (no covariates). Results indicate 41% (ICC=.41) of variation in the day-level MVPA
minutes was between-participant and 59% was within-participant. This result justifies the use of
a multilevel model (Bates et al. 2015).
4.2.5.3. Model Building Strategy and Covariates
All four BE measures were person-mean centered to disentangle the between-subject
(person-level) and within-subject (day-level) effects (Curran and Bauer, 2011). Since the total
parks and open space exposure variable contains a large proportion of zeros (about one third

99
across all observation days), it was recoded into a binary variable (1=had any parks and open
space exposure in an observation day or 0 = no exposure). Additionally, the day-level MVPA
outcome variable was log-transformed to ensure normality since the variable showed a right-
skewed distribution. For each set of four BE measures derived using two activity space methods
(250 m KDE, 250 m RB), LMMs were fitted to test person- and day-level effects of each BE
measure on day-level MVPA minutes.
All models were adjusted for women’s baseline age, education, pregnancy and
postpartum periods, and daily wear time. In addition, additional covariates were tested in
separate univariate models and included in the final model if they reached a significance level of
<.1. These additional covariates included maternal parity, employment status, pre-pregnancy
BMI categories, daily averaged temperature (both linear and quadratic terms were tested to
account for its potential non-linear relationships with daily PA outcomes), weekdays versus
weekends, neighborhood safety and cohesion score, and neighborhood deprivation index score.
Moreover, two-way interaction terms between day-level BE measures and
aforementioned temporal factors, individual sociodemographics, and neighborhood covariates
were entered separately into LMMs to test whether they moderated the associations between BE
measures and day-level MVPA minutes. For each significant interaction (p<.05), predicted
trajectories for MVPA min/d related to BE measures at a specific value of moderator (all
categories if categorical; -1SD, mean, +1SD if continuous) were estimated and visualized, with
the slope estimated and its statistical significance tested (i.e., whether an estimated slope of the
day-level MVPA on a specific BE measure at a specific moderator value is different from zero).

100
4.2.5.4. Sensitivity Analyses
Models were also run for sensitivity analyses by using BE measures derived using
smaller buffers for two activity space methods (100 m RB, 100 m KDE) and the RNB method
(800 m and 1,600 m RNB). In addition, three-level LMMs with random intercepts at person- and
period-levels were run as sensitivity analyses to test whether the random effects of pregnancy
and early postpartum periods would confound the associations between BE exposures and PA
outcomes found in two-level LMMs. The R 4.0.2 (R Core Team, 2020) and lme4 package
(version 1.1-27.1) were used for LMM (Bates et al. 2015), and the interaction package (version
1.1.5) was used for simple slope analyses. Since the outcome variable was log-transformed,
exponentiated effect estimates interpreted on a multiplicative scale were reported for all models.
4.3. Results
4.3.1. Descriptive Statistics
The descriptive statistics for PA outcomes, BE predictors, and covariates of the study
participants are summarized in Table 4.1. Participants’ mean age at study entry was 29.00 years
(SD=6.1; range:18.3-45.4). All the participants were Hispanic. Over one-third (34.6%) had some
college or above education. 36.2% were employed during the 1st trimester compared to 39.00%
during the 3rd trimester, and 21.2% at 4-6 months postpartum. At recruitment, 29.1% were
pregnant with their first child, 72.7% were overweight or obese according to their reported pre-
pregnancy BMI. The recruited participants lived in neighborhoods with an average deprivation
index score (on a 1 to 10 scale; where 1=least deprived) of 6.3 (SD=1.8, Range:2.00-9.00). The
average neighborhood safety and cohesion score (on a 1 to 5 scale; where 1=least safe and
cohesive) self-reported by women was 3.00 (SD=.69, Range:1.00-4.4) during the 1st trimester,

101
3.1 during the 3rd trimester (SD=.81; Range:1.2-4.7), and 3.3 (SD=.9, Range:1.4-4.8) at 4-6
months postpartum.
The average MVPA min/d across all three study periods were 30.7 (SD=22.4), which was
higher (p=.71) during the 3
rd
trimester (32.00 min/d; SD=20.4) than during the 1
st
trimester (30.8
min/d; SD=23.5) and at 4-6 months postpartum (29.5 min/d; SD=23.1). Less than half of the
data collection days were on weekends (45.4%), which had lower (p<.01) day-level MVPA (26.8
min/d; SD=19.6) compared to weekdays (34.2 min/d; SD=24.2). The mean average daily
temperature in Celsius was 19.92 °C (SD=4.4; Range:5.2-31.4).
In terms of neighborhood greenness, women were exposed to walkable road networks
with an average 23.3% green space coverage at daily activity locations visited (SD=8.3), which
was lower than the Los Angeles County average of 32.4% (mean percent green space along all
walkable roads in the County). The exposure estimates of the greenness exposure measure varied
slightly using the RB method. As for parks and open space access, women’s mean distance from
any point in the daily activity spaces to the nearest park entrance was 791 m (SD=399;
corresponding to 15-20-minute walk). The estimates differed slightly with the RB method.
Additionally, women were exposed to parks and open space in activity spaces in approximately
three out of every four observation days (73.4%; 257 out of 350 days), according to the KDE
method. This percentage was slightly smaller if measured by the RB method. Lastly, regarding
walkability, the mean walkability index score was 15.00 (SD=1.7), which was 1.5 units higher
than the Los Angeles County average. The walkability exposure estimates differed slightly
between the two activity space methods.

102
Table 4.1. Descriptive statistics for physical activity (PA) outcomes, built-environment (BE)
predictors, and covariates of the study participants.

Overall
(N=55 participants,
N=350 person-days)
1
st
Trimester
(N=47 participants,
N=133 person-days)
3
rd
Trimester
(N=41 participants,
N=106 person-days)
4-6 Months
Postpartum
(N=38 participants,
N=112 person-
days)
Variable Mean(SD) or n(%)
Outcome
Day-level MVPA
minutes
30.73 (22.45) 30.76 (23.48) 31.99 (20.44) 29.48 (23.13)
Predictors
% Green space along walkable roads
RB 23.21 (5.87) 23.73 (5.38) 23.26 (5.91) 22.54 (6.37)
KDE 23.30 (8.27) 24.06 (8.47) 22.80 (7.65) 22.88 (8.58)
Days with parks and open space exposure (yes/no)
RB 254 (72.6%) 97 (72.9%) 81 (76.4%) 76 (68.5%)
KDE 257 (73.4%) 99 (74.4%) 82 (77.4%) 76 (68.5%)
Distance to the nearest park entrance (in m)
RB 868.03 (342.07) 826.44 (323.96) 873.76 (328.80) 912.38 (371.49)
KDE 790.80 (398.47) 748.10 (346.55) 807.72 (424.17) 825.79 (429.16)
Walkability index score (range from 1 to 20)
RB 15.09 (1.39) 15.05 (1.54) 15.24 (1.24) 14.98 (1.35)
KDE 15.01 (1.68) 14.99 (1.78) 15.12 (1.58) 14.95 (1.64)
Covariates
Age 29.04 (6.14) - - -
Education
High school or less 36 (65.45%) - - -
Some
college/Graduate
19 (34.55%) - - -
Parity
Second or greater
birth
39 (70.91%) - - -
First-born 16 (29.09%) - - -
Pre-pregnancy BMI
categories

Normal 15 (27.27%) - - -
Overweight/Obesity

40 (72.73%) - - -
Employment status
a

Unemployed – 30 (63.83%) 25 (60.98%) 26 (78.79%)
Employed – 17 (36.17%) 16 (39.02%) 7 (21.21%)
Missing – 0 0 5

103
Table 4.1. (Cont.)

Overall
(N=55 participants,
N=350 person-
days)
1
st
Trimester
(N=47 participants,
N=133 person-
days)
3
rd
Trimester
(N=41 participants,
N=106 person-
days)
4-6 Months
Postpartum
(N=38
participants,
N=112 person-
days)
Variable Mean(SD) or n(%)
Neighborhood cohesion
and safety score (range
from 1 to 5)
a
– 3.02 (.69) 3.07 (.77) 3.28 (.86)
Missing – 1 0 5
Neighborhood
deprivation index (range
from 1 to 10)
6.34 (1.78)
Missing 2
Daily accelerometer
wearing hours
13.79 (2.63) 13.32 (1.82) 14.19 (3.03) 13.96 (2.95)
Type of day
Weekday 191 (54.57%) 76 (57.14%) 60 (56.60%) 55 (49.55%)
Weekend 159 (45.43%) 57 (42.86%) 46 (43.40%) 56 (50.45%)
Average daily
temperature (°C)
19.92 (4.40) 20.37 (4.36) 20.41 (4.28) 18.95 (4.45)
Missing 9 9 0 0
a
Summary statistic not available since the employment status variable is measured at each study period.
Note: KDE = Kernel Density Estimation. RB = route buffer.

4.3.2. Predictors of Day-Level PA Outcomes in Pregnant Women
LMM results examining within-subject (day-level) associations between BE measures
derived using the 250 m KDE and RB methods and women’s day-level MVPA minutes are
shown in Table 4.2. In both models, results showed that women engaged in more MVPA on
weekdays versus weekend days. In addition, women engaged in more MVPA on days with
higher averaged daily air temperatures and this relationship was linear. Education attainment was
also associated with women’s MVPA min/d so that women who had college or graduate degrees
engaged in less MVPA compared to those who had high school or less education.

104
Table 4.2. Between-subject (BS) and within-subject (WS) effects of exposures to built-
environment (BE) characteristics derived using the 250 m kernel density estimation (KDE) and
250 m route buffer (RB) methods on daily moderate-to-vigorous physical activity (MVPA)
outcomes.
Daily MVPA
minutes(log-transformed)
250 m KDE model 250 RB model
Predictors Estimates (95%CI)1
(Intercept) 8.00*** (3.23 – 19.79) 8.62*** (3.43 – 21.66)
% Green space along walkable roads (BS) 1.01 (.99 – 1.03) 1.03 (1.00 – 1.06)
Distance to the nearest park entrance (BS) 1.01 (.98 – 1.05) 1.04 (1.00 – 1.10)
Walkability index score (BS) 1.05 (.97 – 1.14) 1.18* (1.04 – 1.33)
Daily parks and open space exposure (BS) 1.64 (1.00 – 2.70) 1.47 (.92 – 2.34)
% Green space along walkable roads (WS) 1.00 (.98 – 1.02) 1.01 (.99 – 1.03)
Distance to the nearest park entrance (WS) 1.00 (.94 – 1.05) .99 (.96 – 1.02)
Walkability index score (WS) 1.01 (.89 – 1.15) 1.02 (.94 – 1.12)
Daily parks and open space exposure (WS)2 1.21* (1.01 – 1.46) 1.22* (1.02 – 1.47)
Maternal age .99 (.97 – 1.01) .99 (.96 – 1.01)
Education: Some college/Graduate .68** (.52 – .89) .66** (.50 – .88)
Pre-pregnancy BMI category: Overweight .93 (.69 – 1.25) .95 (.72 – 1.27)
Pre-pregnancy BMI category: Obesity .69* (.51 – .92) .66** (.50 – .87)
Average daily temperature (°C) 1.02* (1.00 – 1.04) 1.02* (1.00 – 1.04)
Type of day: Weekend .81** (.71 – .93) .80** (.70 – .92)
The 3rd trimester day .96 (.81 – 1.13) .95 (.80 – 1.12)
4-6 months postpartum day .90 (.75 – 1.08) .91 (.76 – 1.08)
Neighborhood cohesion and safety score (range from 1 to
5)
1.02 (.91 – 1.14) 1.03 (.92 – 1.15)
Daily accelerometry wearing hours 1.09*** (1.06 – 1.13) 1.09*** (1.06 – 1.13)
* p<.05 ** p<.01 *** p<.001
1 Exponentiated effect estimates interpreting on a multiplicative scale were reported for all models
2 The binary variable was not person-mean centered for the ease of interpretation. The person-mean centered
versions of the variable were also tested, and results remained invariant.

After controlling for all covariates, results showed (see Table 4.2 – 250 m KDE model)
that the women engaged in more MVPA min/d (b=1.21; 95%CI:1.01-1.46; p<.05) on days they
were exposed to parks and open space in their activity spaces. This result did not change in the

105
250 m RB model (see Table 4.2). In addition, women who were exposed to activity spaces with
higher walkability score, on average compared to other women, engaged in more MVPA min/d
(b=1.18; 95%CI:1.04-1.33; p<.05) in the 250 m RB model. Associations were not detected
between other BE variables and women’s day-level MVPA outcomes.
Sensitivity analyses that applied BE variables derived using the 100 m KDE and RB, and
800 m and 1,600 m RNB methods were conducted. These results are summarized in Appendix
G. In all models, both positive associations between daily parks and open space exposure and
MVPA min/d and between daily walkability index score and MVPA min/d were no longer
significant, although the associations still approached statistical significance (p<.1) in the 100 m
KDE and 100 m RB models. In addition, the associations between BE measures derived using
the 250 m KDE and RB and MVPA min/d via three-level LMMs with random intercepts at
person- and period-level (i.e., 1
st
and 3
rd
trimesters and 4-6 months postpartum) were also tested,
and the aforementioned significant results on positive associations between daily parks and open
space exposures and MVPA min/d did not change.
4.3.3. Moderators of Day-Level Associations between BE and PA
The LMM results after adding two-way interactions between within-subject (day-level)
BE measures and a list of temporal factors, individual sociodemographics, and neighborhood
characteristics are visualized in Figure 4.1 and presented in table form in Appendix H. The
simple slope analysis results for each significant interaction term are summarized in Table 4.3 as
well. Overall, these results show the associations between daily BE exposures and MVPA min/d
differed by weekdays versus weekends, late pregnancy versus two other periods, early
postpartum versus two other periods, the average daily temperature, maternal parity numbers,

106

Figure 4.1. The significant interactions between built-environment (BE) characteristics and
temporal factors, individual sociodemographics, and neighborhood characteristics in predicting
day-level moderate-to-vigorous physical activity (MVPA) outcomes.
Note: Timepoints include the 1
st
trimester, the 3
rd
trimester, and 4-6 months postpartum.

107

Figures 4.1. (Cont.)

Note: Timepoints include the 1
st
trimester, the 3
rd
trimester, and 4-6 months postpartum.

108

Figures 4.1. (Cont.)

Note: Three Timepoints = the 1
st
trimester, the 3
rd
trimester, and 4-6 months postpartum.

109
Table 4.3. Simple slope analyses results for significant two-way interactions between built-
environment (BE) exposures and temporal, individual, and neighborhood moderators in
predicting daily moderate-to-vigorous physical activity (MVPA) outcomes.
Interaction Terms Estimates (95%CI)
Activity space walkability x Daily averaged temperature (°C)
Daily averaged temperature: Low (-1SD) .93 (.82-1.04)
Daily averaged temperature: Medium (Mean) 1.01 (.93-1.11)
Daily averaged temperature: High (+1SD) 1.11 (1.00-1.23)*
Activity space walkability x Weekends
Weekend days 1.14 (1.01-1.28)*
Weekdays .91 (.79-1.04)
Activity space walkability x The 3rd trimester
The 3rd trimester days .96 (.93-1.00)
Non 3rd trimester days 1.06 (.99-1.14)
Activity space walkability x 4-6 months postpartum
The 4-6 months postpartum days 1.12 (.95-1.31)
The 4-6 months postpartum days .81 (.63-1.03)
Activity space walkability x Pre-pregnancy Body Mass Index (BMI) category
Pre-pregnancy BMI category: Overweight .99 (.77-1.27)
Pre-pregnancy BMI category: Obese 1.26 (1.02-1.56)*
Pre-pregnancy BMI category: Normal .85 (.64-1.13)
Activity space walkability x Maternal parity
Maternal parity: 1st born 1.18 (1.01-1.38)*
Maternal parity: 2nd or greater birth .95 (.86-1.06)
Activity space park exposure x Neighborhood safety and cohesion
Neighborhood cohesion and safety: Low (-1SD) 1.56 (1.20-2.03)***
Neighborhood cohesion and safety: Medium (Mean) 1.27 (1.05-1.52)*
Neighborhood cohesion and safety: High (+1SD) 1.03 (.81-1.30)
Activity space walkability x Neighborhood safety and cohesion
Neighborhood cohesion and safety: Low (-1SD) 1.16 (1.03-1.32)*
Neighborhood cohesion and safety: Medium (Mean) 1.02 (.94-1.11)
Neighborhood cohesion and safety: High (+1SD) .89 (.78-1.02)
Activity space park proximity x Neighborhood safety and cohesion
Neighborhood cohesion and safety: Low (-1SD) 1.02 (.97-1.07)
Neighborhood cohesion and safety: Medium (Mean) .98 (.95-1.02)
Neighborhood cohesion and safety: High (+1SD) .95 (.90-.99)*
* p<.05 ** p<.01 *** p<.001
Note: CI = Confidence Interval

110
pre-pregnancy BMI categories, and participant-reported neighborhood cohesion and safety
scores.
In terms of temporal factors, Figures 4.1a-b show that the effects of activity space
walkability on women’s MVPA min/d differed by the daily averaged temperature (b=1.02;
95%CI:1.01 – 1.04; p<.01) and weekdays versus weekends (b=1.25; 95%CI: 1.04-1.51; p<.05).
The simple slope analysis results (Table 4.3) show that women engaged in more MVPA when
they were exposed to activity spaces with higher walkability index score than usual in weekends
(b=1.11; 95%CI: 1.00-1.23; p<.05) and warmer days (b=1.14; 95%CI: 1.01-1.28; p<.05).
Moreover, Figures 4.1c-d show how the effects of activity space park proximity on women’s
MVPA differed by the 3
rd
trimester versus two other time points (b=1.10; 95%CI:1.01-1.20;
p<.05) and the effects of activity space walkability on women’s MVPA differed by 4-6 months
postpartum versus two other time points (b=.72; 95%CI:.54-.97; p<.05). However, none of the
predicted slope values for each stratum reported in Table 4.3 was significant.
Turning next to individual sociodemographics, the effects of activity space walkability on
women’s MVPA differed by the first born versus second or greater birth (b=1.23; 95%CI:1.02-
1.49; p<.05), so that the women who had given birth to their first child engaged in more MVPA
(b=1.18; 95%CI:1.01-1.38, p<.05; see Figure 4.1f and Table 4.3) on days when they were
exposed to activity spaces with higher walkability. Additionally, the effects of activity space
walkability (b=1.49; 95%CI:1.04-2.12; p<.05) on women’s MVPA differed with pre-pregnancy
BMI categories. According to Figure 4.1e and Table 4.3., Women who were obese engaged in
more MVPA on days when they were exposed to activity spaces with higher walkability index
scores (b=1.26; 95%CI:1.02-1.56; p<.05) than usual.

111
Lastly, the results show that neighborhood cohesion and safety moderated the effects of
BE measures on women’s MVPA for measures including park proximity, park coverage, and
walkability. The predicted trajectories for MVPA min/d related to each measure were stratified
by -1SD, mean, and +1SD of neighborhood cohesion and safety score (see Figures 4.1g-i), but
the positive associations between women’s daily BE exposure and MVPA were only significant
among low neighborhood cohesion and safety score groups. For example (see Figure 4.1h),
women who reported their neighborhoods as less safe engaged in more MVPA (b=1.16;
95%CI:1.03-1.32; p<.05) on days when they were exposed to activity spaces with higher
walkability index scores than usual. Similar interpretations can be made for interaction effects as
visualized in Figures 4.1g and 4.1i.
4.4. Discussion
This study used highly resolved smartphone location data collected from a sample of 55
pregnant Hispanic women in urban Los Angeles to measure their dynamic daily exposures to BE
characteristics during the 1
st
and 3
rd
trimesters and at 4-6 months postpartum and examined
relationships of these exposure measures to women’s day-level accelerometer-assessed PA. This
study also examined whether the static versus activity space exposure measurement methods
applied influence the statistical power in detecting the day-level associations between BE
exposures and PA outcomes. Lastly, the study explored the interactive effects of BE measures
applied and the temporal-, and individual- and neighborhood-level moderators in predicting day-
level PA outcomes. The results provided evidence of within-day (day-level) associations
between BE exposures and women’s PA outcomes during and after pregnancy, particularly for
those of low SES or racial/ethnic minority groups. Finally, by applying GPS-based BE exposure
assessment methods, this study was able to reduce the potential spatial misclassification biases

112
associated with prior studies, which measured BE exposures at residential locations at a single
point in time (usually near giving birth). The significance of these results is discussed in more
detail below.
4.4.1. Day-Level Associations between BE and PA
The positive daily association found between 250 m activity space-based (both KDE and
RB) parks and open space exposure and pregnant women’ PA outcomes can be added to prior
studies reporting mixed results. For example, Nichani et al. (2016) found the residential green
space exposure was not associated with PA during pregnancy, while Richardsen et al. (2016)
reported women who lived in neighborhoods with good access to recreational areas accumulated
more MVPA compared to women who had limited access. In sensitivity analyses of this study,
residential parks and open space exposure derived using the 800 m and 1,600 m residential
(RNB) methods were not associated with women’s day-level MVPA outcomes during pregnancy
and early postpartum. Thus, the discrepancy in results observed in different studies might be
attributed to the static, residential-based green space exposure assessment approaches used both
studies because these approaches might not capture women’s exposure in non-residential
locations and therefore biased the study results.
Additionally, daily associations between activity space parks and open space exposure
and pregnant women’s MVPA were not found for models that applied 100 m KDE and RB BE
measures. This suggests that the results are sensitive to the choice of activity space configuration
parameters (i.e., buffer sizes). The choice of 100 m to represent the viewshed exposure along
daily movement paths and the null association suggest roadside exposure alone did not influence
pregnant women’s PA behaviors. In this realm, Kwan (2012) points to the existence of uncertain
geographic context problem (UGCoP) in health studies, which refers to the uncertainties of

113
contextual exposures that exert influences on health behaviors or outcomes. Among the few
studies that explored this issue, Laatikainen et al. (2018) studied 844 adults in Finland and found
the association between BE exposure and their mental health and wellbeing differed by activity
space method applied. In a study of 403 adults in Guangzhou, China, Zhao et al. (2018)
demonstrated that the associations between the BE and obesity were significantly influenced by
seven activity space definitions. This study further demonstrates the presence of UCGoP in
associations between the relationships of BE exposures and PA outcomes.
Moreover, given the multilevel modeling approach applied, the positive association found
in the study (i.e., days with parks and open space exposure were associated with more daily
MVPA) should be interpreted as the effect of women’s day-to-day variations in parks and open
space exposure compared to their usual exposure on their daily PA outcomes. This is different
from the large population-based studies (Nichani et al. 2016; Richardsen et al. 2016; Porter et al.
2019) which focused on examining the influences of population-level differences in residential
neighborhood park exposure (e.g., women with park access versus those without) on pregnant
women’s PA outcomes. The evidence on the day-level relationship between BE exposure and PA
outcomes of pregnant women is still limited, future studies need to continue exploring this
relationship at the day-, within-day- or minute- level to further elucidate the different ways in
which the temporality of environmental exposures might influence women’s PA behaviors
during and after pregnancy.
4.4.2. Differences in Day-Level BE-PA Associations by Temporal Factors
This study also found that weekdays versus weekend days, and the pregnancy and early
postpartum periods moderated the day-level association between BE exposures and MVPA
outcomes in pregnant women. In terms of pregnancy and postpartum periods, the day-level

114
effects of activity space walkability on women’s MVPA differed by postpartum (i.e., 4-6 months
postpartum) versus pregnancy, although predicted slopes for both periods were insignificant.
According to the interaction plot, a trending positive association between daily activity space
walkability and day-level MVPA was found during pregnancy periods. This finding was
consistent with Porter et al. (2019), in which authors found the residential neighborhood
walkability assessed by the environmental audit tool was associated with both overall and
recreational PA during pregnancy. Moreover, the results reported in this chapter demonstrate the
importance of having access to a walkable environment at non-home locations or along daily
paths to promote pregnant women’ PA behaviors.
However, these results also show an inverse association between walkability and
women’s MVPA at 4-6 months postpartum, which was contrary to the original hypothesis
specified for this study and the results of the aforementioned prior study. The walkability
measure used in this chapter was directly adopted from the walkability index score created by the
US Environmental Protection Agency (EPA), which includes components such as street
intersection density and distance to the nearest public transit stop (Pickard et al. 2015). As a
result, the walkability measure used in this chapter may be better at predicting the utilitarian type
of PA than the recreational type. The result in Chapter 2 also reported that women in the
MADRES sample performed less pedestrian-based trips at 4-6 months postpartum. As a result,
the utilitarian PA in the MADRES sample may have significantly decreased at the postpartum
period and consequently attenuated the predictive powers of that walkability measure that was
used here.
Interestingly, in another interaction analysis that compared the effects of activity space
walkability on women’s MVPA at the late pregnancy (3
rd
trimester) versus two other timepoints,

115
the results showed positive associations between walkability and MVPA that was nearly
significant during the 3
rd
trimester. This finding, together with the non-significant relationship
between walkability and MVPA across two pregnancy trimesters reported in the prior paragraph,
suggests activity space walkability may be especially critical for pregnant women to maintain PA
during late pregnancy, when they tend to feel less comfortable being active in outdoor
recreational places and have mobility (Evenson et al. 2009).
The positive association reported between weekend activity space walkability and
women’s MVPA may be due to women having more time and opportunities (e.g., family
activities, events) to facilitate PA during weekend days (Evenson et al. 2004), which results in
higher PA on these days. In addition, the finding that the positive association between activity
space walkability and PA only existed during warmer days suggested BE itself, as a part of the
physical environment, may not influence pregnant women’s PA alone, and therefore it is
important to incorporate other time-varying physical environment factors such as weather and air
pollution in the efforts of disentangling the influences of multiple environmental contextual
factors on women’s PA during pregnancy and postpartum.
4.4.3. Differences in Day-Level BE-PA Associations by Individual Sociodemographics and
Neighborhood Characteristics
The day-level association between BE exposure and women’s MVPA also differed by
maternal parity number, pre-pregnancy BMI category, and women’s self-reported neighborhood
cohesion and safety score. When it comes to daily activity space walkability, the results showed
that its positive association with day-level MVPA was limited to those women who had given
birth to their first child. It may be that women who had two or more children were involved in
more care-giving activities than those who had their 1
st
born; as a result, their PA behaviors may
be less susceptible to changes in BE characteristics (Borodulin, Evenson, and Herring 2009).

116
Additionally, Melzer et al. (2010) in a descriptive study reported a larger proportion of
primiparous women were engaged in >30 min per day of MVPA than multiparous women,
although the differences were not significant. Together, these factors may explain the null
association between exposure to high walkability areas and PA outcomes among women who
had two or more children in the current study.
Additionally, the results in this study indicated that the associations between activity
space walkability and daily MVPA were different for obese women versus normal weight
groups, but not different for overweight women versus normal weight groups. Specially,
exposures to activity spaces with higher walkability were associated with significantly higher
day-level MVPA in obese women during pregnancy and early postpartum. A longitudinal study
reported women who were obese before pregnancy had the largest decreases in PA during early
pregnancy compared to one-year prior (Fell et al. 2009). In addition, a qualitative study that
interviewed 14 overweight and obese pregnant women in UK reported the lack of sidewalks in
local neighborhoods were one of the major obstacles for their PA behaviors (Weir et al. 2010).
Together, these results suggest walkability as a BE feature could mitigate the negative impacts of
being obese on women’s PA outcomes during pregnancy and postpartum.
Moreover, the women’s self-reported neighborhood safety and cohesion score modified
the positive effects of park proximity, exposure, and walkability on women’s PA outcomes.
Specifically, all these relationships were only significant for women who lived in less safe and
cohesive neighborhoods. Prior studies have reported a negative relationship between
neighborhood safety and pregnant women’s PA (Evenson et al. 2009; Laraia et al. 2007).
Therefore, it may be that the exposure to favorable BE features around non-home activity

117
locations and along travel routes create additional PA opportunities for those who were not
active in their own neighborhoods given the safety concerns.
4.4.4. Study Strengths and Limitations
To the best of my knowledge, this is the first study that examines GPS-based dynamic
exposures to BE characteristics and accelerometry-based PA in women during the pregnancy and
early postpartum periods. A major strength is the estimation of GPS-derived activity space
exposure to greenness, parks and open space, and walkability, and the objective assessment of
PA outcomes by repeatedly collecting highly resolved smartphone location and accelerometry
data across the 1
st
and 3
rd
trimesters of pregnancy and at 4-6 months postpartum. Consequently,
the work overcame recall biases inherent in self-reported BE exposures and PA outcomes and
provided insights into longitudinal changes in both exposures and outcomes. Additionally, the
longitudinal design used for this study allows us to examine both the between-subject effects
(variations in BE exposures between pregnant women) and the within-subject effects (day-to-day
variations within women) of exposure to BE characteristics on pregnant women’s PA outcomes.
The work reported in this chapter also applies spatial averaging (RB) and time-weighted spatial
averaging (KDE) methods to capture women’s activity space exposures to BE characteristics and
study the relationships of these dynamic exposure to pregnant women’s PA outcomes. As a
result, this study was expected to reduce the spatial misclassification between BE exposures and
PA outcomes and thereby mitigate the UGCoP, a common issue in residential-based studies.
Finally, this study is among the first to examine the effects of person- and day-level BE exposure
and its interactions with temporal factors, individual sociodemographics, and neighborhood
characteristics in predicting PA outcomes of pregnant women, and, to the best of my knowledge,
the first one to focus on a low-income Hispanic population. The findings of a positive association

118
between parks and open space exposures and walkability in daily activity spaces and women’s
PA outcomes, as well as the moderating roles of weekdays versus weekends, pregnancy and
postpartum periods, average daily temperature, maternal parity, pre-pregnancy BMI, and
neighborhood cohesion and safety, can provide further guidance on formulating effective PA
promotion strategies for pregnant women such as targeting at at-risk groups or days to improve
maternal and infant health outcomes after pregnancy, particularly among understudied health
disparity populations.
This study also has a few limitations. First, in addition to the GPS missingness issue and
its potential influences on BE exposures that was discussed in Chapter 3, the accelerometry data
also had missingness issues (211 of 655 data collection days were invalid according to the ≥10 h
of wearing threshold. To reduce the potential biases this may bring on the study results (i.e.,
missing not at random), individual and temporal covariates (e.g., weekend days) that
significantly predicted the day-level validity of accelerometry data were controlled in the final
statistical models, and the main results did not change. Despite this effort, there may exist other
unknown factors that may also influence the day-level validity of accelerometry data and thereby
bias the study results. Second, this study was unable to differentiate recreational PA (e.g., leisure
walks in a park) from utilitarian PA (e.g., commuting walks to workplaces or for grocery
shopping) using the accelerometer data collected. Without fully separating these two types of
PA, the mismatch between the BE exposure and PA outcome may still occur, as different BE
characteristics may be more related to a specific type of PA (Jankowska, Schipperijn, and Kerr
2015; Sallis et al. 2012).
Third, due to the cross-sectional nature of analyses in this study and the lack of
information in the decision-making processes of individual PA scenarios, the results may be

119
subject to selective daily mobility bias, which refers to the circumstance that one is unsure
whether an exposure (e.g., saw a park) led to women’s PA activity (e.g., a walking session) or
the exposure (e.g., a park visit) is in fact part of a pre-planned activity (e.g., a family outdoor
activity session). Future studies may mitigate this bias by excluding anchor points (places which
people organize their daily activities such as home, work, grocery stores) from environmental
exposure measurement so that the effects of novel exposures on behaviors can be uncovered
(Chaix 2013). Third, this study did not measure the quality of parks and open space, which has
been associated with park use behaviors of predominantly Hispanic communities and thus may
mediate or moderate the associations explored in this study (Dolash et al., 2015). Future studies
are recommended to measure other aspects of parks and open space accessibility measures (e.g.,
amenities, accessibility, maintenance) and examine how these additional measures influence
pregnant women’s PA behaviors.
Lastly, this study focused on a health disparity group of low-income, Hispanic women, a
population that has been understudied and has disproportionally low PA outcomes compared to
non-Hispanic white population. Thus, the study results may not be generalized to pregnant
women in other regions or SES or racial/ethnic groups; nevertheless, it does shed light on BE-PA
relationships in an important population, and it may pave the way for future studies to continue
exploring this important relationship during pregnancy and postpartum.
4.5. Conclusions
Pregnancy and early postpartum are critical periods for women’s health. This study has
shown that daily exposure to parks and open space, and walkable environments may affect
women’s PA outcomes during pregnancy and early postpartum, which in turn would affect both
short- and long-term maternal and infant health outcomes. More importantly, this study has

120
demonstrated that the choice of exposure assessment method may greatly influence the ability to
detect meaningful relationship between BE exposures and PA outcomes in pregnant women.
Consequently, future studies examining the impact of BE on pregnant women’s PA should
consider their spatiotemporal movement patterns when measuring their daily exposures to BE
characteristics, in order to reduce the spatial mismatch between exposures and outcomes. Lastly,
this study found that the BE-PA relationships were stronger on weekends, during late pregnancy,
and for those in low SES neighborhoods, which provided further guidance in formulating
effective PA promotion strategies for increasing women’s PA across pregnancy and postpartum
and decreasing PA-related health risks and outcomes.

121
Chapter 5 Conclusions
Maternal obesity during and after pregnancy is a major risk factor for mothers’ short- and long-
term health outcomes such as diabetes, cancer, and cardiovascular diseases. Hispanic women of
childbearing age have disproportionally high obesity risks and low PA levels in compared to
non-Hispanic white population. In this dissertation research, through three dissertation studies, I
examined daily mobility patterns (Chapter 2), dynamic BE exposures (Chapter 3), and the
associations between BE exposures and PA outcomes (Chapter 4) in a group of predominantly
Hispanic low-income women of childbearing age. Results of these studies further elucidate the
influences of women’s daily mobility and BE exposures on PA behaviors in this understudied
health disparity population.
The three dissertation studies have important implications for future studies that intend to
examine the environmental contextual influences on women’s PA behaviors during and after
pregnancy, especially women of low SES and specific racial/ethnic groups. The first two
sections of this chapter discuss this work’s main contributions to the existing literature and some
directions for future research. The last section summarizes major takeaways from this work.
5.1. Contributions and Connections
To start, this research contributes to the existing literature through its appliance of
cutting-edge exposure assessment methods. Leveraging highly resolved geolocation and
accelerometry-assessed movement data from the MADRES real-time data collection and
sampling study, this research was able to derive dynamic BE exposures through the application
of advanced geospatial analytical methods such as KDE-based spatiotemporal clustering
detection. These exposure metrics weighted the exposure values based on the duration a woman
spent in each daily activity location and for each trip (time-weighted spatial averaging approach).

122
Consequently, it overcame the spatial misclassification bias in prior residential-based studies (as
demonstrated by the correlation analysis results in Chapter 3) and captured the “true contextual
unit” better than traditional activity space methods such as RB, which ignores temporal
information. To my best knowledge, this is the first study that measures women’s dynamic daily
exposures to BE characteristics using highly resolved smartphone location data, and the study
results demonstrate the merit of incorporating spatiotemporal movement patterns into the
environmental exposure assessment to reduce measurement error.
With the application of both residential- and activity space-based exposure assessment
methods, this dissertation study had a unique opportunity to investigate and mitigate important
analytical problems that exist in studies that examine the effects of area-based environmental
contextual variables on individual behaviors or outcomes, especially UCGoP. In Chapter 4, I
examined the robustness of associations between daily exposure to greenness, parks and open
space, and walkability and women’s day-level PA outcomes through applying BE exposure
measures derived using the residential (RNB), spatial averaging (RB) and time-weighted spatial
averaging (KDE) methods (with two different buffer sizes for each method). I found the
association between park exposure and women’s PA outcomes only existed in measures derived
using two activity space methods with 250 m buffers. This shows how the configuration choices
embedded in different exposure measurement methods can greatly influence the model strength
in detecting meaningful BE-PA relationships. This study, to my best of knowledge, is the first
study that evaluates the magnitude of UGCoP in environmental contextual influences on PA
outcomes. The three studies in this dissertation augment limited existing studies (Zhao, Kwan,
and Zhou 2018; Delmelle et al. 2022; Laatikainen, Hasanzadeh, and Kyttä 2018) in showcasing

123
the sensitivity of study results to the geospatial methods that are applied in emerging activity
space-based neighborhood health studies.
Moreover, to my best knowledge, this dissertation is one of a handful of that has applied
highly resolved smartphone geolocation and accelerometry-assessed movement data to examine
the associations between daily mobility patterns, BE exposures, and PA outcomes in pregnant
women, and the first one that focuses on low income, Hispanic women of childbearing age.
Through three dissertation studies, I found pregnant women in my sample overall made few
daily visits to parks and open space (chapter 2) and had little exposure to parks and open space in
their daily activity spaces (Chapter 3), and these findings did not differ by pregnancy and early
postpartum periods. The latter of the two studies therefore provides further evidence for the
existence of an “ethno-racial” disparity in parks and green space access found by previous
studies which either examined largely Hispanic communities in Los Angeles or Hispanic
pregnant women in other US cities (Byrne 2012; Borodulin, Evenson, and Herring 2009; Derose
et al. 2015).
Furthermore, my finding of positive effects of activity space parks and open space
exposure on women’s PA outcomes during and after pregnancy (Chapter 4) provides a silver
lining and suggests a path forward for improving PA promotion strategies for pregnant women of
low SES and specific ethnic/racial minority groups. Specifically, the negative impact of low park
access in these women’s residential neighborhoods on their PA may be offset by increasing
women’s awareness of and visits to parks and open space close to their frequently visited non-
home activity locations (e.g., workplaces, grocery stores). As a result, future PA interventions for
pregnant women’s PA should take their daily non-home parks and open space exposure into

124
consideration to improve its efficacy, particularly if these interventions were targeting women of
low SES or specific racial/ethnic minority groups.
Lastly, my study also found women’s trip behaviors, daily exposures to urban form
characteristics (e.g., walkability), and influences of urban form exposures on their PA outcomes
to greatly vary by pregnancy and postpartum periods, and levels of neighborhood safety and
cohesion. For example, women decreased their time spent at commercial and service locations
and reduced the total number of vehicular-based trips at 4-6 months postpartum (Chapter 2). In
the same chapter, lower neighborhood cohesion and safety score was associated with increased
vehicular trips across pregnancy and postpartum. In examining the day-level effects of BE
exposures on PA outcomes, I found activity space walkability was only positively associated
with women’s PA during the late pregnancy period and for women who reported their
neighborhoods as less safe (Chapter 4). Consequently, this research complements two prior
residential-based studies in demonstrating the pregnancy and early postpartum period may be a
critical modifier in the relationships of BE exposures to women’s PA outcomes (Porter et al.
2019; Richardsen et al. 2016). In addition, this research adds to the previous evidence by
showing the importance of providing access to walkable environments for women of low SES
and specific racial/ethnic minority groups in their daily activity spaces to increase their PA
outcomes (Perez, Ruiz, and Berrigan 2019). Moreover, this dissertation research was able to
examine day-level influences of BE exposure on pregnant women’s PA outcomes during and
after pregnancy by disentangling between-subject (person-level) and within-subject (day-level)
effects, which provides novel insights on the BE-PA association at a fine temporal granularity.

125
5.2. Limitations and Directions for Future Studies
Despite the aforementioned merits, this research is limited in terms of its data, methods,
analytical approaches, and generalizability. To start, the missingness in GPS and accelerometry
data applied in this dissertation research may be a concern. I made multiple efforts to mitigate
the potential impact of missingness on study outcomes. These efforts included imputing missing
GPS data using participant-reported sleep/wake time and EMA-reported physical context
(Chapter 2), exploring diurnal patterns of GPS data missingness (Chapter 2), performing
sensitivity analyses using different cut-offs for labeling valid GPS and accelerometry observation
days (Chapter 2, 3, and 4), and controlling variables (e.g., weekdays versus weekend days) that
predicted invalid days (Chapter 4). However, other unaccounted data missingness scenarios may
exist and possibly bias study results. For instance, the missingness of GPS data may be
correlated with spatial context (e.g., tall buildings, trees) that could obstruct receiver signals. As
a result, BE characteristics (e.g., proximity to public transit stops) in these spatial contexts were
not captured and therefore day-level exposure results may be under- or over-estimated. Future
studies are recommended to compare different advanced missing data imputation approaches,
evaluate their imputation rates and accuracies, and chose the one that achieves the highest
accuracies for their specific setting (Barnett and Onnela 2020).
Further, given GPS and accelerometry data were collected using separate devices, the
missingness from one or both data may greatly reduce the sample sizes and powers for the
statistical models to detect associations between BE exposure and PA outcomes. Increasingly,
studies have advocated an “all-in-one” approach such as applying Smartphones or Smartwatches
to simultaneously collect geolocation and movement data as well as other self-reported
psychosocial variables such as stress and affect (Straczkiewicz, James, and Onnela 2021; Gal et

126
al. 2018). This approach may be more promising in reducing the rates of data missingness than
studies applying multiple sensors, although at a potential cost of data accuracy (Piccinini,
Martinelli, and Carbonaro 2020).
Besides the data missingness issue, more research is needed to continue improving the
activity space-based exposure measurement method applied in this research. For example, the
influences of environmental contextual factors on PA behaviors may vary by locations (home
versus workplace), time of the day (daytime versus evening), and days of the week (weekdays
versus weekends) (Kwan 2018; Koohsari et al. 2016). As a result, an adaptive buffer size
approach which derives minute-level BE exposures based on buffer sizes corresponding to time
of the day, weekdays versus weekend, and trip modes may be utilized to better capture BE
exposures for women during and after pregnancy, given their drastic changes in time-activity and
mobility patterns as shown in Chapter 2. In addition, recognizing traditional BE data (e.g.,
NDVI) is limited in capturing the pedestrian’s interactions with green features on the streets, in
this research, additional greenness measures such as % green space and tree cover along
walkable roads were derived to better capture these interactions. However, both measures still
could not capture other crucial street design features such as crosswalks, lighting fixtures, and
sidewalk maintenance that were associated with walking behaviors (Yin et al. 2015). Emerging
studies have started to extract street-level BE features from Google Street View products via
deep learning techniques (Yin et al. 2015; Yin 2017). Future studies should explore these novel
approaches and strive to incorporate them when examining the influences of BE exposures on
pregnant women’s PA behaviors.
Besides its limitations in data and exposure assessment methods, this dissertation study
remains subject to the selective daily mobility bias. This bias refers to the fact that an

127
unmeasured factor may have causal effects on both exposures and outcomes (Chaix et al. 2012;
2013). For instance, a pregnant woman with a good knowledge of benefits of exercise may
choose to visit a park to exercise. In this case, the measured effect of park access on her exercise
outcomes will be spurious since they are both caused by her motivations to exercise. To mitigate
the selective daily mobility bias, future studies should apply more advanced methods such as
excluding locations where PA occurred from the exposure assessment or only measuring the
exposure at or around anchor points (e.g., key locations where one organizes life activities such
as home, work, and neighborhood grocery store), which can be extracted over longer periods
(e.g., 12-months) of geolocation monitoring (Chaix et al. 2013; Perchoux et al. 2015, Zenk et al.
2018). These methods should be explored by future studies aim to examine the association
between GPS-based BE exposure and women’s PA behaviors across pregnancy and postpartum.
Furthermore, the analyses in this research are still cross-sectional in nature, which limit
the opportunities to infer causal relationships. Numerous studies have associated BE
characteristics with PA outcomes, but the behavioral mechanisms from BE exposure to PA
outcomes remain largely unknown (Travert, Annerstedt, and Daivadanam 2019). In this realm,
future studies should consider applying context-sensitive (e.g., during or after a PA episode,
during a walking session in the park) EMA survey tools to collect psychosocial information (e.g.,
perceived safety, stress, motivation, mood) as events unfold so that the connection between
objectively-measured and perceived BE exposures and these psychosocial variables in
determining PA behaviors can be further examined (Dunton 2018; Huang et al. 2016).
Finally, when conducting this dissertation study, I leveraged the GPS and accelerometry
data collected by the MADRES real-time and personal sampling study. This study focused on
understanding the causes of excessive weight gain and retention such as environmental exposure

128
and social stressors among low-income Hispanic women to reduce the disproportionate burden
of disease they bear; as a result, it only enrolled Hispanic women to limit between-group
racial/ethnic differences in its design. Due to this unique study design, results of this dissertation
study may not be generalized to pregnant women in other regions or SES or racial/ethnic group.
However, given Hispanic and low-income women of childbearing age as a group forms a large
and understudied group that has been disproportionally exposed to various environmental
hazards, my explorations of associations between daily mobility, BE exposures, and PA
outcomes may help to guide future studies for this important health disparity population or
potentially other populations of similar sociodemographic characteristics.
5.3. Major Takeaways
In conclusion, this dissertation study used highly resolved smartphone location and
accelerometry-assessed activity data collected from 62 Hispanic women of childbearing age in
urban Los Angeles, CA, during pregnancy and the early postpartum period to investigate the
associations among women’s daily mobility patterns (Chapter 2), dynamic built-environment
(BE) exposures (Chapter 3), and PA outcomes (Chapter 4). The results of three empirical case
studies revealed an exceedingly low parks and open space exposure for this group at their daily
activity locations and along travel paths during and after pregnancy, which were associated with
women’s lower day-to-day PA outcomes. In addition, important modifiers (e.g., late pregnancy,
early postpartum, high pre-pregnancy BMI, low neighborhood safety) of women’s daily
mobility, BE exposures, and their associations with PA outcomes were identified. Lastly, the
second and third studies evaluated measurement error and bias resulting from applying
traditional residential-based measures were evaluated and their implications on uncovering the
relationships between BE exposures and PA outcomes were investigated. Future studies are

129
suggested to perform qualitative analyses of the BE features (e.g., parks) in which women’s PA
was performed, mitigate selective daily mobility bias, and apply real-time surveying techniques
such as EMA to elucidate psychosocial pathways from BE exposures to PA outcomes. Future PA
promotion interventions for pregnancy women are recommended to target at-risk pregnancy and
postpartum periods and sociodemographic groups to improve their efficacy, especially for those
of low socioeconomic status and specific racial/ethnic minority groups.

130
References
Abatzoglou, J. T. 2013. “Development of Gridded Surface Meteorological Data for Ecological
Applications and Modelling.” International Journal of Climatology 33, no. 1 (January):
121–31. https://doi.org/10.1002/joc.3413.
Algoblan, A., M. Alalfi, and M. Khan. 2014. “Mechanism Linking Diabetes Mellitus and
Obesity.” Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy, December
(December), 587. https://doi.org/10.2147/DMSO.S67400.
Almanza, E., M. Jerrett, G. Dunton, E. Seto, and M. A. Pentz. 2012. “A Study of Community
Design, Greenness, and Physical Activity in Children Using Satellite, GPS and
Accelerometer Data.” Health & Place 18, no. 1 (January): 46–54.
https://doi.org/10.1016/j.healthplace.2011.09.003.
Anabitarte, A., M. Subiza-Pérez, J. Ibarluzea, K. Azkona, G. García-Baquero, C. Miralles-
Guasch, J. Irazusta, K. W. Whitworth, G. Vich, and A. Lertxundi. 2020. “Testing the
Multiple Pathways of Residential Greenness to Pregnancy Outcomes Model in a Sample
of Pregnant Women in the Metropolitan Area of Donostia-San Sebastián.” International
Journal of Environmental Research and Public Health 17, no. 12 (June): 4520.
https://doi.org/10.3390/ijerph17124520.
Arredondo, E. M., D. Sotres-Alvarez, M. Stoutenberg, S. M. Davis, N. C. Crespo, M. R.
Carnethon, S. F. Castañeda, et al. 2016. “Physical Activity Levels in U.S.
Latino/Hispanic Adults.” American Journal of Preventive Medicine 50, no. 4 (April):
500–508. https://doi.org/10.1016/j.amepre.2015.08.029.
August, K. J., and D. H. Sorkin. 2011. “Racial/Ethnic Disparities in Exercise and Dietary
Behaviors of Middle-Aged and Older Adults.” Journal of General Internal Medicine 26,
no. 3 (March): 245–50. https://doi.org/10.1007/s11606-010-1514-7.
Balakrishnan, K., S. Sambandam, P. Ramaswamy, S. Ghosh, V. Venkatesan, G. Thangavel, K.
Mukhopadhyay, et al. 2015. “Establishing Integrated Rural-Urban Cohorts to Assess Air
Pollution-Related Health Effects in Pregnant Women, Children and Adults in Southern
India: An Overview of Objectives, Design and Methods in the Tamil Nadu Air Pollution
and Health Effects (TAPHE) Study.” BMJ Open 5, no. 6 (June): e008090–e008090.
https://doi.org/10.1136/bmjopen-2015-008090.
Banay, R., C. Bezold, P. James, J. Hart, and F. Laden. 2017. “Residential Greenness: Current
Perspectives on Its Impact on Maternal Health and Pregnancy Outcomes.” International
Journal of Women’s Health Volume 9, no. February (February): 133–44.
https://doi.org/10.2147/IJWH.S125358.
Barnett, I., and J.-P. Onnela. 2020. “Inferring Mobility Measures from GPS Traces with Missing
Data.” Biostatistics 21, no. 2 (April): e98–112.
https://doi.org/10.1093/biostatistics/kxy059.

131
Bastain, T. M., T. Chavez, R. Habre, M. S. Girguis, B. Grubbs, C. Toledo-Corral, M. Amadeus,
et al. 2019. “Study Design, Protocol and Profile of the Maternal And Developmental
Risks from Environmental and Social Stressors (MADRES) Pregnancy Cohort: A
Prospective Cohort Study in Predominantly Low-Income Hispanic Women in Urban Los
Angeles.” BMC Pregnancy and Childbirth 19, no. 1 (December): 189.
https://doi.org/10.1186/s12884-019-2330-7.
Bates, D., M. Mächler, B. Bolker, and S. Walker. 2015. “Fitting Linear Mixed-Effects Models
Using Lme4.” Journal of Statistical Software 67, no. 1.
https://doi.org/10.18637/jss.v067.i01.
Bell, M. L., and K. Belanger. 2012. “Review of Research on Residential Mobility during
Pregnancy: Consequences for Assessment of Prenatal Environmental Exposures.”
Journal of Exposure Science & Environmental Epidemiology 22, no. 5 (September): 429–
38. https://doi.org/10.1038/jes.2012.42.
Bell, R., P. W.G. Tennant, C. McParlin, M. S. Pearce, A. J. Adamson, J. Rankin, and S. C.
Robson. 2013. “Measuring Physical Activity in Pregnancy: A Comparison of
Accelerometry and Self-Completion Questionnaires in Overweight and Obese Women.”
European Journal of Obstetrics & Gynecology and Reproductive Biology 170, no. 1
(September): 90–95. https://doi.org/10.1016/j.ejogrb.2013.05.018.
Berntsen, S., K. R. Richardsen, K. Mørkrid, L. Sletner, K. I. Birkeland, and A. K. Jenum. 2014.
“Objectively Recorded Physical Activity in Early Pregnancy: A Multiethnic Population-
Based Study: Physical Activity in Multiethnic Pregnant Women.” Scandinavian Journal
of Medicine & Science in Sports 24, no. 3 (June): 594–601.
https://doi.org/10.1111/sms.12034.
Besser, L. M., and A. L. Dannenberg. 2005. “Walking to Public Transit: Steps to Help Meet
Physical Activity Recommendations.” American Journal of Preventive Medicine 29, no.
4: 273–80. https://doi.org/10.1016/j.amepre.2005.06.010.
Blanchard, O., S. Deguen, W. Kihal-Talantikite, R. François, and D. Zmirou-Navier. 2018.
“Does Residential Mobility during Pregnancy Induce Exposure Misclassification for Air
Pollution?” Environmental Health 17, no. 1 (December): 72.
https://doi.org/10.1186/s12940-018-0416-8.
Boll, L. M., R. Khamirchi, L. Alonso, E. Llurba, O. J. Pozo, M. Miri, and P. Dadvand. 2020.
“Prenatal Greenspace Exposure and Cord Blood Cortisol Levels: A Cross-Sectional
Study in a Middle-Income Country.” Environment International 144, no. November
(November): 106047. https://doi.org/10.1016/j.envint.2020.106047.
Booth, K. M., M. M. Pinkston, and W. S. Carlos Poston. 2005. “Obesity and the Built
Environment.” Journal of the American Dietetic Association 105, no. 5 SUPPL.
https://doi.org/10.1016/j.jada.2005.02.045.

132
Borodulin, K., K. R. Evenson, and A. H. Herring. 2009. “Physical Activity Patterns during
Pregnancy through Postpartum.” BMC Women’s Health 9, no. 1 (December): 32.
https://doi.org/10.1186/1472-6874-9-32.
Borodulin, K. M., K. R. Evenson, F. Wen, A. H. Herring, and A. M. Benson. 2008. “Physical
Activity Patterns during Pregnancy.” Medicine and Science in Sports and Exercise 40,
no. 11 (November): 1901–8. https://doi.org/10.1249/MSS.0b013e31817f1957.
Brawarsky, P., N.E. Stotland, R.A. Jackson, E. Fuentes-Afflick, G.J. Escobar, N. Rubashkin, and
J.S. Haas. 2005. “Pre-Pregnancy and Pregnancy-Related Factors and the Risk of
Excessive or Inadequate Gestational Weight Gain.” International Journal of Gynecology
& Obstetrics 91, no. 2 (November): 125–31. https://doi.org/10.1016/j.ijgo.2005.08.008.
Brooks, M., E., K. Kristensen, K. Benthem J.,van, A. Magnusson, C. W. Berg, A. Nielsen, H.
Skaug, M. Mächler, and B. M. Bolker. 2017. “GlmmTMB Balances Speed and
Flexibility Among Packages for Zero-Inflated Generalized Linear Mixed Modeling.” The
R Journal 9, no. 2: 378. https://doi.org/10.32614/RJ-2017-066.
Brunson, J. 2020. “Ggalluvial: Layered Grammar for Alluvial Plots.” Journal of Open Source
Software 5, no. 49 (May): 2017. https://doi.org/10.21105/joss.02017.
Byrne, J. 2012. “When Green Is White: The Cultural Politics of Race, Nature and Social
Exclusion in a Los Angeles Urban National Park.” Geoforum 43, no. 3 (May): 595–611.
https://doi.org/10.1016/j.geoforum.2011.10.002.
Byrne, N. M., A. M. Groves, H. D. McIntyre, L. K. Callaway, and for the BAMBINO group.
2011. “Changes in Resting and Walking Energy Expenditure and Walking Speed during
Pregnancy in Obese Women.” American Journal of Clinical Nutrition 94, no. 3
(September): 819–30. https://doi.org/10.3945/ajcn.110.009399.
Carlson, J. A., B.E. Saelens, J. Kerr, J. Schipperijn, T.L. Conway, L.D. Frank, J.E. Chapman, K.
Glanz, K. L. Cain, and J.F. Sallis. 2015. “Association between Neighborhood Walkability
and GPS-Measured Walking, Bicycling and Vehicle Time in Adolescents.” Health and
Place 32: 1–7. https://doi.org/10.1016/j.healthplace.2014.12.008.
Chaix, B. 2018. “Mobile Sensing in Environmental Health and Neighborhood Research.” Annual
Review of Public Health 39, no. 1 (April): 367–84. https://doi.org/10.1146/annurev-
publhealth-040617-013731.
Chaix, B., Y. Kestens, C. Perchoux, N. Karusisi, J. Merlo, and K. Labadi. 2012. “An Interactive
Mapping Tool to Assess Individual Mobility Patterns in Neighborhood Studies.”
American Journal of Preventive Medicine 43, no. 4 (October): 440–50.
https://doi.org/10.1016/j.amepre.2012.06.026.
Chaix, B., J. Méline, S. Duncan, C. Merrien, N. Karusisi, C. Perchoux, A. Lewin, K. Labadi, and
Y. Kestens. 2013. “GPS Tracking in Neighborhood and Health Studies: A Step Forward
for Environmental Exposure Assessment, a Step Backward for Causal Inference?” Health
& Place 21, no. May (May): 46–51. https://doi.org/10.1016/j.healthplace.2013.01.003.

133
Chasan-Taber, L., M. D. Schmidt, P. Pekow, B. Sternfeld, C. G. Solomon, and G. Markenson.
2008. “Predictors of Excessive and Inadequate Gestational Weight Gain in Hispanic
Women.” Obesity 16, no. 7 (July): 1657–66. https://doi.org/10.1038/oby.2008.256.
Chen, L, E. M. Bell, A. R. Caton, C. M. Druschel, and S. Lin. 2010. “Residential Mobility
during Pregnancy and the Potential for Ambient Air Pollution Exposure
Misclassification.” Environmental Research 110, no. 2 (February): 162–68.
https://doi.org/10.1016/j.envres.2009.11.001.
Cronan, M. K., K. J. Shinew, I. Schneider, S. A. W. Stanis, and D. Chavez. 2008. “Physical
Activity Patterns and Preferences Among Latinos in Different Types of Public Parks.”
Journal of Physical Activity and Health 5, no. 6 (November): 894–908.
https://doi.org/10.1123/jpah.5.6.894.
Curran, P. J., and D. J. Bauer. 2011. “The Disaggregation of Within-Person and Between-Person
Effects in Longitudinal Models of Change.” Annual Review of Psychology 62, no. 1
(January): 583–619. https://doi.org/10.1146/annurev.psych.093008.100356.
Currie, L. M., C. G. Woolcott, D. B. Fell, B. A. Armson, and L. Dodds. 2014. “The Association
Between Physical Activity and Maternal and Neonatal Outcomes: A Prospective Cohort.”
Maternal and Child Health Journal 18, no. 8 (October): 1823–30.
https://doi.org/10.1007/s10995-013-1426-3.
Dadvand, P., A. de Nazelle, F. Figueras, X. Basagaña, J. Su, E. Amoly, M. Jerrett, M. Vrijheid,
J. Sunyer, and M.J. Nieuwenhuijsen. 2012. “Green Space, Health Inequality and
Pregnancy.” Environment International 40, no. 1: 110–15.
https://doi.org/10.1016/j.envint.2011.07.004.
Dadvand, P., A. de Nazelle, M. Triguero-Mas, A. Schembari, M. Cirach, E. Amoly, F. Figueras,
X. Basagaña, B. Ostro, and M. Nieuwenhuijsen. 2012. “Surrounding Greenness and
Exposure to Air Pollution During Pregnancy: An Analysis of Personal Monitoring Data.”
Environmental Health Perspectives 120, no. 9 (September): 1286–90.
https://doi.org/10.1289/ehp.1104609.
Daley, A. J., C. Psychol, C. MacArthur, and H. Winter. 2007. “The Role of Exercise in Treating
Postpartum Depression: A Review of the Literature.” Journal of Midwifery & Women’s
Health 52, no. 1 (January): 56–62. https://doi.org/10.1016/j.jmwh.2006.08.017.
Delmelle, E. M., M. R. Desjardins, P. Jung, C. Owusu, Y. Lan, A. Hohl, and C. Dony. 2022.
“Uncertainty in Geospatial Health: Challenges and Opportunities Ahead.” Annals of
Epidemiology 65, no. January (January): 15–30.
https://doi.org/10.1016/j.annepidem.2021.10.002.
Diez R. A. V., and C. Mair. 2010. “Neighborhoods and Health: Neighborhoods and Health.”
Annals of the New York Academy of Sciences 1186, no. 1 (February): 125–45.
https://doi.org/10.1111/j.1749-6632.2009.05333.x.

134
Dolash, K., M. He, Z. Yin, and E. T. Sosa. 2015. “Factors That Influence Park Use and Physical
Activity in Predominantly Hispanic and Low-Income Neighborhoods.” Journal of
Physical Activity and Health 12, no. 4 (April): 462–69.
https://doi.org/10.1123/jpah.2013-0226.
Dunton, G. F. 2017. “Ecological Momentary Assessment in Physical Activity Research.”
Exercise and Sport Sciences Reviews 45, no. 1 (January): 48–54.
https://doi.org/10.1249/JES.0000000000000092.
Dunton, G. F. 2018. “Sustaining Health-Protective Behaviors Such as Physical Activity and
Healthy Eating.” JAMA 320, no. 7 (August): 639.
https://doi.org/10.1001/jama.2018.6621.
Evenson, K. R., R. Barakat, W. J. Brown, P. Dargent-Molina, M. Haruna, E. M. Mikkelsen, M.
F. Mottola, K. M. Owe, E. K. Rousham, and S. Yeo. 2014. “Guidelines for Physical
Activity During Pregnancy: Comparisons From Around the World.” American Journal of
Lifestyle Medicine 8, no. 2 (March): 102–21. https://doi.org/10.1177/1559827613498204.
Evenson, K. R., D. A. Savitz, and S. L. Huston. 2004. “Leisure-Time Physical Activity among
Pregnant Women in the US.” Paediatric and Perinatal Epidemiology 18, no. 6
(November): 400–407. https://doi.org/10.1111/j.1365-3016.2004.00595.x.
Evenson, K. R., M.-K. Moos, K. Carrier, and A. M. Siega-Riz. 2009. “Perceived Barriers to
Physical Activity Among Pregnant Women.” Maternal and Child Health Journal 13, no.
3 (May): 364–75. https://doi.org/10.1007/s10995-008-0359-8.
Evenson, K. R., and F. Wen. 2010. “National Trends in Self-Reported Physical Activity and
Sedentary Behaviors among Pregnant Women: NHANES 1999–2006.” Preventive
Medicine 50, no. 3 (March): 123–28. https://doi.org/10.1016/j.ypmed.2009.12.015.
Fan, J., Y. Song, Y. Chen, R. Hui, and W. Zhang. 2013. “Combined Effect of Obesity and
Cardio-Metabolic Abnormality on the Risk of Cardiovascular Disease: A Meta-Analysis
of Prospective Cohort Studies.” International Journal of Cardiology 168, no. 5
(October): 4761–68. https://doi.org/10.1016/j.ijcard.2013.07.230.
Fell, D. B., K. S. Joseph, B. A. Armson, and L. Dodds. 2009. “The Impact of Pregnancy on
Physical Activity Level.” Maternal and Child Health Journal 13, no. 5 (September):
597–603. https://doi.org/10.1007/s10995-008-0404-7.
Fields, R., A. T. Kaczynski, M. Bopp, and E. Fallon. 2013. “Built Environment Associations
With Health Behaviors Among Hispanics.” Journal of Physical Activity and Health 10,
no. 3 (March): 335–42. https://doi.org/10.1123/jpah.10.3.335.
Frank, L. D., E. H. Fox, J. M. Ulmer, J. E. Chapman, S. E. Kershaw, J. F. Sallis, T. L. Conway,
et al. 2017. “International Comparison of Observation-Specific Spatial Buffers:
Maximizing the Ability to Estimate Physical Activity.” International Journal of Health
Geographics 16, no. 1 (December): 4. https://doi.org/10.1186/s12942-017-0077-9.

135
Fraser, A., K. Tilling, C. Macdonald-Wallis, R. Hughes, N. Sattar, S. M Nelson, and D. A
Lawlor. 2011. “Associations of Gestational Weight Gain with Maternal Body Mass
Index, Waist Circumference, and Blood Pressure Measured 16 y after Pregnancy: The
Avon Longitudinal Study of Parents and Children (ALSPAC).” The American Journal of
Clinical Nutrition 93, no. 6 (June): 1285–92. https://doi.org/10.3945/ajcn.110.008326.
Gal, R., A. M. May, E. J. van Overmeeren, M. Simons, and E. M. Monninkhof. 2018. “The
Effect of Physical Activity Interventions Comprising Wearables and Smartphone
Applications on Physical Activity: A Systematic Review and Meta-Analysis.” Sports
Medicine - Open 4, no. 1 (December): 42. https://doi.org/10.1186/s40798-018-0157-9.
Gilleard, W. L. 2013. “Trunk Motion and Gait Characteristics of Pregnant Women When
Walking: Report of a Longitudinal Study with a Control Group.” BMC Pregnancy and
Childbirth 13, no. 1 (December): 71. https://doi.org/10.1186/1471-2393-13-71.
Giuliano, G. 2005. “Low Income, Public Transit, and Mobility.” Transportation Research
Record: Journal of the Transportation Research Board 1927, no. 1 (January): 63–70.
https://doi.org/10.1177/0361198105192700108.
Ha, S., C. Nobles, J. Kanner, S. Sherman, S.-H. Cho, N. Perkins, A. Williams, et al. 2020. “Air
Pollution Exposure Monitoring among Pregnant Women with and without Asthma.”
International Journal of Environmental Research and Public Health 17, no. 13 (July):
4888. https://doi.org/10.3390/ijerph17134888.
Hales, C. M., C. D. Fryar, M. D. Carroll, D. S. Freedman, and C. L. Ogden. 2018. “Trends in
Obesity and Severe Obesity Prevalence in US Youth and Adults by Sex and Age, 2007-
2008 to 2015-2016.” JAMA 319, no. 16 (April): 1723.
https://doi.org/10.1001/jama.2018.3060.
Hannam, K., R. McNamee, F. De Vocht, P. Baker, C. Sibley, and R. Agius. 2013. “A
Comparison of Population Air Pollution Exposure Estimation Techniques with Personal
Exposure Estimates in a Pregnant Cohort.” Environmental Science: Processes & Impacts
15, no. 8: 1562–72. https://doi.org/10.1039/c3em00112a.
Harrison, C. L., R. G. Thompson, H. J. Teede, and C. B. Lombard. 2011. “Measuring Physical
Activity during Pregnancy.” International Journal of Behavioral Nutrition and Physical
Activity 8, no. 1 (December): 19. https://doi.org/10.1186/1479-5868-8-19.
Hausenblas, H,, P. Giacobbi, B. Cook, R. Rhodes, and A. Cruz. 2011. “Prospective Examination
of Pregnant and Nonpregnant Women’s Physical Activity Beliefs and Behaviours.”
Journal of Reproductive and Infant Psychology 29, no. 4 (September): 308–19.
https://doi.org/10.1080/02646838.2011.629993.
Headen, I. E., E. M. Davis, M. S. Mujahid, and B. Abrams. 2012. “Racial-Ethnic Differences in
Pregnancy-Related Weight.” Advances in Nutrition 3, no. 1 (January): 83–94.
https://doi.org/10.3945/an.111.000984.

136
Hodgson, S., P. W.W. Lurz, M. D.F. Shirley, M. Bythell, and J. Rankin. 2015. “Exposure
Misclassification Due to Residential Mobility during Pregnancy.” International Journal
of Hygiene and Environmental Health 218, no. 4 (June): 414–21.
https://doi.org/10.1016/j.ijheh.2015.03.007.
Huang, T. K., K. E. Wyka, E. B. Ferris, J. Gardner, K. R. Evenson, D. Tripathi, G. M. Soto, et al.
2016. “The Physical Activity and Redesigned Community Spaces (PARCS) Study:
Protocol of a Natural Experiment to Investigate the Impact of Citywide Park Redesign
and Renovation.” BMC Public Health 16, no. 1 (December): 1160.
https://doi.org/10.1186/s12889-016-3822-2.
Hughes, J. P., M. A. McDowell, and D. J. Brody. 2008. “Leisure-Time Physical Activity Among
US Adults 60 or More Years of Age: Results From NHANES 1999–2004.” Journal of
Physical Activity and Health 5, no. 3 (May): 347–58.
https://doi.org/10.1123/jpah.5.3.347.
James, P., D. Berrigan, J. E. Hart, J. A. Hipp, C. M. Hoehner, J. Kerr, J. M. Major, M. Oka, and
F. Laden. 2014. “Effects of Buffer Size and Shape on Associations between the Built
Environment and Energy Balance.” Health & Place 27, no. May (May): 162–70.
https://doi.org/10.1016/j.healthplace.2014.02.003.
James, P., M. Jankowska, C. Marx, J. E. Hart, D. Berrigan, J. Kerr, P. M. Hurvitz, J. A. Hipp,
and F. Laden. 2016. ‘Spatial Energetics’: Integrating Data From GPS, Accelerometry,
and GIS to Address Obesity and Inactivity.” American Journal of Preventive Medicine
51, no. 5 (November): 792–800. https://doi.org/10.1016/j.amepre.2016.06.006.
Jankowska, M. M., L. Natarajan, S. Godbole, K. Meseck, D. D. Sears, R. E. Patterson, and J.
Kerr. 2017. “Kernel Density Estimation as a Measure of Environmental Exposure
Related to Insulin Resistance in Breast Cancer Survivors.” Cancer Epidemiology
Biomarkers & Prevention 26, no. 7 (July): 1078–84. https://doi.org/10.1158/1055-
9965.EPI-16-0927.
Jankowska, M. M., J. Schipperijn, and J. Kerr. 2015. “A Framework for Using GPS Data in
Physical Activity and Sedentary Behavior Studies.” Exercise and Sport Sciences Reviews
43, no. 1 (January): 48–56. https://doi.org/10.1249/JES.0000000000000035.
Jankowska, M. M., J.-A. Yang, N. Luo, C. Spoon, and T. Benmarhnia. 2021. “Accounting for
Space, Time, and Behavior Using GPS Derived Dynamic Measures of Environmental
Exposure.” Health & Place, November (November), 102706.
https://doi.org/10.1016/j.healthplace.2021.102706.
Jenum, A. K., K. R. Richardsen, S. Berntsen, and K. Mørkrid. 2013. “Gestational Diabetes,
Insulin Resistance and Physical Activity in Pregnancy in a Multi-Ethnic Population – a
Public Health Perspective.” Norsk Epidemiologi 23, no. 1 (June).
https://doi.org/10.5324/nje.v23i1.1602.
Jiang, B., D. Li, L. Larsen, and W. C. Sullivan. 2016. “A Dose-Response Curve Describing the
Relationship Between Urban Tree Cover Density and Self-Reported Stress Recovery.”

137
Environment and Behavior 48, no. 4 (May): 607–29.
https://doi.org/10.1177/0013916514552321.
Kaczynski, A. T., and A. J. Mowen. 2011. “Does Self-Selection Influence the Relationship
between Park Availability and Physical Activity?” Preventive Medicine 52, no. 1
(January): 23–25. https://doi.org/10.1016/j.ypmed.2010.10.003.
Kaczynski, A. T., L. R. Potwarka, and B. E. Saelens. 2008. “Association of Park Size, Distance,
and Features With Physical Activity in Neighborhood Parks.” American Journal of
Public Health 98, no. 8 (August): 1451–56. https://doi.org/10.2105/AJPH.2007.129064.
Kakinami, L., R. Wissa, R. Khan, G. Paradis, T. A. Barnett, and L. Gauvin. 2018. “The
Association between Income and Leisure-Time Physical Activity Is Moderated by
Utilitarian Lifestyles: A Nationally Representative US Population (NHANES 1999–
2014).” Preventive Medicine 113, no. August (August): 147–52.
https://doi.org/10.1016/j.ypmed.2018.05.013.
Kershaw, K.N., D.J. Marsh, E.G. Crenshaw, R.B. McNeil, V.L. Pemberton, S.A. Cordon, D.M.
Haas, et al. 2021. “Associations of the Neighborhood Built Environment with Physical
Activity across Pregnancy.” Journal of Physical Activity and Health 18, no. 5: 541–47.
https://doi.org/10.1123/JPAH.2020-0510.
Kim, J., and M.-P. Kwan. 2021. “How Neighborhood Effect Averaging Might Affect
Assessment of Individual Exposures to Air Pollution: A Study of Ozone Exposures in
Los Angeles.” Annals of the American Association of Geographers 111, no. 1 (January):
121–40. https://doi.org/10.1080/24694452.2020.1756208.
Kind, A. J. H., and W. R. Buckingham. 2018. “Making Neighborhood-Disadvantage Metrics
Accessible — The Neighborhood Atlas.” New England Journal of Medicine 378, no. 26
(June): 2456–58. https://doi.org/10.1056/NEJMp1802313.
Kirby, R. S., E. Delmelle, and J. M. Eberth. 2017. “Advances in Spatial Epidemiology and
Geographic Information Systems.” Annals of Epidemiology 27, no. 1 (January): 1–9.
https://doi.org/10.1016/j.annepidem.2016.12.001.
Koohsari, M. J., N. Owen, E. Cerin, B. Giles-Corti, and T. Sugiyama. 2016. “Walkability and
Walking for Transport: Characterizing the Built Environment Using Space Syntax.”
International Journal of Behavioral Nutrition and Physical Activity.
https://doi.org/10.1186/s12966-016-0448-9.
Krans, E. E., and J. C. Chang. 2011. “A Will Without a Way: Barriers and Facilitators to
Exercise During Pregnancy of Low-Income, African American Women.” Women &
Health 51, no. 8 (November): 777–94. https://doi.org/10.1080/03630242.2011.633598.
Kwan, M.-P. 2012. “The Uncertain Geographic Context Problem.” Annals of the Association of
American Geographers 102, no. 5 (September): 958–68.
https://doi.org/10.1080/00045608.2012.687349.

138
Kwan, M.-P. 2018. “The Limits of the Neighborhood Effect: Contextual Uncertainties in
Geographic, Environmental Health, and Social Science Research.” Annals of the
American Association of Geographers 108, no. 6 (November): 1482–90.
https://doi.org/10.1080/24694452.2018.1453777.
Laatikainen, T. E., K. Hasanzadeh, and M. Kyttä. 2018. “Capturing Exposure in Environmental
Health Research: Challenges and Opportunities of Different Activity Space Models.”
International Journal of Health Geographics 17, no. 1 (December): 29.
https://doi.org/10.1186/s12942-018-0149-5.
Laraia, B., L. Messer, K. Evenson, and J. S. Kaufman. 2007. “Neighborhood Factors Associated
with Physical Activity and Adequacy of Weight Gain During Pregnancy.” Journal of
Urban Health 84, no. 6 (November): 793–806. https://doi.org/10.1007/s11524-007-9217-
z.
Larsen, B. A, D. Pekmezi, B. Marquez, T. J. Benitez, and B. H. Marcus. 2013. “Physical Activity
in Latinas: Social and Environmental Influences.” Women’s Health 9, no. 2 (March):
201–10. https://doi.org/10.2217/WHE.13.9.
Lee, K., and M.-P. Kwan. 2019. “The Effects of GPS-Based Buffer Size on the Association
between Travel Modes and Environmental Contexts.” ISPRS International Journal of
Geo-Information 8, no. 11 (November): 514. https://doi.org/10.3390/ijgi8110514.
Leggett, S. S., J. M. Wesley, A. Myla, P. McCoy, and N. Betson. 2018. “Availability and Quality
of Physical Activity Resources in Neighborhood Parks for Pregnant Women and Women
of Childbearing Age.” Preventing Chronic Disease 15, no. June (June): 170411.
https://doi.org/10.5888/pcd15.170411.
Liao, J., X. Chen, S. Xu, Y. Li, B. Zhang, Z. Cao, Y. Zhang, S. Liang, K. Hu, and W. Xia. 2019.
“Effect of Residential Exposure to Green Space on Maternal Blood Glucose Levels,
Impaired Glucose Tolerance, and Gestational Diabetes Mellitus.” Environmental
Research 176, no. September (September): 108526.
https://doi.org/10.1016/j.envres.2019.108526.
Loh, V. H. Y., J. Veitch, J. Salmon, E. Cerin, L. Thornton, S. Mavoa, K. Villanueva, and A.
Timperio. 2019. “Built Environment and Physical Activity among Adolescents: The
Moderating Effects of Neighborhood Safety and Social Support.” International Journal
of Behavioral Nutrition and Physical Activity 16, no. 1 (December): 132.
https://doi.org/10.1186/s12966-019-0898-y.
Lovasi, G. S., M. A. Hutson, M. Guerra, and K. M. Neckerman. 2009. “Built Environments and
Obesity in Disadvantaged Populations.” Epidemiologic Reviews 31, no. 1 (November):
7–20. https://doi.org/10.1093/epirev/mxp005.
MacLeod, K. E., L. Shi, D. Zhang, L. Chen, and S. M. Chao. 2018. “Association between
Vehicle Time during Pregnancy and Mental Health among Women of Different Income
Groups.” Journal of Transport & Health 8, no. March (March): 106–11.
https://doi.org/10.1016/j.jth.2018.01.003.

139
Mama, S. K., S. A. McCurdy, A. E. Evans, D. I. Thompson, P. M. Diamond, and R. E. Lee.
2015. “Using Community Insight to Understand Physical Activity Adoption in
Overweight and Obese African American and Hispanic Women: A Qualitative Study.”
Health Education & Behavior 42, no. 3 (June): 321–28.
https://doi.org/10.1177/1090198114557128.
Marshall, J. D., M. Brauer, and L. D. Frank. 2009. “Healthy Neighborhoods: Walkability and Air
Pollution.” Environmental Health Perspectives 117, no. 11 (November): 1752–59.
https://doi.org/10.1289/ehp.0900595.
Matthews, S. A., and T.-C. Yang. 2013. “Spatial Polygamy and Contextual Exposures
(SPACEs): Promoting Activity Space Approaches in Research on Place And Health.”
American Behavioral Scientist 57, no. 8 (August): 1057–81.
https://doi.org/10.1177/0002764213487345.
McEachan, R. R. C., S. L. Prady, G. Smith, L. Fairley, B. Cabieses, C. Gidlow, J. Wright, P.
Dadvand, D. van Gent, and M. J. Nieuwenhuijsen. 2016. “The Association between
Green Space and Depressive Symptoms in Pregnant Women: Moderating Roles of
Socioeconomic Status and Physical Activity.” Journal of Epidemiology and Community
Health 70, no. 3 (March): 253–59. https://doi.org/10.1136/jech-2015-205954.
McNeill, L. H., M. W. Kreuter, and S.V. Subramanian. 2006. “Social Environment and Physical
Activity: A Review of Concepts and Evidence.” Social Science & Medicine 63, no. 4
(August): 1011–22. https://doi.org/10.1016/j.socscimed.2006.03.012.
Melzer, K., Y. Schutz, N. Soehnchen, V. Othenin-Girard, B. Martinez de Tejada, O. Irion, M.
Boulvain, and B. Kayser. 2010. “Effects of Recommended Levels of Physical Activity on
Pregnancy Outcomes.” American Journal of Obstetrics and Gynecology 202, no. 3
(March): 266.e1-266.e6. https://doi.org/10.1016/j.ajog.2009.10.876.
Moskowitz, D. S., and S. N. Young. 2006. “Ecological Momentary Assessment: What It Is and
Why It Is a Method of the Future in Clinical Psychopharmacology.” Journal of
Psychiatry & Neuroscience: JPN 31, no. 1 (January): 13–20.
Nethery, E, S E Leckie, K Teschke, and M Brauer. 2008. “From Measures to Models: An
Evaluation of Air Pollution Exposure Assessment for Epidemiological Studies of
Pregnant Women.” Occupational and Environmental Medicine 65, no. 9 (September):
579–86. https://doi.org/10.1136/oem.2007.035337.
Nethery, E., M. Brauer, and P. Janssen. 2009. “Time–Activity Patterns of Pregnant Women and
Changes during the Course of Pregnancy.” Journal of Exposure Science &
Environmental Epidemiology 19, no. 3 (March): 317–24.
https://doi.org/10.1038/jes.2008.24.
Nichani, V., K. Dirks, B. Burns, A. Bird, S. Morton, and C. Grant. 2016. “Green Space and
Physical Activity in Pregnant Women: Evidence from the Growing up in New Zealand
Study.” Journal of Physical Activity and Health 13, no. 12: 1341–50.
https://doi.org/10.1123/jpah.2016-0013.

140
O’Brien, O. A., K. L. Lindsay, M. McCarthy, A. F. McGloin, M. Kennelly, H. A. Scully, and F.
M. McAuliffe. 2017. “Influences on the Food Choices and Physical Activity Behaviours
of Overweight and Obese Pregnant Women: A Qualitative Study.” Midwifery 47, no.
April (April): 28–35. https://doi.org/10.1016/j.midw.2017.02.003.
O’Connor, S. G., R. Habre, T. M. Bastain, C. M. Toledo-Corral, F. D. Gilliland, S. P. Eckel, J.
Cabison, et al. 2019. “Within-Subject Effects of Environmental and Social Stressors on
Pre- and Post-Partum Obesity-Related Biobehavioral Responses in Low-Income Hispanic
Women: Protocol of an Intensive Longitudinal Study.” BMC Public Health 19, no. 1
(December): 253. https://doi.org/10.1186/s12889-019-6583-x.
Ogden, C. L., M. D. Carroll, H. G. Lawman, C. D. Fryar, D. Kruszon-Moran, B. K. Kit, and K.
M. Flegal. 2016. “Trends in Obesity Prevalence Among Children and Adolescents in the
United States, 1988-1994 Through 2013-2014.” JAMA 315, no. 21 (June): 2292.
https://doi.org/10.1001/jama.2016.6361.
Ouidir, M., L. Giorgis-Allemand, S. Lyon-Caen, X. Morelli, C. Cracowski, S. Pontet, I. Pin, J/
Lepeule, V. Siroux, and R. Slama. 2015. “Estimation of Exposure to Atmospheric
Pollutants during Pregnancy Integrating Space–Time Activity and Indoor Air Levels:
Does It Make a Difference?” Environment International 84, no. November (November):
161–73. https://doi.org/10.1016/j.envint.2015.07.021.
Pennington, A. D., M. J. Strickland, M. Klein, X. Zhai, A. G. Russell, C. Hansen, and L. A.
Darrow. 2017. “Measurement Error in Mobile Source Air Pollution Exposure Estimates
Due to Residential Mobility during Pregnancy.” Journal of Exposure Science &
Environmental Epidemiology 27, no. 5 (September): 513–20.
https://doi.org/10.1038/jes.2016.66.
Perchoux, C., B. Chaix, and Y. Kestens. 2019. “Activity Spaces in Place and Health Research:
Novel Exposure Measures, Data Collection Tools, and Designs.” Health & Place 58, no.
July (July): 102130. https://doi.org/10.1016/j.healthplace.2019.05.008.
Perchoux, C., Y. Kestens, R. Brondeel, and B. Chaix. 2015. “Accounting for the Daily Locations
Visited in the Study of the Built Environment Correlates of Recreational Walking (the
RECORD Cohort Study).” Preventive Medicine 81, no. December (December): 142–49.
https://doi.org/10.1016/j.ypmed.2015.08.010.
Perez, L. G., J. Carlson, D. Slymen, K. Patrick, J. Kerr, S. Godbole, J. Elder, G. Aamodt, and E.
Arredondo. 2016. “Does the Social Environment Moderate Associations of the Built
Environment with Latinas’ Objectively-Measured Neighborhood Outdoor Physical
Activity?” Preventive Medicine Reports 4, no. December (December): 551–57.
https://doi.org/10.1016/j.pmedr.2016.10.006.
Perez, L. G., J. M. Ruiz, and D. Berrigan. 2019. “Neighborhood Environment Perceptions among
Latinos in the U.S.” International Journal of Environmental Research and Public Health
16, no. 17 (August): 3062. https://doi.org/10.3390/ijerph16173062.

141
Piccinini, F., G. Martinelli, and A. Carbonaro. 2020. “Accuracy of Mobile Applications versus
Wearable Devices in Long-Term Step Measurements.” Sensors 20, no. 21 (November):
6293. https://doi.org/10.3390/s20216293.
Pickard, B. R., J. Daniel, M. Mehaffey, L. E. Jackson, and A. Neale. 2015. “EnviroAtlas: A New
Geospatial Tool to Foster Ecosystem Services Science and Resource Management.”
Ecosystem Services 14, no. August (August): 45–55.
https://doi.org/10.1016/j.ecoser.2015.04.005.
Piercy, K. L., R. P. Troiano, R. M. Ballard, S. A. Carlson, J, E. Fulton, D. A. Galuska, S. M.
George, and R. D. Olson. 2018. “The Physical Activity Guidelines for Americans.”
JAMA 320, no. 19 (November): 2020. https://doi.org/10.1001/jama.2018.14854.
Porter, A. K., D. A. Rodríguez, B. G. Frizzelle, and K. R. Evenson. 2019. “The Association
between Neighborhood Environments and Physical Activity from Pregnancy to
Postpartum: A Prospective Cohort Study.” Journal of Urban Health 96, no. 5 (October):
703–19. https://doi.org/10.1007/s11524-019-00376-y.
Pretty, J., J. Peacock, M. Sellens, and M. Griffin. 2005. “The Mental and Physical Health
Outcomes of Green Exercise.” International Journal of Environmental Health Research
15, no. 5 (October): 319–37. https://doi.org/10.1080/09603120500155963.
Rainham, D., I. McDowell, D. Krewski, and M. Sawada. 2010. “Conceptualizing the
Healthscape: Contributions of Time Geography, Location Technologies and Spatial
Ecology to Place and Health Research.” Social Science & Medicine 70, no. 5 (March):
668–76. https://doi.org/10.1016/j.socscimed.2009.10.035.
Derose, K.P., B. Han, S. Williamson, and D. A. Cohen. 2015. “Racial-Ethnic Variation in Park
Use and Physical Activity in the City of Los Angeles.” Journal of Urban Health 92, no. 6
(December): 1011–23. https://doi.org/10.1007/s11524-015-9994-8.
Rasmussen, K. M, P. M. Catalano, and A. L. Yaktine. 2009. “New Guidelines for Weight Gain
during Pregnancy: What Obstetrician/Gynecologists Should Know.” Current Opinion in
Obstetrics & Gynecology 21, no. 6 (December): 521–26.
https://doi.org/10.1097/GCO.0b013e328332d24e.
Redmond, M. L., F. Dong, and L. M. Frazier. 2015. “Does the Extended Parallel Process Model
Fear Appeal Theory Explain Fears and Barriers to Prenatal Physical Activity?” Women’s
Health Issues 25, no. 2 (March): 149–54. https://doi.org/10.1016/j.whi.2014.11.009.
Renault, K., K. Nørgaard, N. J. Secher, K. R. Andreasen, B. Baldur-Felskov, and L. Nilas. 2012.
“Physical Activity during Pregnancy in Normal-Weight and Obese Women: Compliance
Using Pedometer Assessment.” Journal of Obstetrics and Gynaecology 32, no. 5 (July):
430–33. https://doi.org/10.3109/01443615.2012.668580.
Richardsen, K. R., I. Mdala, S. Berntsen, Y. Ommundsen, E. Wilhelm Martinsen, L. Sletner, and
A. K. Jenum. 2016. “Objectively Recorded Physical Activity in Pregnancy and
Postpartum in a Multi-Ethnic Cohort: Association with Access to Recreational Areas in

142
the Neighbourhood.” International Journal of Behavioral Nutrition and Physical Activity
13, no. 1 (December): 78. https://doi.org/10.1186/s12966-016-0401-y.
Robertson, C., and R. Feick. 2018. “Inference and Analysis across Spatial Supports in the Big
Data Era: Uncertain Point Observations and Geographic Contexts.” Transactions in GIS
22, no. 2 (April): 455–76. https://doi.org/10.1111/tgis.12321.
Saelens, B. E., A. V. Moudon, B. Kang, P. M. Hurvitz, and C. Zhou. 2014. “Relation Between
Higher Physical Activity and Public Transit Use.” American Journal of Public Health
104, no. 5 (May): 854–59. https://doi.org/10.2105/AJPH.2013.301696.
Sallis, J. F., R. B. Cervero, W. Ascher, K. A. Henderson, M. K. Kraft, and J. Kerr. 2006. “An
Ecological Approach to Creating Active Living Communities.” Annual Review of Public
Health 27, no. 1 (April): 297–322.
https://doi.org/10.1146/annurev.publhealth.27.021405.102100.
Sallis, J. F., M. F. Floyd, D. A. Rodríguez, B. E. Saelens, D. A. Rodriguez, and B. E. Saelens.
2012. “The Role of Built Environments in Physical Activity, Obesity, and CVD.”
Circulation 125, no. 5: 729–37.
https://doi.org/10.1161/CIRCULATIONAHA.110.969022.
Sallis, J. F., B. E. Saelens, L. D. Frank, T. L. Conway, D. J. Slymen, K. L. Cain, J. E. Chapman,
and J. Kerr. 2009. “Neighborhood Built Environment and Income: Examining Multiple
Health Outcomes.” Social Science & Medicine 68, no. 7 (April): 1285–93.
https://doi.org/10.1016/j.socscimed.2009.01.017.
Sallis, J. F., D. J. Slymen, T. L. Conway, L. D. Frank, B. E. Saelens, K. Cain, and J. E.
Chapman. 2011. “Income Disparities in Perceived Neighborhood Built and Social
Environment Attributes.” Health & Place 17, no. 6 (November): 1274–83.
https://doi.org/10.1016/j.healthplace.2011.02.006.
Sampson, R. J., S. W. Raudenbush, and F. Earls. 1997. “Neighborhoods and Violent Crime: A
Multilevel Study of Collective Efficacy.” Science 277, no. 5328: 918–24.
https://doi.org/10.1126/science.277.5328.918.
Schmidt, M. D., P. Pekow, P. S. Freedson, G. Markenson, and L. Chasan-Taber. 2006. “Physical
Activity Patterns during Pregnancy in a Diverse Population of Women.” Journal of
Women’s Health 15, no. 8 (October): 909–18. https://doi.org/10.1089/jwh.2006.15.909.
Sherman, J. E., J. Spencer, J. S Preisser, W. M Gesler, and T. A. Arcury. 2005. “A Suite of
Methods for Representing Activity Space in a Healthcare Accessibility Study.”
International Journal of Health Geographics 4, no. 1 (December): 24.
https://doi.org/10.1186/1476-072X-4-24.
Shrestha, Samjhana, Yan Kestens, Frédérique Thomas, Tarik El Aarbaoui, and Basile Chaix.
2019. “Spatial Access to Sport Facilities from the Multiple Places Visited and Sport
Practice: Assessing and Correcting Biases Related to Selective Daily Mobility.” Social

143
Science & Medicine 236, no. September (September): 112406.
https://doi.org/10.1016/j.socscimed.2019.112406.
Silva, S. G. da, K. R. Evenson, U. Ekelund, I. C. M. da Silva, M. R. Domingues, B. G. C. da
Silva, M. de Almeida Mendes, G. I. N. Cruz, and P. C. Hallal. 2019. “How Many Days
Are Needed to Estimate Wrist-Worn Accelerometry-Assessed Physical Activity during
the Second Trimester in Pregnancy?” Edited by Adewale L. Oyeyemi. PLOS ONE 14,
no. 6 (June): e0211442. https://doi.org/10.1371/journal.pone.0211442.
Sinclair, I., M. St-Pierre, G. Elgbeili, P. Bernard, C. Vaillancourt, S. Gagnon, and K. Needham
Dancause. 2019. “Psychosocial Stress, Sedentary Behavior, and Physical Activity during
Pregnancy among Canadian Women: Relationships in a Diverse Cohort and a
Nationwide Sample.” International Journal of Environmental Research and Public
Health 16, no. 24 (December): 5150. https://doi.org/10.3390/ijerph16245150.
Singh, G. K., M. D. Kogan, P. C. Van Dyck, and M. Siahpush. 2008. “Racial/Ethnic,
Socioeconomic, and Behavioral Determinants of Childhood and Adolescent Obesity in
the United States: Analyzing Independent and Joint Associations.” Annals of
Epidemiology 18, no. 9 (September): 682–95.
https://doi.org/10.1016/j.annepidem.2008.05.001.
Sister, C., J. Wolch, and J. Wilson. 2010. “Got Green? Addressing Environmental Justice in Park
Provision.” GeoJournal 75, no. 3 (June): 229–48. https://doi.org/10.1007/s10708-009-
9303-8.
Smith, L., L. Foley, and J. Panter. 2019. “Activity Spaces in Studies of the Environment and
Physical Activity: A Review and Synthesis of Implications for Causality.” Health &
Place 58, no. July (July): 102113. https://doi.org/10.1016/j.healthplace.2019.04.003.
Soltani, H., and R. B. Fraser. 2000. “A Longitudinal Study of Maternal Anthropometric Changes
in Normal Weight, Overweight and Obese Women during Pregnancy and Postpartum.”
British Journal of Nutrition 84, no. 1 (July): 95–101.
https://doi.org/10.1017/S0007114500001276.
Straczkiewicz, M., P. James, and J.-P. Onnela. 2021. “A Systematic Review of Smartphone-
Based Human Activity Recognition Methods for Health Research.” Npj Digital Medicine
4, no. 1 (December): 148. https://doi.org/10.1038/s41746-021-00514-4.
Taveras, E. M., M. W. Gillman, K. Kleinman, J. W. Rich-Edwards, and S. L. Rifas-Shiman.
2010. “Racial/Ethnic Differences in Early-Life Risk Factors for Childhood Obesity.”
Pediatrics 125, no. 4 (April): 686–95. https://doi.org/10.1542/peds.2009-2100.
Thierry, B., B. Chaix, and Y. Kestens. 2013. “Detecting Activity Locations from Raw GPS Data:
A Novel Kernel-Based Algorithm.” International Journal of Health Geographics 12, no.
1 (December): 14. https://doi.org/10.1186/1476-072X-12-14.

144
Thomson, J. L., M. H. Goodman, and A. S. Landry. 2019. “Assessment of Neighborhood Street
Characteristics Related to Physical Activity in the Lower Mississippi Delta.” Health
Promotion Perspectives 9, no. 1 (January): 24–30. https://doi.org/10.15171/hpp.2019.03.
Torres M. T., M. Miri, L. Alonso, M. Dolores Gómez-Roig, M. Foraster, and P. Dadvand. 2020.
“Exposure to Greenspace and Birth Weight in a Middle-Income Country.” Environmental
Research 189, no. October (October): 109866.
https://doi.org/10.1016/j.envres.2020.109866.
Travert, A.-S., K. Annerstedt, and M. Daivadanam. 2019. “Built Environment and Health
Behaviors: Deconstructing the Black Box of Interactions—A Review of Reviews.”
International Journal of Environmental Research and Public Health 16, no. 8 (April):
1454. https://doi.org/10.3390/ijerph16081454.
Troiano, R. P., D. Berrigan, K. W. Dodd, L. C. Mâsse, T. Tilert, and M. Mcdowell. 2008.
“Physical Activity in the United States Measured by Accelerometer.” Medicine &
Science in Sports & Exercise 40, no. 1 (January): 181–88.
https://doi.org/10.1249/mss.0b013e31815a51b3.
Tsai, W. L., A. J. S. Davis, and L. E. Jackson. 2019. “Associations between Types of Greenery
along Neighborhood Roads and Weight Status in Different Climates.” Urban Forestry &
Urban Greening 41, no. May (May): 104–17. https://doi.org/10.1016/j.ufug.2019.03.011.
Van Dyck, D., G. Cardon, B. Deforche, J. F. Sallis, N. Owen, and I. De Bourdeaudhuij. 2010.
“Neighborhood SES and Walkability Are Related to Physical Activity Behavior in
Belgian Adults.” Preventive Medicine 50, no. January (January): S74–79.
https://doi.org/10.1016/j.ypmed.2009.07.027.
Varshavsky, J., A. Smith, A. Wang, E. Hom, M. Izano, H. Huang, A. Padula, and T. J.
Woodruff. 2020. “Heightened Susceptibility: A Review of How Pregnancy and Chemical
Exposures Influence Maternal Health.” Reproductive Toxicology 92, no. March (March):
14–56. https://doi.org/10.1016/j.reprotox.2019.04.004.
Walker, L. O., E. R. Fowles, and B. S. Sterling. 2011. “The Distribution of Weight‐Related Risks
Among Low‐Income Women During the First Postpartum Year.” Journal of Obstetric,
Gynecologic & Neonatal Nursing 40, no. 2 (March): 198–205.
https://doi.org/10.1111/j.1552-6909.2011.01231.x.
Walter, J. R., W. Perng, K. P. Kleinman, S. L. Rifas-Shiman, J. W. Rich-Edwards, and E. Oken.
2015. “Associations of Trimester-Specific Gestational Weight Gain with Maternal
Adiposity and Systolic Blood Pressure at 3 and 7 Years Postpartum.” American Journal
of Obstetrics and Gynecology 212, no. 4 (April): 499.e1-499.e12.
https://doi.org/10.1016/j.ajog.2014.11.012.
Wang, B., L. Gao, and Z. Juan. 2017. “A Trip Detection Model for Individual Smartphone-Based
GPS Records with a Novel Evaluation Method.” Advances in Mechanical Engineering 9,
no. 6 (June): 168781401770506. https://doi.org/10.1177/1687814017705066.

145
Weir, Z., J. Bush, S. C Robson, C. McParlin, J. Rankin, and R. Bell. 2010. “Physical Activity in
Pregnancy: A Qualitative Study of the Beliefs of Overweight and Obese Pregnant
Women.” BMC Pregnancy and Childbirth 10, no. 1 (December): 18.
https://doi.org/10.1186/1471-2393-10-18.
Wit, L., J. G. M. Jelsma, M. N. M. van Poppel, A. Bogaerts, D. Simmons, G. Desoye, R. Corcoy,
et al. 2015. “Physical Activity, Depressed Mood and Pregnancy Worries in European
Obese Pregnant Women: Results from the DALI Study.” BMC Pregnancy and Childbirth
15, no. 1 (December): 158. https://doi.org/10.1186/s12884-015-0595-z.
Wolch, J., J. P. Wilson, and J. Fehrenbach. 2005. “Parks and Park Funding in Los Angeles: An
Equity-Mapping Analysis.” Urban Geography 26, no. 1 (February): 4–35.
https://doi.org/10.2747/0272-3638.26.1.4.
Wu, J., C. Jiang, G. Jaimes, S. Bartell, A. Dang, D. Baker, and R. J. Delfino. 2013. “Travel
Patterns during Pregnancy: Comparison between Global Positioning System (GPS)
Tracking and Questionnaire Data.” Environmental Health 12, no. 1 (December): 86.
https://doi.org/10.1186/1476-069X-12-86.
Yi, L., J. P. Wilson, T. B. Mason, R. Habre, S. Wang, and G. F. Dunton. 2019. “Methodologies
for Assessing Contextual Exposure to the Built Environment in Physical Activity Studies:
A Systematic Review.” Health & Place 60, no. November (November): 102226.
https://doi.org/10.1016/j.healthplace.2019.102226.
Yin, L., 2017. “Street Level Urban Design Qualities for Walkability: Combining 2D and 3D GIS
Measures.” Computers, Environment and Urban Systems 64, no. July (July): 288–96.
https://doi.org/10.1016/j.compenvurbsys.2017.04.001.
Yin, L., Q. Cheng, Z. Wang, and Z. Shao. 2015. “‘Big Data’ for Pedestrian Volume: Exploring
the Use of Google Street View Images for Pedestrian Counts.” Applied Geography 63
(2015): 337-345. https://doi.org/10.1016/j.apgeog.2015.07.010.
Zeger, S. L., D. Thomas, F. Dominici, J. M. Samet, J. Schwartz, D. Dockery, and A. Cohen.
"Exposure measurement error in time-series studies of air pollution: concepts and
consequences." Environmental Health Perspectives 108, no. 5 (2000): 419-426. http://
doi: 10.1289/ehp.00108419
Zenk, S. N., S. A. Matthews, A. N. Kraft, and K. K. Jones. 2018. “How Many Days of Global
Positioning System (GPS) Monitoring Do You Need to Measure Activity Space
Environments in Health Research?” Health and Place.
https://doi.org/10.1016/j.healthplace.2018.02.004.
Zenk, S. N., A. J. Schulz, S. A. Matthews, A. Odoms-Young, J. Wilbur, L. Wegrzyn, K. Gibbs,
C. Braunschweig, and C. Stokes. 2011. “Activity Space Environment and Dietary and
Physical Activity Behaviors: A Pilot Study.” Health & Place 17, no. 5 (September):
1150–61. https://doi.org/10.1016/j.healthplace.2011.05.001.

146
Zhan, Y., J. Liu, Z. Lu, H. Yue, J. Zhang, and Y. Jiang. 2020. “Influence of Residential
Greenness on Adverse Pregnancy Outcomes: A Systematic Review and Dose-Response
Meta-Analysis.” Science of The Total Environment 718, no. May (May): 137420.
https://doi.org/10.1016/j.scitotenv.2020.137420.
Zhao, P., M.-P. Kwan, and S. Zhou. 2018. “The Uncertain Geographic Context Problem in the
Analysis of the Relationships between Obesity and the Built Environment in
Guangzhou.” International Journal of Environmental Research and Public Health 15, no.
2 (February): 308. https://doi.org/10.3390/ijerph15020308.
Zhu, Q., B. Xia, Y. Zhao, H. Dai, Y. Zhou, Y. Wang, Q. Yang, et al. 2019. “Predicting
Gestational Personal Exposure to PM2.5 from Satellite-Driven Ambient Concentrations
in Shanghai.” Chemosphere 233, no. October (October): 452–61.
https://doi.org/10.1016/j.chemosphere.2019.05.251.

147
Appendix A: Confidence flagging algorithm for signal loss scenarios using the Maternal And Developmental
Risks from Environmental and Social stressors (MADRES) real-time and personal sampling study Global
Positioning System (GPS) data

Levels of Confidence Criteria Observations (N)
Extremely High Dprev ≤1 m AND Dnext ≤1 m 5,246
Very High Dnext ≤45 m/s* Tnext AND Tnext ≤30 min 23,935
High Dnext ≤45 m/s* Tnext AND 30< Tnext ≤120 min 23,335
Medium
Dnext >45 m/s* Tnext AND Dnext ≤45 m/s* (Tnext + Tsigloss)
AND Tnext ≤120 min
1,558
Low
[Dnext >45 m/s*Tnext AND Dnext ≤45 m/s*(Tnext + Tsigloss) AND Tnext >120 min] OR [Dnext >45 m/s* Tnext AND Dnext
>45 m/s* (Tnext + Tsigloss)]
139,784
No Information No distance/duration data available 383,338

Note: D = distance. T = Time. N = 577,196 observations where signal was lost. Signal loss scenarios = segments with signals from either GPS or network
source that are lost for ≥1min.

148
Appendix B: Rules used to classify contexts of stays detected with the Maternal And Developmental Risks
from Environmental and Social stressors (MADRES) real-time and personal sampling study Global
Positioning System (GPS) data
Spatial Contexts Rules
Home residential
1
Longest stay of the corresponding pregnancy/postpartum period (≥2 h)
Non-home residential
If a stay was assigned one of the following 2016 Southern California Association of Governments (SCAG) land use codes:
single-family residential, multi-family residential, mobile homes and trailer parks, mixed residential, and rural residential
2

Commercial and services
If a stay was assigned the following 2016 SCAG land use codes: general office, commercial and services, industrial land
uses
2

Parks and open space If a stay was assigned the following 2016 SCAG land use code: “open space and recreation”
Schools and public facilities If a stay was assigned the following 2016 SCAG land use codes: “Facilities,” “Education.”
Other
If a stay was assigned a 2016 SCAG land use code besides the above mentioned codes or did not get a 2016 SCAG land use
code assigned.
Indoor/Outdoor Locations Rules
Home indoor
If a stay was classified as home residential and fell inside the 1 m buffer of the nearest 2014 Los Angeles County building
footprint polygon
3
Home outdoor
If a stay was classified as home residential and fell outside the 1 m buffer of the nearest 2014 Los Angeles County building
footprint polygon
3

Non-home indoor
If a stay was classified as one of non-home contexts and fell inside the 1 m buffer of the nearest 2014 Los Angeles County
building footprint polygon
3

Non-home outdoor
If a stay was classified as one of non-home contexts and fell outside the 1 m buffer of the nearest 2014 Los Angeles County
building footprint polygon
3

Out of Los Angeles County
If a stay was out of Los Angeles County, its indoor/outdoor status was not classified since building footprints Geographic
Information System (GIS) data were unavailable.
1
The “home residential” context was assigned at the last step, which was defined as the longest stay (a minimum of 2 h) detected within a
pregnancy/postpartum period regardless of prior-assigned contexts.
2
An existing land use was assigned to a stay if the stay point fell within the 15 m buffer of land use parcel boundaries.
3
1 m rather than the exact boundary was chosen to account for scenarios when indoor activities occurred mainly on corners of the building, resulting in a stay
that fell outside the building footprint polygon and was misclassified as outdoor.

149
Appendix C: An illustration of the GPS imputation rules based on existing data and home locations detected
from context classification.

1
Home day is defined as the day during which mothers answered home (indoor) / home (outdoor) in the physical context question in all four or five survey
prompts.
Note: GPS = Global Positioning System. Obs. = Observations.

150

Appendix D: Individual sociodemographics, neighborhood characteristics, and temporal factors included
in the study along with their measures and data sources
Categories List Measure Source
Individual
Sociodemographic
Age Continuous
MADRES maternal demographics
data
Education
Marriage Living together/married versus others
Acculturation US born versus Foreign born
Parity 0 versus >1 child
Employment status Employed versus unemployed
MADRES 1
st
trimester, 3
rd

trimester, and 6-month
questionnaires
Safety and cohesion
score
Neighborhood cohesion scale (on a scale of 1 to 5 with higher score
indicates better cohesion) developed by Sampson (1997)
MADRES 2
nd
trimester and 6-month
questionnaires
Neighborhood
characteristics
National Walkability
Index score
On a scale of 1 to 20 with higher indicates more walkable
neighborhoods
EPA EnviroAtlas
Area Deprivation Index
score
Ranking relatively to other neighborhoods within state of California
from domains of income, education, employment, and housing
quality (in a scale of 1 to 10 with higher means more disadvantaged)
2018 Area Deprivation Index from
Neighborhood Atlas®, University of
Wisconsin-Madison
Temporal factors
Universal age
1

Continuous ages calculated from pre- to post-birth, with zero
indicates fetus delivery date
MADRES questionnaires
Pregnancy/postpartum
periods
1
st
trimester, 3
rd
trimester, 4-6 month postpartum MADRES questionnaires
Weekends 1 = weekend days Derived from GPS timestamp
1
Universal age is calculated as a continuous variable (starting from conception till end of monitoring period) as an alternative to pregnancy/postpartum periods
for sensitivity analysis purposes.

151
Appendix E: Zero-inflated Generalized Linear Mixed-Effects Model (GLMM) results for changes in time-
activity patterns by pregnancy and postpartum period

Non-Home
Residential
(min/day)
Commercial
and Services
(min/day)
Industrial
and Office
Spaces
(min/day)
Schools and Public
Facilities (min/day)
Parks and
Open Spaces
(min/day)
Home
Residence
Outdoor
(min/day)
All Non-Home
Contexts Outdoor
(min/day)
Predictors
Count Model Incidence Rate Ratio (95%CI)
Period: 1
st

Trimester
1.21
(.69-2.12)
.98
(.75-1.28)
1.29
(.85-1.97)
.73
(.43-1.24)
1.77
(.65-4.81)
.87
(.74-1.02)
1.15
(.85-1.58)
Period: 4-6 Months
Postpartum
1.61
(.92-2.84)
.82
(.62-1.09)
1.33
(.87-2.02)
.75
(.44-1.28)
.46
(.12-1.78)
.9
(.75-1.1)
1.07
(.78-1.47)
Valid GPS
observation (h/d)
1.01
(.94-1.08)
1.08 **
(1.03-1.13)
1.03
(.98-1.08)
1.02
(.93-1.11)
.95
(.78-1.15)
1.06 ***
(1.04-1.08)
1.03
(.98-1.07)
Zero-Inflated Model Odds Ratio (95%CI)
Period: 1
st

Trimester
.87
(1.63-.46)
.63
(1.12-.36)
.61
(1.14-.33)
.94
(1.79-.5)
1.35
(3.82-.48)
.56
(2.13-.15)
1.03
(1.79-.59)
Period: 4-6 Months
Postpartum
1.06
(1.98-.57)
.42 **
(.77-.23)
.55
(1.05-.29)
.85
(1.65-.44)
1.19
(3.63-.39)
1.02
(4.91-.21)
.73
(1.29-.42)
Valid GPS
observation (h/d)
.3 *
(.82-.11)
.29 *
(.83-.1)
.18 *
(.94-.03)
.15 **
(.52-.04)
.23
(2.65-.02)
.2
(1.84-.02)
.23 *
(.81-.07)
*p<.05. **p<.01. ***p<.001. Exponentiated parameter estimates are shown. Reversed odds ratio (i.e., odds for an outcome to be non-zero) of zero-inflated
models were calculated for easier interpretation.
Note: GPS = Global Positioning System.

152
Appendix F: Zero-inflated Generalized Linear Mixed-Effects Model (GLMM) results for changes in daily
mobility patterns by pregnancy and postpartum period
Pedestrian-based Trip
(min/d)
Vehicular-based Trip
(min/d)
Pedestrian-based Trip
(N/d)
Vehicular-based Trip
(N/d)
Predictors

Count Model Incidence Rate Ratio (95%CI)
Period: 1
st
Trimester 1.03
(.75-1.42)
1.05
(.86-1.28)
1.03
(.72-1.47)
1.03
(.87-1.23)
Period: 4-6 Months
Postpartum
.96
(.69-1.32)
1.17
(.95-1.44)
.82
(.56-1.2)
1
(.84-1.19)
Valid GPS
observation (h/d)
1.02
(.97-1.08)
1.05 **
(1.02-1.09)
.97
(.92-1.03)
1.04 *
(1.01-1.07)
Zero-Inflated Model Odds Ratio (95%CI)
Period: 1
st
Trimester .72
(1.23-.43)
.8
(1.45-.45)
.62
(1.43-.26)
.72
(1.46-.35)
Period: 4-6 Months
Postpartum
.61
(1.05-.35)
.5 *
(.93-.27)
.58
(1.45-.24)
.44 *
(.92-.21)
Valid GPS
observation (h/d)
.21 *
(.86-.05)
.35 *
(.92-.13)
.21
(1.55-.03)
.28
(1.11-.07)
*p<.05. **p<.01. ***p<.001. Exponentiated parameter estimates are shown. Reversed odds ratio (i.e., odds for an outcome to be non-zero) of zero-inflated
models were calculated for easier interpretation.
Note: GPS = Global Positioning System.

153
Appendix G: Sensitivity analyses results of between-subject (BS) and within-subject (WS) effects of built-
environment (BE) exposures on day-level moderate-to-vigorous physical activity (MVPA) outcomes
MVPA min/d
100 m KDE model 100 RB model 800 m RNB model 1,600 m RNB model
Predictors Estimates (95%CI)
(Intercept)
4.43
*

(1.08 – 18.22)
4.07
*

(1.01 – 16.29)
1.19
(.14 – 9.93)
.50
(.04 – 6.23)
% Green space along walkable roads
(BS)
1.00
(.99 – 1.02)
1.01
(.98 – 1.05)
1.01
(.98 – 1.03)
1.01
(.98 – 1.03)
Distance to the nearest park entrance
(BS)
1.01
(.96 – 1.05)
1.04
(.97 – 1.12)
1.00
(1.00 – 1.00)
1.00
(1.00 – 1.00)
Walkability index score (BS)
1.03
(.95 – 1.12)
1.14
(.99 – 1.31)
1.07
(.98 – 1.17)
1.12
(1.00 – 1.25)
Daily parks and open space exposure
(BS)
1.41
(.85 – 2.32)
1.51
(.92 – 2.48)
1.13
(.79 – 1.60)

% Green space along walkable roads
(WS)
.99
(.98 – 1.01)
1.01
(.99 – 1.03)

Distance to the nearest park entrance
(WS)
.99
(.94 – 1.05)
.99
(.96 – 1.02)

Walkability index score (WS)
1.00
(.88 – 1.13)
1.02
(.94 – 1.11)

Daily parks and open space exposure
(WS)
1.14
(.97 – 1.36)
1.17
(.99 – 1.38)

Maternal age
1.00
(.97 – 1.03)
1.00
(.97 – 1.03)
1.00
(.97 – 1.03)
1.00
(.97 – 1.03)
Education: Some college/Graduate
.69
*

(.49 – .97)
.70
*

(.50 – .97)
.73
*

(.54 – .99)
.75
(.55 – 1.01)
Parity: First-born
1.21
(.85 – 1.74)
1.21
(.86 – 1.72)
1.24
(.85 – 1.79)
1.22
(.86 – 1.75)

154
Employment status: Employed
1.03
(.82 – 1.28)
1.02
(.82 – 1.27)
1.06
(.85 – 1.32)
1.07
(.86 – 1.33)
Pre-pregnancy BMI category:
Overweight/Obesity
.77
(.56 – 1.04)
.75
(.56 – 1.01)
.78
(.58 – 1.06)
.79
(.59 – 1.06)
Average daily temperature (°C)
1.02
*

(1.00 – 1.04)
1.02
*

(1.00 – 1.04)
1.02
*

(1.00 – 1.04)
1.02
*

(1.00 – 1.04)
Type of day: Weekend
.81
**

(.70 – .93)
.80
**

(.70 – .92)
.80
**

(.70 – .92)
.80
**

(.70 – .92)
The 3
rd
trimester day
.95
(.79 – 1.13)
.94
(.79 – 1.12)
.95
(.80 – 1.13)
.95
(.80 – 1.14)
4-6 months postpartum day
.89
(.74 – 1.07)
.90
(.75 – 1.08)
.89
(.74 – 1.07)
.89
(.74 – 1.07)
Neighborhood cohesion and safety
score (range from 1 to 5)
1.04
(.92 – 1.17)
1.03
(.91 – 1.16)
1.04
(.92 – 1.17)
1.04
(.92 – 1.17)
Neighborhood deprivation index (range
from 1 to 10)
1.02
(.94 – 1.12)
1.05
(.96 – 1.14)
1.01
(.93 – 1.10)
1.02
(.95 – 1.11)
Daily accelerometry wearing hours
1.09
***

(1.06 – 1.13)
1.09
***

(1.06 – 1.13)
1.10
***

(1.06 – 1.13)
1.10
***

(1.07 – 1.13)
Total valid accelerometry collection
days
1.01
(.96 – 1.06)
1.01
(.97 – 1.06)
1.01
(.97 – 1.06)
1.01
(.97 – 1.06)
* p<.05 ** p<.01 *** p<.001. Estimates were exponentiated for the ease of interpretation.
Note: KDE = Kernel Density Estimation. RB = Route Buffer. RNB = Residential Network Buffer.

155
Appendix H: Significant interactions between day-level built-environment (BE) exposures and temporal,
individual and neighborhood factors in predicting day-level moderate-to-vigorous physical activity (MVPA)
outcomes
MVPA min/d

250 m KDE x
4-6 months
postpartum
250 m KDE x
Pre-pregnancy
BMI
250 m KDE x
Neighborhood
safety
250 m RB
Average daily
temperature
250 m RB x
Weekend days
250 m RB x the
3
rd
trimester
250 m RB x
First-born
Predictors Estimates (95%CI)
(Intercept)
7.10
***

(2.85 – 17.69)
8.72
***

(3.42 – 22.25)
4.40
**

(1.54 – 12.58)
6.83
***

(2.41 – 19.33)
8.76
***

(3.46 – 22.15)
9.45
***

(3.71 – 24.05)
8.70
***

(3.15 – 24.05)

% Green space along
walkable roads (BS)
1.01
(.99 – 1.03)
1.01
(.99 – 1.03)
1.01
(.99 – 1.03)
1.03
(1.00 – 1.06)
1.03
(1.00 – 1.06)
1.03
(1.00 – 1.06)
1.03
(1.00 – 1.06)

Distance to the nearest park
entrance (BS)
1.01
(.98 – 1.05)
1.01
(.98 – 1.05)
1.02
(.98 – 1.05)
1.04
(.99 – 1.10)
1.04
(.99 – 1.09)
1.05
(1.00 – 1.10)
1.04
(1.00 – 1.10)

Walkability index score (BS)
1.05
(.97 – 1.14)
1.05
(.97 – 1.14)
1.05
(.97 – 1.14)
1.18
*

(1.04 – 1.35)
1.17
*

(1.03 – 1.32)
1.20
**

(1.05 – 1.36)
1.18
*

(1.04 – 1.34)

Daily parks and open space
exposure (BS)
1.62
(.98 – 2.67)
1.66
*

(1.01 – 2.74)
1.57
(.95 – 2.59)
1.50
(.93 – 2.40)
1.55
(.97 – 2.48)
1.46
(.91 – 2.33)
1.46
(.91 – 2.34)

% Green space along
walkable roads (WS)
1.00
(.98 – 1.03)
1.03
(.99 – 1.06)
1.07
(1.00 – 1.15)
1.04
(.95 – 1.13)
1.00
(.97 – 1.03)
1.01
(.99 – 1.04)
1.00
(.98 – 1.02)

Distance to the nearest park
entrance (WS)
1.00
(.93 – 1.07)
1.08
(.92 – 1.27)
1.02
(.78 – 1.35)
.92
(.80 – 1.05)
.98
(.93 – 1.03)
.96
(.93 – 1.00)
.97
(.93 – 1.01)

Walkability index score
(WS)
1.12
(.95 – 1.31)
.85
(.64 – 1.13)
1.03
(.58 – 1.84)
.68
*

(.49 – .94)
.91
(.79 – 1.04)
.99
(.90 – 1.09)
.95
(.86 – 1.06)

Daily parks and open space
exposure (WS)
1.30
*

(1.05 – 1.62)
1.09
(.75 – 1.58)
2.72
**

(1.30 – 5.66)
1.63
(.82 – 3.22)
1.23
(.98 – 1.55)
1.21
(.98 – 1.49)
1.23
*

(1.00 – 1.50)

4-6 months postpartum
1.05
(.77 – 1.44)
.94
(.79 – 1.13)
.93
(.78 – 1.11)
.89
(.75 – 1.06)
.91
(.76 – 1.09)
.91
(.76 – 1.08)
.90
(.75 – 1.07)

Average daily
temperature(°C)
1.02
*

(1.00 – 1.04)
1.01
(1.00 – 1.03)
1.02
*

(1.00 – 1.04)
1.03
*

(1.01 – 1.06)
1.02
*

(1.00 – 1.04)
1.02
*

(1.00 – 1.04)
1.02
*

(1.00 – 1.04)

156
MVPA min/d

250 m KDE x
4-6 months
postpartum
250 m KDE x
Pre-pregnancy
BMI
250 m KDE x
Neighborhood
safety
250 m RB
Average daily
temperature
250 m RB x
Weekend days
250 m RB x the
3
rd
trimester
250 m RB x
First-born
Predictors Estimates (95%CI)
Type of day: Weekend
.81
**

(.71 – .93)
.82
**

(.71 – .94)
.81
**

(.71 – .93)
.82
**

(.71 – .93)
.86
(.66 – 1.13)
.80
**

(.70 – .92)
.81
**

(.71 – .93)

The 3
rd
trimester
.96
(.81 – 1.13)
1.01
(.85 – 1.20)
.94
(.80 – 1.12)
.93
(.79 – 1.10)
.96
(.81 – 1.13)
.93
(.69 – 1.27)
.95
(.80 – 1.12)

Maternal age
.99
(.97 – 1.01)
.99
(.97 – 1.01)
.99
(.97 – 1.01)
.99
(.96 – 1.01)
.99
(.96 – 1.01)
.99
(.96 – 1.01)
.99
(.96 – 1.01)

Education: Some
college/Graduate
.70
*

(.53 – .92)
.69
**

(.52 – .91)
.67
**

(.50 – .88)
.67
**

(.50 – .89)
.66
**

(.50 – .88)
.66
**

(.49 – .88)
.66
**

(.49 – .89)

Pre-pregnancy BMI: Obesity
.92
(.68 – 1.25)
.92
(.60 – 1.42)
.90
(.67 – 1.22)
.95
(.71 – 1.27)
.95
(.71 – 1.27)
.96
(.72 – 1.29)
.94
(.71 – 1.26)

Pre-pregnancy BMI:
Overweight
.66
**

(.49 – .89)
.55
**

(.35 – .87)
.69
*

(.52 – .93)
.66
**

(.49 – .87)
.65
**

(.49 – .86)
.66
**

(.50 – .88)
.66
**

(.49 – .87)

Neighborhood cohesion and
safety score (range from 1
to 5)
1.04
(.93 – 1.16)
1.05
(.93 – 1.17)
1.23
*

(1.01 – 1.50)
1.02
(.92 – 1.14)
1.03
(.92 – 1.15)
1.04
(.93 – 1.16)
1.02
(.91 – 1.15)

Daily accelerometer
wearing hours
1.09
***

(1.06 – 1.12)
1.09
***

(1.06 – 1.13)
1.09
***

(1.06 – 1.13)
1.09
***

(1.06 – 1.13)
1.09
***

(1.06 – 1.13)
1.08
***

(1.05 – 1.12)
1.09
***

(1.06 – 1.13)

% Green space along
walkable roads (WS) x 4-6
months postpartum
.98
(.94 – 1.03)

Distance to the nearest park
entrance (WS) x 4-6 months
postpartum
.96
(.85 – 1.09)

Walkability index score
(WS) x 4-6 months
postpartum x 4-6 months
postpartum
.72
*

(.54 – .97)

157
MVPA min/d

250 m KDE x
4-6 months
postpartum
250 m KDE x
Pre-pregnancy
BMI
250 m KDE x
Neighborhood
safety
250 m RB
Average daily
temperature
250 m RB x
Weekend days
250 m RB x the
3
rd
trimester
250 m RB x
First-born
Predictors Estimates (95%CI)
Daily parks and open space
exposure (WS) x 4-6 months
postpartum
.80
(.56 – 1.15)

Parity: First-born

.97
(.63 – 1.50)

% Green space along
walkable roads (WS) x
Parity: First-born

1.04
(.99 – 1.09)

Distance to the nearest park
entrance (WS) x Parity:
First-born

1.04
(.96 – 1.12)

Walkability index score
(WS) x 4-6 months
postpartum x Parity: First-
born

1.23
*

(1.02 – 1.49)

Daily parks and open space
exposure (WS) x Parity:
First-born

1.04
(.69 – 1.58)

% Green space along
walkable roads (WS) x Pre-
pregnancy BMI: Obesity

.92
**

(.87 – .97)

% Green space along
walkable roads (WS) x Pre-
pregnancy BMI: Overweight

.97
(.93 – 1.02)

Distance to the nearest park
entrance (WS) x Pre-
pregnancy BMI: Obesity

1.01
(.84 – 1.20)

158
MVPA min/d

250 m KDE x
4-6 months
postpartum
250 m KDE x
Pre-pregnancy
BMI
250 m KDE x
Neighborhood
safety
250 m RB
Average daily
temperature
250 m RB x
Weekend days
250 m RB x the
3
rd
trimester
250 m RB x
First-born
Predictors Estimates (95%CI)
Distance to the nearest park
entrance (WS) x Pre-
pregnancy BMI: Overweight

.85
(.71 – 1.03)

Walkability index score
(WS) x 4-6 months
postpartum x Pre-pregnancy
BMI: Obesity

1.49
*

(1.04 – 2.12)

Walkability index score
(WS) x 4-6 months
postpartum x Pre-pregnancy
BMI: Overweight

1.17
(.80 – 1.70)

Daily parks and open space
exposure (WS) x Pre-
pregnancy BMI: Obesity

1.01
(.64 – 1.59)

Daily parks and open space
exposure (WS) x Pre-
pregnancy BMI: Overweight

1.33
(.84 – 2.10)

% Green space along
walkable roads (WS) x
Neighborhood cohesion and
safety

.98
*

(.96 – 1.00)

Distance to the nearest park
entrance (WS) x
Neighborhood cohesion and
safety

.99
(.90 – 1.09)

Walkability index score
(WS) x 4-6 months
postpartum x Neighborhood
cohesion and safety

.99
(.81 – 1.21)

159
MVPA min/d

250 m KDE x
4-6 months
postpartum
250 m KDE x
Pre-pregnancy
BMI
250 m KDE x
Neighborhood
safety
250 m RB
Average daily
temperature
250 m RB x
Weekend days
250 m RB x the
3
rd
trimester
250 m RB x
First-born
Predictors Estimates (95%CI)
Daily parks and open space
exposure (WS) x
Neighborhood cohesion and
safety

.78
*

(.62 – .97)

% Green space along
walkable roads (WS) x
Average daily temperature

1.00
(.99 – 1.00)

Distance to the nearest park
entrance (WS) x Average
daily temperature

1.00
(1.00 – 1.01)

Walkability index score
(WS) x Average daily
temperature

1.02
**

(1.01 – 1.04)

Daily parks and open space
exposure (WS) x Average
daily temperature

.99
(.95 – 1.02)

% Green space along
walkable roads (WS) x
Weekend days

1.02
(.98 – 1.06)

Distance to the nearest park
entrance (WS) x Weekend
days

1.01
(.94 – 1.08)

Walkability index score
(WS) x Weekend days

1.25
*

(1.04 – 1.51)

Daily parks and open space
exposure (WS) x Weekend
days

.92
(.67 – 1.25)

160
MVPA min/d

250 m KDE x
4-6 months
postpartum
250 m KDE x
Pre-pregnancy
BMI
250 m KDE x
Neighborhood
safety
250 m RB
Average daily
temperature
250 m RB x
Weekend days
250 m RB x the
3
rd
trimester
250 m RB x
First-born
Predictors Estimates (95%CI)
% Green space along
walkable roads (WS) x the
3
rd
trimester

1.00
(.96 – 1.05)

Distance to the nearest park
entrance (WS) x the 3
rd

trimester

1.10
*

(1.01 – 1.20)

Walkability index score
(WS) x the 3
rd
trimester

1.20
(.95 – 1.53)

Daily parks and open space
exposure (WS) x the 3
rd

trimester

1.01
(.71 – 1.43)

* p<.05 ** p<.01 *** p<.001. Estimates were exponentiated for the ease of interpretation.
Note: KDE = Kernel Density Estimation. MVPA = Moderate-to-Vigorous Physical Activity. RB = Route Buffer.

Abstract (if available)

Abstract Increasing rates of maternal obesity during and after pregnancy are associated with numerous short- and long-term health risks and outcomes in women postpartum. Hispanic women of childbearing age have disproportionally high obesity risks and low physical activity (PA) levels compared to non-Hispanic white populations. This dissertation study used 4-day objective, highly resolved smartphone location and accelerometry-assessed activity data collected from 62 Hispanic women of childbearing age in urban Los Angeles, CA, during pregnancy and the early postpartum period to investigate the associations among women’s daily mobility patterns (Chapter 2), dynamic built-environment (BE) exposures (Chapter 3), and PA outcomes (Chapter 4). Results of these three empirical case studies revealed exceedingly low parks and open space exposure for this group at their daily activity locations and along travel paths during and after pregnancy, which had negative impacts on their day-to-day PA outcomes. In addition, important modifiers (e.g., late pregnancy, early postpartum, high pre-pregnancy BMI, low neighborhood safety) of women’s daily mobility and BE exposures, and their associations with PA outcomes were identified. Lastly, measurement error and bias resulting from applying traditional residential-based measures were evaluated and their implications for uncovering the relationships between BE exposures and PA outcomes were investigated. Future studies should conduct qualitative analyses of the BE features (e.g., parks) in which women’s PA were performed, mitigate selective daily mobility bias, and apply real-time surveying techniques such as ecological momentary assessment (EMA) to elucidate psychosocial pathways from BE exposures to PA outcomes. Future PA promotion interventions for pregnant women should target at-risk pregnancy periods and sub-population groups to improve their efficacy, especially for those of low socioeconomic status and specific racial/ethnic minority groups.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Personal PM2.5 exposure during pregnancy in an environmental health disparities population

PDF

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

PDF

Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data

PDF

Influences of specific environmental domains on childhood obesity and related behaviors

PDF

Air pollution, mitochondrial function, and growth in children

PDF

Personal exposure to particulate matter PM2.5 sources during pregnancy and birthweight

PDF

Effects of near-roadway air pollution exposure on obesity, obesity-related behavior, and neurobehavioral deficits during peripuberty

PDF

Prenatal environmental exposures and fetal growth in the MADRES cohort

PDF

Susceptibility to autism likelihood associated with prenatal ambient air pollution exposures: identifying vulnerable populations

PDF

Associations of ambient air pollution exposures with perceived stress in the MADRES cohort

PDF

Association of maternal and environmental factors with infant feeding behaviors in a birth cohort study

PDF

Pregnancy in the time of COVID-19: effects on perinatal mental health, birth, and infant development

PDF

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

PDF

Disparities in gallbladder, intra-hepatic bile duct, and other biliary cancers among multi-ethnic populations: a California Cancer Registry study

PDF

Characterizing self-reported spatial and provider barriers to maternal health care utilization in Malawi

PDF

Carcinogenic exposures in racial/ethnic groups

PDF

Assessing the impact of air pollution on adverse birth outcomes in a low resource setting

PDF

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

PDF

The role of physicians in the late-stage diagnosis of melanoma among: a population-based study in Los Angeles County

PDF

Effect of biomass fuel exposure on infant respiratory health outcomes in Bangladesh

Asset Metadata

Creator Yi, Li (author)

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

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Health and Place,Population

Publication Date 03/21/2022

Defense Date 03/02/2022

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

Tag activity space,built environment,greenness,Hispanic women,maternal health,OAI-PMH Harvest,obesity,park,physical activity,Pregnancy,public transit,walkability

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Dunton, Genevieve (committee chair), Wilson, John (committee chair), Boeing, Geoff (committee member), Habre, Rima (committee member)

Creator Email lyi632@usc.edu,yili0910@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC110816578

Unique identifier UC110816578

Document Type Dissertation

Format application/pdf (imt)

Rights Yi, Li

Type texts

Source 20220323-usctheses-batch-917 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

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

Repository Email cisadmin@lib.usc.edu