University of Southern California Dissertations and Theses

System dynamics of in-home family eating behavior: insights from intensive longitudinal data using Ecological Momentary Assessment and wearable sensors

(USC Thesis Other)

System dynamics of in-home family eating behavior: insights from intensive longitudinal data using Ecological Momentary Assessment and wearable sensors

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SYSTEM DYNAMICS OF IN- HOME FAMILY EATING BEHAVIOR:
INSIGHTS FROM INTENSIVE LONGITUDINAL DATA USING ECOLOGICAL
MOMENTARY ASSESSMENT AND WEARABLE SENSORS
by
Brooke M. Bell
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PREVENTIVE MEDICINE (HEALTH BEHAVIOR RESEARCH))
May 2021
Copyright 2021 Brooke M. Bell

ii

DEDICATION
To the brave, brilliant, and revolutionary women in science who sacrificed, resisted, and
broke down barriers so that I could thrive.

iii

ACKNOWLEDGEMENTS
I would firstly like to acknowledge and thank my doctoral advisors and committee co-
chairs, Dr. Donna Spruijt-Metz and Dr. Kayla de la Haye. You both have provided immense
support and encouragement throughout my time as a doctoral student. Your brilliance,
thoughtfulness, insightfulness, and kindness inspire me every day. I am incredibly appreciative
of the time you have invested in and continue to invest in me and my future, and promise to pay
it forward. I truly couldn’t have asked for better advisors.
I would also like to acknowledge and thank my committee members: Dr. Thomas
Valente, Dr. Britni Belcher, and Dr. Shinyi Wu. I am grateful for your willingness to happily
share your expertise and for your support throughout the entire dissertation process.
I am grateful for the members of the entire M2FED team, including Dr. Jack Stankovic,
Dr. John Lach, Ridwan Alam, Abu Mondol, Ifat Emi, Meiyi Ma, Sarah Preum, Zeya Chen,
Mohsin Ahmad, Asif Salekin, Luz Antunez-Castillo, Yadira Garcia, Grace Do, Jessica Rayo,
Daisy Gonzalez, Megri Kartounian, and Matt Shi.
I would like to thank other faculty, colleagues, and co-authors – both at USC and beyond
– for their involvement in my education and growth as a scholar: Dr. Jennifer Unger, Dr. Akilah
Dulin, Dr. Ricky Bluthenthal, Dr. Jimi Huh, Dr. Adam Leventhal, Dr. Stefan Schneider, Dr.
Nabil Alshurafa, Dr. Erik Heckler, Dr. Cheng Wen, Dr. George Vega Yon, and Dr. Abigail
Horn.
A huge thank you to all of my doctoral friends and colleagues: Sydney O’Connor, Jessi
Tobin, Karen Ra, Afton Kechter, Jen Zink, Shelia Yu, Sam Cwalina, Georgia Christodoulou,
Anuja Majmundar, Artur Galimov, Christine Naya, Cynthia Ramirez, Bridgette Do, Shirlene
Wang, Maria Bolshakova, Cynthia Begay, Sydney Miller, Carol Ochoa, Sarah Piombo, Sheila

iv

Pakdaman, Joycelyn Yip, Bryce Tappan, and Kris Coombs. I wouldn’t have gotten through this
program without all of you.
I would also like to acknowledge and thank the staff members and IT support within the
Department of Preventive Medicine, including Andrew Zaw and Gregory (Ryan) Wilkerson.
A huge thank you to the entire USC Civic Engagement team, whose presence in my life
greatly enriched my experience at USC. A special thank you to Dulce Acosta, Carolina Castillo,
and Rocio Flores.
Thank you to my best friend, Zainab Taymuree, for being a radiant light in my life.
Finally, thank you to my partner Ryan Rose for your constant support and love.
This dissertation was supported by the National Science Foundation (1521740).

v

TABLE OF CONTENTS
Page

DEDICATION................................................................................................................... ii
ACKNOWLEDGEMENTS ............................................................................................ iii
LIST OF TABLES .............................................................................................................x
LIST OF FIGURES ......................................................................................................... xi
ABSTRACT ..................................................................................................................... xii
CHAPTER 1: INTRODUCTION .....................................................................................1
BACKGROUND AND SIGNIFICANCE ...............................................................1
Overweight and Obesity ..............................................................................1
Prevalence and trends ......................................................................1
Implications of overweight and obesity ...........................................2
Health disparities .............................................................................3
Past childhood obesity prevention and treatment efforts .................4
Gaps in the literature ........................................................................6
Eating Behavior as a Primary Risk Factor for Overweight/Obesity ............6
Etiology of overweight and obesity .................................................6
Dietary intake and health .................................................................7
Shifting focus: from diet to eating behavior and context .................8
Gaps in the literature ........................................................................9
Influences on Eating Behavior .....................................................................9
Complex systems framework ...........................................................9
Individual influences (Intra-personal)............................................11
Social environmental influences (Inter-personal) ..........................13
Family influences ...........................................................................15
Gaps in the literature ......................................................................19
Dietary Assessment Methodology .............................................................19
Overview of current dietary assessment tools ..............................20
Methodological limitations and advantages of current dietary
assessment tools .................................................................21

vi

Summary of technology-assisted dietary assessment tools ...........22
Methodological limitations and advantages of technology-assisted
dietary assessment tools .....................................................24
Gaps in the literature ......................................................................26
SUMMARY OF GAPS IN THE LITERATURE ..................................................26
SPECIFIC AIMS ...................................................................................................28
CHAPTER 2: INVESTIGATING THE VALIDITY AND FEASIBILITY OF THE
MONITORING AND MODELING FAMILY EATING DYNAMICS (M2FED)
SYSTEM TO DETECT IN-FIELD FAMILY EATING BEHAVIOR .......................30
ABSTRACT ...........................................................................................................30
INTRODUCTION .................................................................................................31
Challenges to Dietary Assessment .............................................................31
Shifting Focus: From Dietary Intake to Eating Behavior and Context ......32
Technology-Assisted Dietary Assessment .................................................33
M2FED Study ............................................................................................34
Assessing Validity of Wearable Sensors ...................................................34
Assessing Feasibility of EMA ...................................................................35
Study Aims.................................................................................................36
METHODS ............................................................................................................36
Participants and Recruitment .....................................................................36
M2FED System Overview .........................................................................37
Procedures ..................................................................................................41
Measures ....................................................................................................43
Analytic Approach .....................................................................................44
RESULTS ..............................................................................................................49
Individual- and Family-Level Characteristics ...........................................49
EMA Descriptives ......................................................................................50
Participant Compliance (Aim 1a) ..............................................................51
Predictors of Compliance (Aim 1b) ...........................................................51
Smartwatch Algorithm Evaluation (Aim 2a) .............................................53
Differences in Eating Event Detection (Aim 2b) .......................................53

vii

DICUSSION ..........................................................................................................54
Evaluating EMA Compliance ....................................................................54
Evaluating Automatic Dietary Assessment (ADA) ...................................55
Limitations and Strengths ..........................................................................57
Future Directions .......................................................................................58
CONCLUSION ......................................................................................................60
TABLES ................................................................................................................61
FIGURES ...............................................................................................................69
CHAPTER 3: UTILIZING WEARABLE SENSORS AND ECOLOGICAL
MOMENTARY ASSESSMENT (EMA) TO AUTOMATICALLY DETECT
CONTEXTUALIZED EATING EVENTS IN THE HOME: A MULTI-LEVEL
LATENT CLASS ANALYSIS ........................................................................................79
ABSTRACT ...........................................................................................................79
INTRODUCTION .................................................................................................80
Global Obesity Epidemic ...........................................................................80
Dietary Intake, Eating Behavior, and Health .............................................81
Emerging Technologies .............................................................................83
Intra-personal (Individual) Context ...........................................................84
Inter-personal (Social) Context ..................................................................84
Latent Class Analysis .................................................................................86
Study Aims.................................................................................................87
METHODS ............................................................................................................87
Participants and Recruitment .....................................................................87
M2FED System Overview .........................................................................88
Procedures ..................................................................................................90
Measures ....................................................................................................92
Analytic Approach .....................................................................................97
RESULTS ............................................................................................................100
Sample Characteristics .............................................................................100
Eating Event Descriptives ........................................................................100
Multi-Level Latent Class Analysis (Aim 1).............................................101

viii

Predictors of Latent Classes (Aim 2) .......................................................102
DISCUSSION ......................................................................................................103
General Discussion ..................................................................................103
Limitations and Advantages ....................................................................107
Implications..............................................................................................108
Future Directions and Conclusion ...........................................................109
TABLES ..............................................................................................................110
FIGURES .............................................................................................................116
CHAPTER 4: INVESTIGATING THE DYNAMICS OF IN-HOME FAMILY
EATING BEHAVIOR: A RELATIONAL EVENT MODELING APPROACH ....118
ABSTRACT .........................................................................................................118
INTRODUCTION ...............................................................................................119
Obesity as a Complex System..................................................................119
A Family Systems Approach ...................................................................121
Affective States and Eating Behavior ......................................................122
M2FED Study: A Systems Approach to Modeling Family Eating
Dynamics .....................................................................................123
Relational Event Modeling (REM) Framework ......................................125
Study Aims...............................................................................................125
METHODS ..........................................................................................................127
Participants and Recruitment ...................................................................127
M2FED System Overview .......................................................................128
Procedures ................................................................................................131
Measures ..................................................................................................132
Analytic Approach ...................................................................................134
RESULTS ............................................................................................................139
Sample Characteristics .............................................................................139
Event Descriptives ...................................................................................139
Relational Event Models (Aim 1) ............................................................140
Differences by Family Role (Aim 2) .......................................................142
DISCUSSION ......................................................................................................142

ix

General Discussion ..................................................................................142
Intervention Implications .........................................................................145
Methodological Implications ...................................................................146
Limitations and Advantages ....................................................................147
Future Directions .....................................................................................148
CONCLUSION ....................................................................................................149
TABLES ..............................................................................................................151
Chapter 5: DISCUSSION..............................................................................................156
SUMMARY OF FINDINGS ...............................................................................156
IMPLICATIONS .................................................................................................158
Methodological Implications ...................................................................158
Theoretical and Intervention Implications ...............................................159
OVERALL DISSERTATION LIMITATIONS ..................................................160
FUTURE RESEARCH DIRECTIONS ...............................................................161
mHealth and Behavioral Health Fields ....................................................161
Broader Public Health Field.....................................................................163
Systems approach to address health disparities ...........................163
The intersection of climate change, eating behaviors, and health165
REFERENCES ...............................................................................................................168

x

LIST OF TABLES
TABLE 1. M2FED ECOLOGICAL MOMENTARY ASSESSMENT (EMA) ITEMS ............................................................................................. 61
TABLE 2. COMPARISON OF RECRUITED SAMPLE AND ANALYTIC SAMPLES.................................................................................................. 63
TABLE 3. INDIVIDUAL-LEVEL CHARACTERISTICS OF M2FED ANALYTIC SAMPLE (N=58), BY FAMILY MEMBER ROLE
(DISSERTATION STUDIES 1 AND 3) ........................................................................................................................................................ 64
TABLE 4. FAMILY-LEVEL AND DEPLOYMENT-LEVEL CHARACTERISTICS OF M2FED STUDY FAMILIES (N=20) ................................ 65
TABLE 5. EMA SUMMARY STATISTICS AFTER APPLYING PARTICIPATION ALGORITHM, BY PROMPT TYPE ........................................... 66
TABLE 6. EMA COMPLIANCE RATES AFTER APPLYING PARTICIPATION ALGORITHM, BY PROMPT TYPE .............................................. 67
TABLE 7. LOGISTIC REGRESSION MODEL RESULTS, EXAMINING PREDICTORS OF COMPLIANCE .............................................................. 68
TABLE 8. INDIVIDUAL-LEVEL CHARACTERISTICS OF M2FED ANALYTIC SAMPLE (N=46), BY FAMILY MEMBER ROLE
(DISSERTATION STUDY 2)..................................................................................................................................................................... 110
TABLE 9. SUMMARY STATISTICS OF EATING EVENT INDICATOR VARIABLES AND COVARIATES (N=290) ........................................ 111
TABLE 10. FIT INDICES FOR MULTILEVEL LATENT CLASSES OF EATING EVENTS, WITHOUT COVARIATES......................................... 112
TABLE 11. CLASS COUNTS AND PROPORTIONS FOR 4-CLASS SOLUTION (N=290) ............................................................................... 113
TABLE 12. FIT INDICES FOR CONDITIONAL 4-CLASS MODEL OF EATING EVENTS, INCLUDING COVARIATES (N=290) .................. 114
TABLE 13. MULTI-LEVEL LATENT CLASS ANALYSIS RESULTS FOR 4-CLASS SOLUTION, INCLUDING COVARIATES ........................... 115
TABLE 14. COUNTS AND ROW PROPORTIONS OF OBSERVED AFFECTIVE STATES AND EATING EVENTS ............................................. 151
TABLE 15. RESULTS FROM THE EGOCENTRIC RELATIONAL EVENT MODELS (ORDINAL LIKELIHOOD) ............................................... 152
TABLE 16. RESULTS FROM FINAL RELATIONAL EVENT MODEL, STRATIFIED BY FAMILY ROLE ............................................................ 154

xi

LIST OF FIGURES
FIGURE 1. OVERVIEW OF M2FED CYBERPHYSICAL SYSTEM ......................................................................................................................... 69
FIGURE 2. EXAMPLES OF A (A) TIME-TRIGGERED AND (B) EATING EVENT-TRIGGERED MOBILE QUESTIONNAIRE RECEIVED ON A
PARTICIPANT ’S PHONE............................................................................................................................................................................... 70
FIGURE 3. EATING EVENT-TRIGGERED EMA QUESTION LOGIC ..................................................................................................................... 72
FIGURE 4. DECISION TREE TO DETERMINE WHEN STUDY PARTICIPANTS ARE PARTICIPATING AT HOME ............................................. 73
FIGURE 5. EXAMPLE OF ‘PARTICIPATION ’ TIME INTERVALS FOR A PARTICIPANT ...................................................................................... 74
FIGURE 6. FLOW OF PARTICIPANTS IN THE M2FED STUDY .......................................................................................................................... 75
FIGURE 7. DISTRIBUTION OF EMAS RECEIVED ACROSS THE TIME OF DAY (HOUR), BY EMA SURVEY TYPE ....................................... 76
FIGURE 8. DISTRIBUTION OF EMAS RECEIVED, BY FAMILY ROLE, BY EMA SURVEY TYPE ...................................................................... 77
FIGURE 9. AVERAGE (A) FAMILY-LEVEL AND (B) INDIVIDUAL-LEVEL COMPLIANCE ................................................................................ 78
FIGURE 10. OVERVIEW OF SAMPLING PROTOCOL FOR (A) TIME-TRIGGERED AND (B) EATING EVENT-TRIGGERED EMAS .......... 116

xii

ABSTRACT
This dissertation utilized data from the Monitoring and Modeling Family Eating
Dynamics (M2FED) study, which developed and deployed novel methods for in-home sensing
that could accurately monitor and model family eating dynamics using intensive longitudinal
data from Ecological Momentary Assesment (EMA) and wearable sensors. This dissertation
investigated the influence of intra-personal and inter-personal factors, and the synergy of these
factors, on both individual and family eating behavior in order to increase our understanding of
how children and families influence one another’s eating behaviors and to ultimately inform
future obesity prevention and intervention strategies. The broad aims of this dissertation were to:
(1) investigate the validity and feasibility of the M2FED cyberphysical system to detect in-field
family eating behavior; (2) identify and characterize “contextualized” latent classes of
participants’ eating events that were automatically detected in the home; and (3) investigate the
dynamics of in-home intra-personal events (affective states) and behavioral events (eating
events) within a family system over a two-week period of time using a relational event modeling
approach. Study 1 found that both individual- and family-level compliance rates to the M2FED
study’s EMA protocols were relatively high (both > the recommended 80%); and the wearable
device (i.e., wrist-worn smartwatch) that was used to automatically detect in-home eating events
performed on par with other in-field devices reported in the literature. Study 2 identified four
distinct classes of eating events – two “healthy” classes (characterized primarily by mindful
eating and satiety) and two “unhealthy” classes (characterized primarily by negative affect and
eating in the absence of hunger), the latter of which was significantly less likely to occur in
social contexts (i.e., when others were present). Lastly, Study 3 revealed that various types of
“inertia” effects, including affective state followed by affective state, eating event followed by

xiii

eating event, and affective state followed by eating event followed by affective state were
observed in the family systems. Overall, results from this dissertation demonstrate that EMA is a
feasible tool to collect ground-truth eating activity and thus evaluate the performance of
wearable sensors that automatically detect eating in the field. Furthermore, the results from this
dissertation provide insight into the ways in which intra-personal and inter-personal factors, and
the synergy of these factors, influence individual and family eating behavior in the wild; and
enable the identification of temporally specific processes (e.g., inertia) and events within the
family system that can be targeted for personalized, context-specific, real-time feedback that will
help to promote healthier eating behaviors and avoid unhealthy eating behaviors within families.

1
CHAPTER 1: INTRODUCTION
BACKGROUND AND SIGNIFICANCE
Overweight and Obesity
Prevalence and trends
In the past few decades, there has been a substantial upward trend of overweight and
obesity
1
prevalence in youth and adults, both in the United States (U.S.) and globally (Han,
Lawlor, & Kimm, 2010; Ng et al., 2014; Skinner, Ravanbakht, Skelton, Perrin, & Armstrong,
2018; The GBD 2015 Obesity Collaborators, 2017), thus making the global obesity epidemic a
major public health concern (World Health Organization, 2011).
Obesity prevalence for U.S. youth aged 2-19 years has nearly doubled in the past few
decades, increasing from 10.0% in 1988-1994 to 18.5% in 2015-2016 (Hales, Fryar, Carroll,
Freedman, & Ogden, 2018; Ogden, Carroll, Lawman, & et al., 2016). Notably, the prevalence
among children aged 2-5 years has increased from 7.2% in 1988-1994 to 13.9% in 2015-2016
(Hales et al., 2018; Ogden et al., 2016), equating to nearly 1 in 7 U.S. children aged 2-5 years old
being obese. Similarly, obesity prevalence for U.S. adults increased from 22.9% in 1988-1994 to
39.6% in 2015-2016 (Flegal, Carroll, Ogden, & Johnson, 2002; Hales et al., 2018). Although it
has been reported in some studies that obesity prevalence has decreased or leveled off for certain
age groups in recent years (e.g., children aged 2 to 5 years old and children aged 6 to 11 years

1
Obesity is typically measured with the use of Body Mass Index (BMI): a person’s weight in
kilograms divided by the square of height in meters squared. For children, obesity is defined as
having a BMI equal to or higher the 95
th
percentile for children and teens of the same age and
sex (these percentiles are calculated from data on children in the U.S. who participated in
national surveys that were conducted from 1963-65 to 1988-944) (Centers for Disease Control
and Prevention, 2018). For adults, a BMI of 30 or above is defined as obese (Centers for Disease
Control and Prevention, 2020).

2
old, respectively) (Ogden et al., 2016), this trend has not been consistently found in the literature
(Skinner et al., 2018).
Implications of overweight and obesity
Physical health. Children who are overweight and obese are at an increased risk of
remaining overweight into adulthood (Singh, Mulder, Twisk, van Mechelen, & Chinapaw,
2008); and a recent systematic review found that child and adolescent overweight or obesity
were associated with significantly increased risk of adverse physical health outcomes, including
diabetes, stroke, cardiovascular disease, and hypertension; and significantly increased risk of
premature mortality (Reilly & Kelly, 2011). Obesity has also been recognized as a risk factor for
a number of cancers, including colon cancer, breast cancer, kidney cancer, thyroid cancer,
pancreatic cancer, and liver cancer (Lauby-Secretan et al., 2016).
Weight stigma. Overweight and obese persons can experience weight-related stigma and
discrimination, in areas including employment, education, health care, and inter-personal
relationships (Puhl & Brownell, 2001; Puhl & Heuer, 2009; Puhl & Latner, 2007). These forms
of weight bias can have a variety of social effects, such as not being hired for a job or not
receiving a promotion, receiving compromised care from healthcare providers, social
marginalization, and bullying and teasing (Puhl & Heuer, 2009; Puhl & Latner, 2007).
The association between obesity and major health risks and weight stigma in childhood
as well as adulthood highlights the urgent need for effective overweight and obesity prevention
and intervention efforts in order to improve population-level health and quality of life.

3
Health disparities
There are reported differences in overweight and obesity rates by demographics,
including race/ethnicity and socio-economic status (SES).
Racial/ethnic disparities. Recent estimates from the National Health and Nutrition
Examination Survey (NHANES) show that, compared to non-Hispanic white U.S. youth
(14.1%), Hispanic (25.8%) and non-Hispanic Black (22.0%) U.S. youth had significantly higher
obesity prevalence in 2015-2016 (Hales, Carroll, Fryar, & Ogden, 2017). Similarly, Hispanic
(47.0%) and non-Hispanic Black (46.8%) U.S. adults had a higher obesity prevalence than non-
Hispanic white U.S. adults (37.9%) (Hales et al., 2017).
Socio-economic disparities. A recent study used two nationally representative health
surveys to examine the changes in obesity among U.S. adolescents by SES: results indicate that,
since 2003-2004, obesity prevalence among low-SES adolescents has increased while prevalence
among high-SES adolescents has decreased (Frederick, Snellman, & Putnam, 2014). For U.S.
adults, obesity prevalence varies by both income and education level. Obesity prevalence was
lowest for those in the highest-income group (31.2%), compared to those in the middle- and
lowest-income groups (40.8% and 39.0%, respectively) (Ogden et al., 2017). U.S. adults with a
college degree also had the lowest obesity prevalence (27.8%) compared to those who had
completed some college (40.6%) and those who were high school graduates or less (40.0%)
(Ogden et al., 2017).
Proposed mechanisms. Several mechanisms have been proposed in the literature to
partially or fully explain these racial/ethnic and socio-economic health disparities. Krueger &
Reither propose that, primarily, the mechanisms fall into three broad categories: health behaviors
(e.g., diet, physical activity), biological factors (e.g., genetics, epigenetics), and environment

4
(e.g., neighborhood contexts, social networks) (Krueger & Reither, 2015). While early research
primarily focused on individual-level demographic characteristics (e.g., age, gender) (Zhang &
Wang, 2004) and behavioral factors (e.g., amount of television watching) (Singh, Kogan, Van
Dyck, & Siahpush, 2008) to explain these differences, a number of recent studies have examined
physical and social environmental factors and found that they also explain these disparities
(Johnson & Johnson, 2015; LaVeist, Pollack, Thorpe, Fesahazion, & Gaskin, 2011; Williams &
Jackson, 2005). One observational study conducted with a large sample of Massachusetts
children found that neighborhood SES and food and physical activity environments partially
attenuated the differences found in BMI z-scores between white and Hispanic children and
between white and Black children (Sharifi et al., 2016). Bleich et al. proposed that the difference
in obesity prevalence between Black and white women (which has been established as a
consistent disparity in nationally-representative datasets) (Flegal, Carroll, Kuczmarski, &
Johnson, 1998; Ogden et al., 2006) would be reduced once socio-environmental exposures were
considered; and their study found no discernable racial disparities between Black and white
women with similar SES who shared the same social context
2
(Bleich et al., 2010). This finding
builds on the body of research that establishes the role of the social and built environment on
obesity prevalence.
Past childhood obesity prevention and treatment efforts
Despite the known complexity of modifiable risk behaviors (e.g., diet, physical activity)
and obesity etiology, the majority of childhood obesity interventions are implemented at the

2
The definition of social context refers to residential segregation; therefore, women in this study
who shared the same ‘social context’ were living in racially integrated communities (Bleich,
Thorpe, Sharif-Harris, Fesahazion, & LaVeist, 2010).

5
individual level (i.e., target individual’s behavioral risk factors) (Al‐Khudairy et al., 2017; T.
Brown et al., 2019; Colquitt et al., 2016; Mead et al., 2017). To date, the impacts of interventions
to prevent and treat pediatric obesity have been modest at best. Interventions may obtain some
changes in behavior, but for the most part fail to influence levels of adiposity (T. Brown et al.,
2019; Oude Luttikhuis et al., 2009; Waters et al., 2011).
Recent evidence has shown that family systems can play a key role in dietary behavioral
interventions (Wilfley, Vannucci, & White, 2010). A family history of obesity strongly predicts
childhood and adult obesity (also referred to as the “intergenerational transmission of obesity”)
due to family members’ shared genetic, environmental, and behavioral risks (Kral & Rauh,
2010). A review found evidence for the familial transmission of certain shared behavioral risks
(i.e., taste preferences, food selections, and eating behaviors), highlighting the importance of
family units as targets for intervention and prevention (Kral & Rauh, 2010).
Family-based interventions that target obesogenic behaviors are critical because they
focus on early prevention in children and have the potential to halt the intergenerational
transmission of obesity. Early prevention is key because overweight and obese children are at
greater risk of being overweight or obese in adulthood (Dattilo et al., 2012), and treating adults
for obesity has yielded limited success (Dombrowski, Knittle, Avenell, Araújo-Soares, &
Sniehotta, 2014).
Family-based interventions have the added benefit of targeting clusters of at-risk
individuals and may also result in parental weight loss (Best et al., 2016; Theim et al., 2012); and
they have been shown to successfully lead to both short- and long-term weight loss in children
(Berge & Everts, 2011; Epstein, Valoski, Wing, & McCurley, 1994; Wilfley et al., 2007).
However, the most recent Cochrane Reviews on interventions to treat and prevent obesity in

6
youth, respectively, show that fewer than 25% of interventions are family-based or include the
family (Oude Luttikhuis et al., 2009) and fewer than 10% incorporate the home as an
intervention setting (T. Brown et al., 2019).
Gaps in the literature
Despite decades of research on childhood obesity prevention and intervention strategies,
the overall prevalence of obesity has continued to rise in both U.S. children and adults. Strategies
that target individual-level behavior have for the most part failed to change one’s long-term level
of adiposity. Evidence from empirical research and systematic reviews (Berge & Everts, 2011;
Sung-Chan, Sung, Zhao, & Brownson, 2013) indicates that family-based interventions are
effective childhood obesity treatments, yet they have not been comprehensively incorporated into
a majority of childhood obesity interventions used to date (T. Brown et al., 2019; Oude
Luttikhuis et al., 2009), nor have they been made sufficiently accessible to the families who need
them.
Eating Behavior as a Primary Risk Factor for Overweight/Obesity
Etiology of overweight and obesity
Traditionally, physical activity, diet, and metabolic factors have been proposed as the
primary factors in body weight management (Weinsier, Hunter, Heini, Goran, & Sell, 1998). The
energy balance equation (Spiegelman & Flier, 2001) is a popular framework used in obesity
prevention and intervention initiatives (Wells & Siervo, 2011). One side of the balance equation
contains energy intake (i.e., eating), while the other side contains energy expenditure (i.e.,
physical activity, basal metabolism, and adaptive thermogenesis) (Spiegelman & Flier, 2001).
According to this equation, obesity will develop only if energy intake exceeds total body

7
expenditure (Spiegelman & Flier, 2001). The majority of obesity prevention and intervention
efforts have been focused on changing an individual’s diet and physical activity behavior, as
those are two of the most important human behaviors that impact body mass index (BMI) and
central adiposity (Spruijt-Metz, 2011). Moreover, intra-personal factors such as an individual’s
knowledge, attitudes, skills, physiology, etc. have been popularly hypothesized as determinants
of obesogenic behaviors (e.g., diet, physical activity), and subsequently, have been proposed as
primary points of intervention (Pate et al., 2013; Thompson, Vamos, & Daley, 2017).
More recently, the etiology of obesity has been recognized as complex, multi-level, and
multi-factorial rather than a simple energy balance problem. Social, political, economic, and
physical environments (and their interactions) undoubtedly play a role in energy balance; many
conceptual models have been proposed to expand the traditional equation, such as an ecological
model (Egger & Swinburn, 1997), a systems-oriented, multi-level model (Huang, Drewnosksi,
Kumanyika, & Glass, 2009), and a complex systems model (Hammond, 2009; Lee et al., 2017).
These models posit that the determinants of obesity are both complex and potentially non-linear,
and therefore provide the opportunity to investigate factors that are at multiple levels (e.g., intra-
personal, inter-personal, community, etc.) and their interdependences.
Dietary intake and health
The field of nutritional epidemiology
3
has produced an abundance of studies that have
examined the role of dietary intake (i.e., what and how much is consumed) in human health and
disease – looking specifically at macronutrients (e.g., fats, carbohydrates), types of food, quality
of food, dietary patterns, and more (Willett, 2013). Decades of research indicate that dietary

3
Nutritional epidemiology is the study of the influence of diet on the occurrence of human
disease.

8
intake is a critical component of chronic disease prevention (Nishida, Uauy, Kumanyika, &
Shetty, 2004). However, the measurement of diet in free-living populations remains a huge
challenge in the field, a topic that will be discussed at greater length at the end of this chapter.
But, even if public health researchers could easily and accurately track dietary intake,
dietary intake patterns are notoriously difficult to change long-term (Wood & Neal, 2016).
Eating occurs frequently throughout the day, and often in stable physical and social contexts
(Khare & Inman, 2006; Liu, Han, & Cohen, 2015). It is strongly shaped by cues and influences
in people’s physical and social environments, thereby making it susceptible to habit formation
and difficult to change (Wood & Neal, 2016) if only focusing on the intra-personal determinants
(e.g., an individual’s knowledge, attitudes, etc.).
Shifting focus: from diet to eating behavior and context
Eating behaviors and patterns (i.e., food choices and motives, feeding practices), and
context (i.e., who is eating, when, where, with whom, etc.) also play a significant role in the
development of obesity and other chronic diseases, including type 2 diabetes and heart disease
(Higgs & Thomas, 2016; Jannasch, Kröger, & Schulze, 2017; Neuhouser, 2018; Reicks et al.,
2015; Robinson, Thomas, Aveyard, & Higgs, 2014; Tourlouki, Matalas, & Panagiotakos, 2009).
For example, research has shown social contexts like eating with family (Neumark-Sztainer,
Wall, Story, & Fulkerson, 2004; Suggs, Della Bella, Rangelov, & Marques-Vidal, 2018; Videon
& Manning, 2003), physical contexts like living in a rural food environment (Lenardson,
Hansen, & Hartley, 2015), psychological contexts like chronic stress levels (Isasi et al., 2015;
Torres & Nowson, 2007), and timing of eating (Wilkinson et al., 2020) have important impacts
on eating patterns and health outcomes.

9
However, our current health behavioral theories do not often take into account an
individual’s biological, social, personal, and environmental states or contexts (Spruijt-Metz et
al., 2015). Our current understanding of eating behavior is largely based on static snapshots of
dietary intake and on linear relationships between behavior change and its predictors. Rather,
people behave within a much larger, complex system, in which their behavior is constantly
evolving across time and contexts (Riley et al., 2011; Spruijt-Metz et al., 2015). New
technologies present an opportunity to collect intensive longitudinal data from mobile phones,
wearable devices, and in-situ sensors in order to inform the development of more dynamic eating
behavior theories (Riley et al., 2011; Spruijt-Metz et al., 2015). These dynamic theories and
models have the potential to explain how the influence processes on eating behavior interact and
unfold over time.
Gaps in the literature
A person’s dietary intake (i.e. what and how much they consume) is both notoriously
hard to measure and notoriously hard to change long-term. Emerging literature indicates that
eating behavior patterns and contexts also play a significant role in obesity and chronic disease
development, but the field is still lacking appropriate behavioral theories that provide a richer
understanding of how eating behaviors vary across contexts and across time.
Influences on Eating Behavior
Complex systems framework
The influences on eating behavior, and consequently, the drivers of obesity, are complex.
Contrary to widespread belief that obesity is a result on one’s lack of willpower and individual
choices, obesity has many complex factors and interdependences. This oversimplification of the

10
problem does not reflect the current state of scientific evidence. Whereas traditional health
behavior models, such as the Health Belief Model (Champion & Skinner, 2008) and the Theory
of Planned Behavior (Ajzen, 1991), suggest that an individual’s behaviors are a result of one’s
thoughts, beliefs, and intentions, more recent models suggest that a much wider variety of factors
influence an individual’s dietary intake and eating behavior.
An increasingly popular conceptualization of the influences on eating behavior and
drivers of obesity is the “Obesity System Map” developed in a report commissioned by the
United Kingdom’s Government Office for Science (Vandenbroeck, Goossens, & Clemens,
2007). This map depicts relevant factors and their interdependences that determine the condition
of obesity for an individual or a group of people. It consists of 108 variables and 304 causal
linkages, which are represented in seven clusters: individual physiology, individual physical
activity, physical activity environment, individual psychology, social psychology, food
consumption, and food production.
Recognizing the obesity epidemic as a complex systems problem necessarily shifts the
paradigm of how public health researchers and professionals approach obesity prevention and
intervention strategies. A complex system can be defined as “one whose properties are not fully
explained by an understanding of its component parts” (p. 79) (Gallagher, Appenzeller, Normile,
Service, & et al., 1999), and have the following properties (as defined by Luke & Stamatakis, p.
358):
i. They are made up of a large number of heterogeneous elements;
ii. These elements interact with each other;
iii. The interactions produce an emergent effect that is different from the effects of the
individual elements;

11
iv. This effect persists over time and adapts to changing circumstances (Luke & Stamatakis,
2012).
Social-ecological and systems science models of health increasingly recognize that a broad
range of interrelated factors impact health by influencing individuals to adopt particular
behaviors. Individual processing of these stimuli and subsequent engagement in behaviors is
therefore central to this system. Identifying and understanding these stimuli that influence eating
behaviors within systems is critical in obesity prevention efforts.
One framework often used to portray the many influences on eating behavior is an
ecological framework. Story et al. present this framework by illustrating the individual factors
(e.g., cognitions, demographics), social environments (e.g., family, friends, social support),
physical environments (e.g., home, school, neighborhoods, food outlets), and macro-level
environments (e.g., economic systems, food marketing and media) that influence what people eat
(Story, Kaphingst, Robinson-O'Brien, & Glanz, 2008).
4

A substantial amount of work has been done to identify and characterize the ways in
which the physical environment influences eating behavior (e.g., food accessibility, food
advertisements, toxic food environment). The characterization of how one’s social environment
influences eating behavior, and moreover, the interaction of one’s individual-level factors and
one’s social environment, has been less of a focus until recently.
Individual influences (Intra-personal)
A number of intra-personal factors (e.g., attitudes, behavior, internal states) have been
shown to strongly influence eating behavior, including: eating in the absence of hunger (Fisher &

4
Physical environmental influences and macro-level influences on eating behavior will not be
exhaustively reviewed in this chapter because they are beyond the scope of this dissertation.

12
Birch, 2002), mood, anger, and stress during eating (Lemmens, Rutters, Born, & Westerterp-
Plantenga, 2011), and the effect of mood, interactions, and other environmental and personal
variables on speed of eating (Volkow, Wang, & Baler, 2011). Many of these intra-personal
contextual states, such as hunger levels, eating in the absence of hunger, and mindful eating,
have been established as associates of eating behavior in cross-sectional studies (Beshara,
Hutchinson, & Wilson, 2013; Fogel et al., 2018; Gilbert & Waltz, 2010; Lansigan, Emond, &
Gilbert-Diamond, 2015; Lavender, Gratz, & Tull, 2011), but the temporal relationship between
these contextual states and eating behavior in the moment are underexplored.
Stress and diet. Decades of research have examined the relationship between stress and
diet (Ganley, 1989; Greeno & Wing, 1994; Sominsky & Spencer, 2014; Stone & Brownell,
1994), however the underlying mechanisms are still unclear to date. Mixed results in the
literature (Dunton et al., 2016; Groesz et al., 2012; Hou et al., 2013) indicate that stress can
either lead to undereating or overeating, possibly varied by stressor severity (e.g., mild vs.
severe, etc.) or stressor duration (e.g., acute vs. chronic) (Torres & Nowson, 2007). Epel and
colleagues (2001) showed that women who exhibited high stress reactivity after being exposed to
an acute stressor (high salivary cortisol output) had increased caloric intake (Epel, Lapidus,
McEwen, & Brownell, 2001). Chronic stress may cue some people to eat highly palatable,
energy-dense foods regardless of hunger (Peters, Kubera, Hubold, & Langemann, 2011). Those
who experience chronic stress have been found to eat more after acute stress events (Yau &
Potenza, 2013), and prefer foods higher in sugar and fat (Yau & Potenza, 2013). However, the
majority of research examining the role of stress on dietary intake has been conducted in adult
populations (Laitinen, Ek, & Sovio, 2002; Torres & Nowson, 2007). Late childhood and
adolescence are known to be a period of increased intense and frequent stress due to

13
physiological, mental, and emotional growth from puberty, heightened school and peer pressure,
and newfound independence (Casey et al., 2010). This period also coincides with a time of
growing autonomy over food choices and exposure to unhealthy food options (Contento,
Williams, Michela, & Franklin, 2006). Therefore, results on stress and unhealthy eating
behaviors from adult populations may not be directly generalizable to youth.
A handful of studies have used Ecological Momentary Assessment (EMA), a data
collection technique in which one’s behavior is repeatedly sampled in the natural environment
(Shiffman, Stone, & Hufford, 2008), to assess the relationship between momentary intra-personal
context (e.g., negative affect, stress) and eating behavior (Engel et al., 2016). Studies have found
that momentary anger, negative affect, and stress are significantly associated with binge eating
episodes in women with eating disorders (Anestis et al., 2010; Engel et al., 2007; Mason et al.,
2018); and fruit and vegetable consumption in the preceding two hours is associated with
happiness in women (Liao et al., 2018). Many of these studies have been conducted in adult
women, typically with an eating disorder; the association between momentary intra-personal
context and eating behavior has yet to be explored in varying populations.
Social environmental influences (Inter-personal)
The study of social influences on eating behavior has rapidly expanded in the past
decade. The field of social psychology suggests that one’s social environment (i.e., the presence
of others) influences one’s thoughts, feelings, and behaviors, including eating behavior. The
evidence reviewed above clearly illustrates that eating is influenced by factors external to the
individual, and social influences may be especially impactful because:
Eating is a social practice. An analysis of the 2006–2008 American Time Use Survey,
which uses a large representative sample of American adults, indicates that the majority of eating

14
and drinking episodes take place with others (i.e., not alone) (Oh, Erinosho, Dunton, M Perna, &
Berrigan, 2014). Eating is an inherently social practice (Delormier, Frohlich, & Potvin, 2009) –
shared meals have been and continue to be the centerpieces of many traditions, celebrations, and
other cultural occasions around the world; but they also occur on a daily basis in places where
we spend a large amount of time such as school and work.
Eating is a habit. Eating behavior is strongly shaped by cues and influences in people’s
social environments. Eating occurs frequently throughout the day, and often in stable social
contexts, resulting in ‘decisions’ to engage in specific eating behaviors being outsourced to cues
in the environment, including social cues (others eating) and physical cues (i.e., food availability,
portion size), thereby making eating behavior susceptible to habit formation (Khare & Inman,
2006; Liu et al., 2015; Verhoeven Aukje, Adriaanse Marieke, Evers, & de Ridder Denise, 2012;
Wood & Rünger, 2016). Social cues can have a strong impact on food intake through several
mechanisms, including mindless imitation of others’ eating behaviors (Bell et al., 2019; Hermans
et al., 2012; Sharps et al., 2015), food matching and modeling (Salvy, de la Haye, Bowker, &
Hermans, 2012), and situational and cultural food norms (Herman & Polivy, 2005).
Modeling of food intake and choice. ‘Modeling’ of food intake refers to the use of others’
eating as a guide for what and how much to eat (Cruwys, Bevelander, & Hermans, 2015). A
multitude of both observational and experimental studies have shown significant modeling
effects with varying populations, foods types, and settings (Cruwys et al., 2015; Herman, Polivy,
Pliner, & Vartanian, 2019; Vartanian, Spanos, Herman, & Polivy, 2015). In the majority of these
studies, participants ate more food when their companion ate more, and ate less food when their
companion ate less (Cruwys et al., 2015; Vartanian et al., 2015); and people were typically
unaware of their modeling behavior.

15
Fewer studies have looked at the modeling of food choice. However, those that have find
similar effects, in which participants are more likely to choose the same type of food as their
companion, although these findings tend not to be as strong as those of food intake modeling
(Cruwys et al., 2015; Herman et al., 2019).
Social facilitation. Relatedly, eating with others leads us to eat more than we normally
would have without others – and this effect becomes stronger as the number of people we eat
with grows (Herman et al., 2019). In one early study, de Castro & de Castro found that meals
eaten with others were 44% larger than meals eaten alone, and this effect was independent of
pre-meal self-reported hunger and the amount of time since the last eating occasion (de Castro,
1995; de Castro & de Castro, 1989). A recent review has found that, in research studies spanning
three decades, this social facilitation phenomenon remains a consistent finding, although the
mechanisms are still unclear (Herman, 2015).
Social norms. Informational eating norms have been proposed as a specific process that
influences one’s eating behavior (Higgs, 2015; Higgs & Thomas, 2016; Robinson et al., 2014).
Informational social norms refer to information about the eating habits of other people (Robinson
et al., 2014). The findings of multiple review papers indicate that social norms influence food
intake and choices, and that there’s a potential to harness social norms to promote healthier
eating habits (Higgs, 2015; Higgs & Thomas, 2016; Robinson et al., 2014).
Family influences
Strong evidence suggests that family networks and systems impact eating behaviors and
obesity risk. The social influence mechanisms described above—i.e., modeling, social
facilitation, and social norms—have also been found to operate among family members. For

16
instance, family members tend to engage in similar behaviors, such as in the areas of food choice
(Ayala et al., 2007; Cameron et al., 2011; Pachucki, Jacques, & Christakis, 2011) and eating
style (e.g., restrained eating) (Munsch et al., 2007). Additionally, one early study not only found
evidence of social facilitation (i.e., meals eaten with others were larger and longer in duration
compared to meals eaten alone), but also found that meals eaten with family were larger and
faster compared to meals with other types of companions (de Castro, 1994).
Parental influence. A large portion of the family influence literature focuses on how
parental behaviors, attitudes, and feeding styles influence child dietary intake. Modeling of
family members’ eating behaviors (Yee, Lwin, & Ho, 2017), home food environment and family
meal practices (e.g., lack of family rules, eat dinner together, eat at least one vegetable) (Lytle et
al, 2011), and parental feeding styles (e.g., restrictive feeding practices, authoritative feeding
styles) are all shown to impact child dietary intake (Savage, Fisher, & Birch, 2007; Vollmer &
Mobley, 2013). Notably, however, few studies have examined the influence in the other direction
(i.e., child’s influence on parent) and/or the bidirectional influences between parents’ diet and
children’s diet intake.
Family features and structure. Other related work has sought to identify the influence of
family features and family structures on various eating outcomes. Patrick & Nicklas (2005)
reviewed the literature on family and social determinants of children’s eating patterns and diet
quality, and they found that parents’ level of education, time constraints, and ethnicity influence
the types of foods a child eats (Patrick & Nicklas, 2005). Pearson et. al. found that adolescents
from dual-parent families ate breakfast on more days per week than adolescents from single-
parent families (Pearson, Atkin, Biddle, Gorely, & Edwardson, 2010).

17
Family meal features. Family meals are a common context for food consumption
(Dallacker, Hertwig, & Mata, 2018; Neumark-Sztainer, Hannan, Story, Croll, & Perry, 2003),
although the frequency of family meals can vary by demographic variables such as child gender,
grade level, and socio-economic status (Neumark-Sztainer, Wall, Fulkerson, & Larson, 2013).
Evidence suggests that greater family meal frequency is associated with better diet quality and
lower body mass index in children (Dallacker et al., 2018), however the mechanisms that explain
this association are still unclear. A recent meta-analysis identified various features of family
meals—e.g., turning off the television during a meal, the child’s involvement with meal
preparation, and a longer meal duration—that are associated with better nutritional health
outcomes in children (Dallacker, Hertwig, & Mata, 2019). But, these features are not well
understood, and what gives rise to “ideal” family meal contexts has not been studied.
Parental/family stress. Parental and/or family stress has been identified as a potential
explanatory variable for certain eating behavior patterns in both children and adults. For
instance, cross-sectional studies have found that higher parent-perceived stress is associated with
more fast-food consumption in children (Parks et al., 2012); and higher work-life stress among
both parents is associated with fewer family meals and with more sugar-sweetened beverage and
fast food consumption by the parents (Bauer, Hearst, Escoto, Berge, & Neumark-Sztainer, 2012).
A large portion of this literature focuses on examining the influence of maternal stress on
her child’s diet. A systematic review reported mixed findings on this relationship between
maternal stress and child dietary intake (O'Connor et al., 2017). The results from one
longitudinal study indicate that mothers’ greater-than-usual perceived stress is associated with
mothers’ healthier eating that day, but is not associated with child’s eating behavior (Dunton et

18
al., 2016). In another longitudinal study, perceived parent stress was not significantly associated
with child consumption of added sugars one year later (Shonkoff et al., 2017).
Recent studies have utilized Ecological Momentary Assessment (EMA) to measure
momentary affect and food intake in the natural environment. One study using EMA found that
mothers and children who reported higher-than-average negative affect consumed more
pastries/sweets; children with higher-than-average negative affect consumed more fast food; and
mothers’ momentary positive affect predicted their own fruit/vegetable consumption (Mason et
al., 2019). Another EMA study found that higher parental stress/depressed mood earlier in the
day predicted fewer homemade foods served at dinner that same night (Berge et al., 2017).
The potential interdependent relationship of stress and obesogenic behaviors between two
family members (typically mother and child) has begun to be explored for eating behavior
(Mason et al., 2019) and physical activity behavior (Yang et al., 2020). However, studies have
not yet explored the independences between more than two family members, nor the possible
dynamic effects of stress on subsequent eating behavior in families.
Family interventions. Parental eating behaviors and feeding practices are an important
feature of children’s eating environments that have immense and lasting influence on diet and
obesity risk in children, suggesting early intervention efforts involving parents and caretakers
may help prevent excess weight gain during childhood (Anzman, Rollins, & Birch, 2010).
Furthermore, as reviewed above, many features of the family system can be barriers or promoters
of healthy eating behaviors. These studies highlight the potential to harness family influence to
reduce obesity risk.
Although the influence of family context and characteristics on childhood obesity,
obesity-related behaviors, and treatment has been recognized for decades (Epstein, Paluch,

19
Roemmich, & Beecher, 2007), the most recent Cochrane Reviews on interventions to treat and
prevent obesity in youth, respectively, show that fewer than 25% of interventions are family-
based or include the family (Oude Luttikhuis et al., 2009) and fewer than 10% incorporate the
home as an intervention setting (T. Brown et al., 2019).
To date, no studies have attempted to monitor or model the family dynamics around
eating in a socio-physical context and in real-time.
Gaps in the literature
Although the field has seen an marked increase in the study of social influences on eating
behavior, many of these studies have notable limitations. Some potentially important features of
the family system have not yet been explored, including multi-level features of family members,
their relationships amongst one another, and the family system as a whole. Relatedly, rarely have
family features been studied together in order to understand their relationship to one another and
to eating behavior. Lastly, much of this research has taken place in laboratory settings (vs. in the
field settings), and has relied on cross-sectional survey data and/or in-lab measurement periods
only spanning a few hours. These limitations prohibit the exploration of the rich dynamics that
may play out over days and weeks as multiple eating events unfold over time.
Dietary Assessment Methodology
To facilitate the study of eating behavior and it’s influences in real-time and in context,
appropriate measurement tools are needed. However, the field of dietary assessment has a long

20
history that has been marked by both controversies and advances.
5
An overview of the field’s
history, controversies, and advances are reviewed below.
Overview of current dietary assessment tools
One persisting challenge to dietary assessment and eating behavior research is the ability
to accurately measure dietary intake. Historically, the assessment of dietary intake and eating
behaviors has utilized self-reporting tools (Shim, Oh, & Kim, 2014; Thompson, Subar, Loria,
Reedy, & Baranowski, 2010). Three commonly used self-report tools are (i) food records (food
diaries), (ii) food frequency questionnaires (FFQ), and (iii) 24-hour dietary recalls (Magarey et
al., 2011; Thompson & Subar, 2017; Willett, 1998).
Food records. The purpose of food records is to measure the types and amounts of food
and beverages consumed over one or more days. Respondents are usually instructed to self-
report this information in their records as close to the time of consumption as possible.
Food frequency questionnaires. The purpose of the food frequency approach is to
measure how often a certain type or types of foods are eaten over a specific period of time (e.g.,
in the past month, in the past year, etc.). In some cases, these data can then be used to estimate
overall nutrient intake.
24-hour dietary recalls. The purpose of the dietary recall is to estimate all the foods and
beverages that were consumed during the preceding 24-hour period. Respondents are asked to
self-report the types and amount of food and beverage they consumed, typically conducted in a
structured interview, although web-based automated self-administered recalls are now available.

5
See: Advances and Controversies in Diet and Physical Activity Measurement and Youth
(Spruijt-Metz et al., 2018).

21
The method of using doubly labelled water (resulting in a measurement of energy
expenditure) is often used to validate these self-reported dietary assessment methods. This
method has been established as the reference standard for validating measurements of energy
intake (Buchowski, 2014), but is limited in scope as it only measures energy expenditure (vs.
other eating outcomes such as intake of particular nutrients or food groups, timing of dietary
intake, etc.).
Methodological limitations and advantages of current dietary assessment tools
All dietary assessment self-report methods have some level of measurement error (the
difference between the measured value and the true value) (Beaton, Burema, & Ritenbaugh,
1997; Thompson et al., 2015). A number of limitations can plague self-report dietary assessment
methods for both children/adolescents and adults (Livingstone, Robson, & Wallace, 2004;
Westerterp & Goris, 2002), such as:
• social desirability bias (caused by a respondent’s social desirability or approval), which
may result in underreported dietary intake (Hebert, Clemow, Pbert, Ockene, & Ockene,
1995).
• recall/memory bias (when a respondent erroneously recalls a past event or behavior),
which may result in under- or over-reported dietary intake (Althubaiti, 2016).
6

• risk of respondent burden (when participation is difficult, time consuming, or emotionally
stressful) ("Encyclopedia of Survey Research Methods," 2008), which may result in
underreporting of dietary intake (Willett, 1998).

6
Notably, one study examining caloric intake data from the National Health and Nutrition
Examination Survey (NHANES) (data collection waves ranging from 1971 to 2010) found that
energy intake data on the majority of respondents (67.3% of women and 58.7% of men) were not
physiologically plausible (Archer, Hand, & Blair, 2013).

22
• reactivity (when a respondent changes their dietary intake as a result of the recording
process itself, typically to lessen the burden on recording their food intake later) (Rebro,
Patterson, Kristal, & Cheney, 1998).
• the extent to which certain types of individuals underreport their energy intake (e.g.,
obese individuals, individuals with body image dissatisfaction, adolescents) (Forrestal,
2011; Novotny et al., 2003; Wehling & Lusher, 2017).
• the extent to which certain types of foods are underreported (Gemming & Ni Mhurchu,
2016).
As a result, the validity of data collected via self-report methods and subsequent research
findings can come under question. As indicated above, these biases may result in the under- and
over-reporting of dietary intake, thereby giving an inaccurate picture of dietary intake patterns,
potentially skewing research findings, and providing inadequate scientific conclusions (Archer &
Blair, 2015; Schoeller, 1995; Schoeller et al., 2013).
It has been argued that, despite the limitations, these traditional tools do have some
methodological advantages, such as the potential to provide detailed information on dietary
intake and portion size, and the potential ability to be easily administered (Satija, Yu, Willett, &
Hu, 2015).
Summary of technology-assisted dietary assessment tools
Emerging technologies present the opportunity to improve these current assessment
methods by using innovations that improve the quality and validity of data that are collected, and
by passively measuring eating activity in naturalistic settings over long periods of time with
minimal user interaction. Technological advances in dietary assessment tools include: (i) web-
based self-administered 24-hour recall tool, (ii) mobile device-assisted ecological momentary

23
assessment (mEMA), (iii) image-assisted and image-based dietary assessments, and (iv)
wearable devices/sensors (Bell et al., 2020; Spruijt-Metz et al., 2018).
Web-based self-administered 24-hour recall. Developed by the National Cancer Institute,
the Automated Self-Administered 24-Hour Recall is a web-based tool that guides a respondent
through an online 24-hour recall in which they report on eating occasions, time of consumption,
food preparation, portion size, and additions (Subar et al., 2012). A user-friendly version for
youth has been recently adapted and validated (Diep et al., 2015).
Mobile device-assisted ecological momentary assessment (mEMA) is a data collection
technique in which one’s behavior is repeatedly sampled in real-time and in context (Shiffman et
al., 2008), and it has been suggested as a promising tool for dietary assessment (Engel et al.,
2016). Studies have used mEMA to measure a variety of dietary outcomes, including frequency
of food intake, intake of specific types of foods (e.g., low glycemic index foods), and energy
intake (Schembre et al., 2018).
Image-assisted dietary assessments utilize images taken with handheld devices or
wearable cameras to supplement traditional dietary assessments approaches (e.g., dietary records,
24-hour dietary recalls) by aiding in the portion size estimation process (Boushey, Spoden, Zhu,
Delp, & Kerr, 2017). Their primary purpose is supplementary, and they are intended to be used
in conjunction with other data sources on dietary intake. Conversely, image-based dietary
assessments are a primary assessment approach, and utilize images to capture all eating
occasions; the image review process can either be completed by a human-trained analyst or an
automated method (Boushey et al., 2017).
Wearable devices with embedded sensors allow for the passive collection of various data
streams (e.g., acoustic, visual, inertial, etc.). The feasibility of using sensors to measure dietary

24
intake and patterns has been tested in both lab and field settings (Bell et al., 2020; Hassannejad et
al., 2017; Heydarian, Adam, Burrows, Collins, & Rollo, 2019; Vu, Lin, Alshurafa, & Xu, 2017).
Studies have used a variety of sensors (e.g., microphones, cameras, smartwatches, EMG
electrodes, etc.) to measure a variety of dietary outcomes, including bites, chewing, swallowing,
and duration of eating occasions (Bell et al., 2020; Boushey et al., 2017; Hassannejad et al.,
2017; Heydarian et al., 2019; Vu et al., 2017).
Combinations of tools. Some studies have reported using multiple types of technology-
assisted dietary assessment tools. For example, in Ye et al., when an eating gesture was
automatically detected via wrist-worn sensor, participants were sent a short message on their
smartwatch to confirm or reject in real-time whether they were eating (Ye, Chen, Gao, Wang, &
Cao, 2016). Similarly, in Gomes & Sousa, when drinking activity was detected via wearable
sensor, participants were sent an alert on their smartphone and could then confirm or reject
whether they were drinking via EMA (Gomes & Sousa, 2019).
Methodological limitations and advantages of technology-assisted dietary assessment tools
These methods are not without limitations. To some extent, they are still susceptible to
the same limitations as the traditional methods including the self-report biases and participant
burden reviewed above (Spruijt-Metz et al., 2018). Additionally, these technology-assisted
dietary assessment tools may experience unique challenges, such as technology usability and
possible malfunctions, cost, suitability for low-literacy populations, and decreased compliance
and attrition rates due to respondent burden (McClung et al., 2018; Spruijt-Metz et al., 2018).
However, technology-assisted dietary assessment tools also possess the following unique
advantages over traditional self-report tools: timing of assessment, location of assessment,
frequency of assessment, levels of assessment, and granularity of assessment.

25
Timing. These novel assessment methods are able to measure behavior near- or just-in-
time, thereby reducing or eliminating the recall bias that plagues retrospective self-report
measures.
Location. Collecting data in the context of daily life, where behavior actually occurs, has
proven to be extremely difficult and burdensome (McClung et al., 2018; Spruijt-Metz et al.,
2018). Consequently, data collected on eating activity are typically collected in a laboratory
setting, in some type of observation setting (e.g., in a cafeteria, etc.), or out of context (e.g.,
recounting at the end of the day what was eaten for breakfast that morning). These technologies
allow for the collection of data not only when the behavior occurs, as described above, but also
where the behavior occurs and can thus improve the ecological validity of research findings
(Shiffman et al., 2008).
Frequency. Frequent measurements of the same variable allow for the possibility to
examine how predictors of behavior and behaviors themselves vary over time.
System-level context. A novel feature of technological tools is their potential ability to
passively measure potentially important contextual features of eating behavior, such as physical
context (e.g., eating at home vs. at work), social context (e.g., social interaction), and more
(Gemming, Doherty, Utter, Shields, & Ni Mhurchu, 2015a; Spruijt-Metz, de la Haye, Lach, &
Stankovic, 2016). However, many of these tools are nascent, and further research is needed to
develop and improve these tools.
Granularity. Traditional methods of eating assessment typically summarize dietary
measures at the day-, week-, or even year-level (Shim et al., 2014; Willett, 1998). Although these
can be helpful in understanding relationships between eating behavior and its predictors,
important micro-level temporal patterns and processes are not measured nor can they be explored

26
with these measures. Technology-assisted dietary assessment tools offer the ability to explore
micro-level eating activities, such as meal microstructure (the dynamic process of eating,
including meal duration, changes in eating rate, chewing frequency, etc.) (Doulah et al., 2017),
food choices (Marcum, Goldring, McBride, & Persky, 2018), and processes (e.g., eating rate
(Ohkuma et al., 2015); eating mimicry (Bell et al., 2019; Sharps et al., 2015); etc.), which is
important because recent literature suggests that they may play an important role on food
selection, dietary intake, and ultimately, obesity and disease risk.
Gaps in the literature
Accurate assessment of diet and eating behaviors has long been a challenge in the field of
obesity research methods. In the past decade, technology-assisted dietary assessment tools,
including wearable devices and mobile device-assisted ecological momentary assessment, have
emerged as a possible solution to objectively and automatically detect human eating activity.
However, the performance of these technological tools needs to be evaluated in the field (vs. in
laboratory settings) with diverse populations before they can be used to understand eating
behavior and context, and used to develop dynamic health behavior theories and models.
SUMMARY OF GAPS IN THE LITERATURE
Family eating dynamics – who is eating, when, where, with whom – have yet to be
measured and modeled dynamically, to better contextualize our understanding of social influence
processes within family systems. This is due in part to the fact that accurate diet monitoring is
difficult and typically relies on self-reported measures (e.g., 24-hour recalls or Food Frequency
Questionnaires), all of which have significant accuracy (validity) and precision (reliability)

27
limitations that make dietary behavior modeling difficult. Furthermore, these methods are not
very well suited to measure:
i. habitual behaviors such as eating because they rely on people being aware of, and then
reporting on, their behaviors;
ii. important temporal processes and interactions among family members in context and in
real-time.
Measurement tools that can minimize these assessment limitations are crucial in order to
accurately detect temporal patterns of dietary intake and to have measurement sensitivity to
variability in eating behavior. The identification of these proximal influences of eating behavior
has the potential to increase our understanding in how children and families influence one
another’s eating behaviors, and could guide future interventions and thus have a positive impact
on diet and ultimately obesity.
Emerging technologies present the opportunity of examining dietary behavior and contextual
factors in real-time, and to apply new analytic methods for modeling the multiple signals from
these complex systems. Wrist-worn and in-situ sensors allow for the examination of dietary
behavior in naturalistic settings over long periods of time. However, these novel technologies
still need to be evaluated in naturalistic field settings and with diverse populations before they
can be deployed in obesity research settings.
To address these gaps in the literature, this dissertation utilized data from the Monitoring
and Modeling Family Eating Dynamics (M2FED) study (Mondol et al., 2020; Spruijt-Metz et al.,
2016). The purpose of the M2FED study was to develop and deploy new methods for in-home
sensing that would accurately monitor and model family eating dynamics. This dissertation
begins the first step toward producing new models that develop behavioral theory, and enable the

28
identification of temporally specific processes and events within the family system that can be
targeted for personalized, context-specific, real-time feedback. Implications for future
interventions include providing this personalized, adaptive, and just-in-time feedback to
encourage healthier family eating dynamics that are likely to have a downstream positive impact
on diet and activity and ultimately obesity.
This dissertation investigated the influence of intra-personal and inter-personal factors, and
the synergy of these factors, on both individual and family eating behavior in order to:
• increase our understanding of how children and families influence one another’s eating
behaviors, and
• inform future obesity prevention and intervention strategies.
SPECIFIC AIMS
Study 1: Investigating the validity and feasibility of the Monitoring and Modeling Family
Eating Dynamics (M2FED) system to detect in-field family eating behavior
1. Evaluate participant compliance with the EMA protocol, (i) overall, (ii) for hourly time-
triggered survey assessments, and (iii) for eating event-triggered survey assessments.
2. Evaluate the impact of time (time of day, day of week, difference between week 1 and 2
of assessment), age, gender, family role, and other family members’ compliance (whether
another participating family member j had answered a survey that had been received
within 15 minutes of focal person i’s survey) on compliance.
3. Evaluate the performance of the wrist-worn smartwatch to automatically detect
participants’ eating events in the home.
4. Determine whether there were systematic differences in the detection of eating events by
age, gender, family role, and height.

29
Study 2: Utilizing wearable sensors and Ecological Momentary Assessment (EMA) to
automatically detect contextualized eating events in the home: a multi-level latent class
analysis
1. Identify and characterize the latent classes of participants’ eating events automatically
detected in the home, using self-reported data on an individual’s hunger, satiety, positive
and negative affect, mindful eating, eating in the absence of hunger, and eating
companionship (social context), collected immediately following an eating event via
EMA.
2. Examine eating event-level predictors of the latent classes of eating events.
Study 3: Investigating the dynamics of in-home family eating behavior: a relational event
modeling approach
1. Investigate the role of four proposed ego-centric mechanisms in predicting the sequence
of in-home intra-personal events (affective states) and behavioral events (eating events)
within a family system over a two-week period of time.
2. Examine whether the presence of these specific event sequences varies by family role
(e.g., parent, child) to establish whether there are consistent temporal patterns across all
types of family members or whether novel patterns emerge for certain types of family
members.

30
CHAPTER 2: INVESTIGATING THE VALIDITY AND FEASIBILITY OF THE
MONITORING AND MODELING FAMILY EATING DYNAMICS (M2FED) SYSTEM
TO DETECT IN-FIELD FAMILY EATING BEHAVIOR
ABSTRACT

Introduction: The field of dietary assessment has a long history marked by both controversies
and advances. Emerging technologies have been offered as a potential solution to address the
limitations of self-report dietary assessment methods. The Monitoring and Modeling Family
Eating Dynamics (M2FED) study utilizes wrist-worn smartwatches to automatically detect real-
time eating activity in the field. Ecological momentary assessment (EMA) methodology is also
employed to confirm whether eating occurred (i.e., ground-truth) and to measure other pertinent
contextual information. The purpose of this paper is to report on participant compliance
(feasibility) to the M2FED study’s two distinct EMA protocols (hourly time-triggered
assessments and eating event-triggered assessments) and on the performance (validity) of the
smartwatch algorithm in automatically detecting eating events in a family-based study.
Methods: 20 families (58 participants) participated in the two-week, observational, M2FED
study. All participants were instructed to wear a smartwatch on their dominant hand and to
respond to time-triggered and eating event-triggered mobile questionnaires via EMA while at
home. EMA data were processed with a “participation algorithm” that identified time intervals in
which participants were likely both at home and actively participating in the study. Compliance
to EMA was calculated (i) overall, (ii) for hourly time-triggered mobile questionnaires, and (iii)
for eating event-triggered mobile questionnaires. Predictors of compliance were determined with
a logistic regression model. The number of true positive and false positive eating events were

31
calculated, as well as the precision of the smartwatch algorithm. The Mann Whitney U Test, the
Kruskal-Wallis Test, and Spearman’s Rank Correlation were used to determine whether there
were differences in the detection of eating events by participant age, gender, family role, and
height.
Results: The overall compliance rate across the 20 deployments was 89.3% for all EMAs, 89.7%
for time-triggered EMAs, and 85.7% for eating event-triggered EMAs. Time of day and whether
other family members had also answered an EMA were significant predictors of compliance to
time-triggered EMAs (both p<0.01). Weekend status and deployment day were significant
predictors of compliance to eating event-triggered EMAs (p<0.05 and p<0.01, respectively).
Approximately three-fourths (76.5%) of detected eating events were confirmed as true positives,
and precision was 0.765. The proportion of correctly detected eating events did not significantly
differ by participant age, gender, family role, nor height (all p>0.05).
Conclusion: This paper demonstrates that EMA is a feasible tool to collect ground-truth eating
activity and thus evaluate the performance of wearable sensors in the field. The combination of a
wrist-worn smartwatch to automatically detect eating and a mobile or wearable device to capture
ground-truth eating activity offers key advantages for the user (participant) and makes mHealth
technologies more accessible to non-engineering behavioral researchers.
INTRODUCTION
Challenges to Dietary Assessment
A prevailing challenge to dietary intake/eating behavior research is the ability to
accurately measure dietary intake. Historically, the assessment of dietary intake and eating
behaviors utilizes self-reporting tools (Shim et al., 2014; Thompson et al., 2010), such as food

32
diaries, food frequency questionnaires (FFQ), and 24-hour dietary recalls (Magarey et al., 2011;
Willett, 1998). All dietary assessment self-report methods have some level of measurement error
(the difference between the measured value and the true value) (Beaton et al., 1997; Thompson
et al., 2015). Dietary data collected via self-report methods may be misreported due biases such
as recall/memory bias (when a respondent erroneously recalls their dietary intake) and social
desirability bias (when a respondent desires to present oneself positively) (Althubaiti, 2016;
Livingstone et al., 2004; Westerterp & Goris, 2002). Studies have also found that those with
certain characteristics (e.g., obese weight status, body image dissatisfaction) are more likely to
underreport their energy intake (Novotny et al., 2003; Wehling & Lusher, 2017).
Shifting Focus: From Dietary Intake to Eating Behavior and Context
The field of nutritional epidemiology has produced an abundance of studies that have
examined the role of dietary intake (i.e., what and how much is consumed) in human health and
disease – looking specifically at macronutrients (e.g., fats, carbohydrates), types of food, quality
of food, dietary patterns, and more (Willett, 2013). Decades of lab-based and observational
research indicates that dietary intake is a critical component of chronic disease prevention
(Nishida et al., 2004). However, the measurement of diet in free-living populations remains a
huge challenge in the field. And, even if public health researchers could easily and accurately
track free-living dietary intake, dietary intake patterns are notoriously difficult to change long-
term (Wood & Neal, 2016).
Eating behaviors and patterns (i.e., food choices and motives, feeding practices) and
context (i.e., who is eating, when, where, with whom, etc.) also play a significant role in the
development of obesity and other chronic diseases, including type 2 diabetes and heart disease
(Higgs & Thomas, 2016; Jannasch et al., 2017; Neuhouser, 2018; Reicks et al., 2015; Robinson

33
et al., 2014; Tourlouki et al., 2009). These findings indicate that the patterns and features of
eating events may be key contexts that shape dietary intake, and thus could be more malleable
features of eating behavior that could be intervened on. However, the field is still lacking
appropriate behavioral theories that provide a richer understanding of how eating behaviors vary
across contexts and across time (Riley et al., 2011; Spruijt-Metz et al., 2015).
Technology-Assisted Dietary Assessment
Emerging technologies offer a potential solution to the accurate assessment of dietary intake
by addressing the limitations of self-report dietary assessment methods. The incorporation of
technologies into dietary assessment can improve the quality and validity of dietary data by
passively measuring eating in naturalistic settings over long periods of time with minimal user
interaction (Spruijt-Metz et al., 2018). Two emerging technological advances in dietary
assessment tools include:
i. ecological momentary assessment (EMA): a data collection technique in which one’s
behavior is repeatedly sampled in real-time and in context (Engel et al., 2016; Schembre
et al., 2018; Shiffman et al., 2008), and
ii. wearable devices/sensors: allow for the passive collection of various data streams from
the physical environment (e.g., acoustic, visual, inertial, etc.) (Bell et al., 2020).
EMA and wearable sensors are able to measure behavior near- or just-in-time, thereby
reducing or eliminating the recall bias that can afflict retrospective self-report measures. In
addition to improving the validity of data, these technologies offer the opportunity to measure
eating behavior frequently and over long periods of time, allowing researchers to examine how it
varies over multiple timescales (e.g., varies over the day, over the week, etc.).

34
M2FED Study
To address the limitations of traditional dietary assessment methods and theories, the
Monitoring and Modeling Family Eating Dynamics (M2FED) study developed a sensor system
that utilized smartphones as well as deployable and wearable sensors to collect synchronized
real-time data on family eating behavior (Spruijt-Metz et al., 2016). This study used (i) wrist-
worn smartwatches containing inertial sensors (accelerometer and gyroscope) to automatically
detect arm movements and hand gestures associated with eating; (i) ecological momentary
assessment via smartphone to confirm whether the eating occurred and to measure other
contextual information, such as who was present during the eating event and the respondent’s
current mood; and (iii) proximity Bluetooth proximity beacons to determine the approximate
location of the smartwatches. Rather than focus on dietary intake (e.g., caloric intake, portion
sizes, etc.), this study took a novel approach by measuring eating behaviors (i.e., food choices
and motives, feeding practices) and context (i.e., who is eating, when, where, with whom, etc.).
Assessing Validity of Wearable Sensors
The feasibility of using wearable sensors to automatically assess eating behavior and context
has been tested in both lab and field settings (Bell et al., 2020; Hassannejad et al., 2017;
Heydarian et al., 2019; Vu et al., 2017), with indication that the performance of the wearable
sensors decreases in naturalistic settings (compared to controlled laboratory settings). Studies
have used a variety of sensors (e.g., microphones, cameras, smartwatches, EMG electrodes, etc.)
to measure a variety of dietary outcomes, including bites, chewing, swallowing, and duration of
eating occasions (Bell et al., 2020; Boushey et al., 2017; Gemming, Utter, & Ni Mhurchu,
2015b; Hassannejad et al., 2017; Heydarian et al., 2019; Vu et al., 2017). A review by Bell and
colleagues indicates that there is still a strong reliance on retrospective self-report methods (e.g.,

35
end-of-day food diaries) to determine ground-truth eating activity to evaluate wearable sensors in
the field (Bell et al., 2020). Given the aforementioned limitations of retrospective self-report
methods to accurately assess diet, the M2FED study used event-contingent EMA to determine
ground-truth eating activity in families. The use of EMA offers unique methodological
advantages, such as:
• the ability to measure behavior near- or just-in-time, thereby reducing recall bias and
reducing participant burden, and;
• the ability to measure behavior at the location in which it actually occurs, thereby
maximizing ecological validity (Shiffman et al., 2008).
The validity of this method has been tested in a few in-field studies (Gomes & Sousa, 2019;
Ye et al., 2016), however it has not yet been tested in a family-based study.
Assessing Feasibility of EMA
One disadvantage of using technologies for data collection is the potential for participant
noncompliance. A recent systematic review and meta-analysis by Wen and colleagues found that
compliance rates among EMA studies in youth samples were suboptimal – the weighted average
compliance rate was 78.3% (Wen, Schneider, Stone, & Spruijt-Metz, 2017), falling under the
recommended 80% compliance rate (Shiffman et al., 2008). Many studies have explored EMA
compliance for a variety of behaviors in various populations (Heron, Everhart, McHale, &
Smyth, 2017; Liao, Skelton, Dunton, & Bruening, 2016; Maher, Rebar, & Dunton, 2018; Nam,
Whittemore, Vlahov, & Dunton, 2019; Wen et al., 2017), but the compliance rate for a family-
based EMA study is underexplored. A recent EMA study involving mothers and their child
found that mothers’ presence may enhance children’s compliance to EMA questionnaires

36
(Dzubur, Huh, Maher, Intille, & Dunton, 2018), suggesting that family members and other social
relations may be leveraged to increase compliance in future EMA studies.
Study Aims
Therefore, the overall aims of this study were to report on participant compliance
(feasibility) to the study’s two distinct EMA protocols (hourly time-triggered assessments and
eating event-triggered assessments), and on the performance (validity) of the wearable sensor in
automatically detecting eating events in a family-based study. Specifically, the primary aims of
this study were to:
Aim 1: (a) Evaluate participant compliance with the EMA protocol, (i) overall, (ii) for
hourly time-triggered survey assessments, and (iii) for eating event-triggered survey assessments.
(b) Evaluate the impact of time (time of day, day of week, deployment day), age, gender, family
role, and other family members’ compliance (whether another participating family member j had
answered a survey that had been received within 15 minutes of focal person i’s survey) on
compliance.
Aim 2: (a) Evaluate the performance of the wrist-worn smartwatch to automatically
detect participants’ eating events in the home. (b) Determine whether there were systematic
differences in the detection of eating events by age, gender, family role, and height.
METHODS
Participants and Recruitment
Eligibility. The research team recruited families that contained at least two members total
(including at least one adult parent and one child between the ages of 11 and 18 years old) living
in Los Angeles County. Families containing children under the age of 11 were eligible to

37
participate, however that child/those children under the age of 11 was/were not permitted to
participate in the study. Families were not eligible to participate if one or more family members
living in the home did not primarily speak English. There were no demographic- or disease-
related exclusion criteria.
Method of recruitment. Families were recruited in public spaces and at public events
within Los Angeles County from May 2017 to August 2019. Snowball sampling was also
employed, such that participating families were offered an additional $20 if they referred other
eligible families that successfully enrolled in the study.
All families that expressed interest and met eligibility requirements were invited to
participate in the study. An intake screening tool was administered over the phone by recruitment
coordination staff to confirm eligibility before enrolling in the study.
This study was approved by the Institutional Review Board of the University of Southern
California. All parents provided informed written consent, and all children provided assent.
M2FED System Overview
The primary objective of the M2FED study was to develop and deploy the M2FED
cyberphysical system (Figure 1) in families’ homes. Cyberphysical systems can be defined as
“physical and engineered systems whose operations are monitored, coordinated, controlled and
integrated by a computing and communication core” (Rajkumar, Lee, Sha, & Stankovic, 2010).
This novel system monitored in-home family eating behavior for all participants. This
system contained four primary components:
1. sensors (including smartwatches, smartphones, and Bluetooth proximity beacons);
2. a base station;
3. an ecological momentary assessment (EMA) subsystem; and

38
4. a remote monitoring subsystem;
all of which were connected through a Wi-Fi router (see Figure 1).
For the scope of this study, all data collected by the system were measured in the home (i.e.,
no data were collected outside of the home).
Sensors. Participants were instructed to wear a Sony Smartwatch 3 (Android Wear
operating system) on their dominant hand during all waking hours that they were in their home.
The smartwatches were used to automatically detect eating-related hand-to-mouth (H-t-M)
gestures for each participant in the home and in real-time. Arm movements and these H-t-M
gestures were detected via an algorithm that utilized motion data from the inertial sensors inside
the smartwatch (accelerometer and gyroscope) (Mondol et al., 2020). If a cluster of at least two
H-t-M gestures were detected within a one-minute timeframe, then the motion data were
processed with a more sophisticated algorithm and these “clusters” were then characterized as an
“eating event”. An eating event can be defined as a set of H-t-M gestures, representing
phenomenon such as consuming a meal, a snack, a drink or a combination of these consumption
behaviors in which H-t-M gestures are clustered temporally. The technical details of the eating
event detection algorithm are provided in detail elsewhere (Mondol et al., 2020). Participants
were instructed to only wear the smartwatch in the home, and to not take it outside or wear it
outside of the home. Consequently, data on H-t-M gestures and eating events that were
determined by the proximity beacons to occur outside of the home were discarded.
Participants were each provided with a Samsung Galaxy S7 smartphone (Android
operating system) pre-programmed with limited functioning. The smartphone application (app)
in which they responded to mobile questionnaires was pinned to the screen so that they could not
access other applications on the smartphone. This smartphone was only intended to be used as a

39
data collection tool. Participants were instructed to keep the smartphone in the home and to not
take it outside of the home. If a smartphone left the home and was not within range of the Wi-Fi
router, then the phone did not receive any mobile questionnaires. Consequently, data on
participants’ states and behaviors outside of the home were not collected.
Estimote Bluetooth Low Energy (BLE) proximity beacons were used to determine the
approximate location of participants’ smartwatches (including approximately which room the
watches were in, and whether they were still in the home) during the length of the study. The
beacons continuously broadcasted “packets” that included the unique MAC address of the
Bluetooth interface, while the smartwatches periodically scanned for these “packets”. Then, the
smartwatches recorded the Received Signal Strength Indicator (RSSI) (signal from the beacons),
which indicated the proximity of the smartwatches to the beacons.
Typically one to two beacons were placed on a wall in each living space in the home
(excluding bathrooms and bedrooms), and required no further action by the participants during
the study.
Base station. A base station is a radio receiver/transmitter that serves as the hub of a local
wireless network (the M2FED system). The base station for the M2FED system was a Lenovo
ThinkPad laptop and that was placed in the family’s home for the duration of the study. The
laptop was placed in a locked cage so that it could not be tampered with. The base station
collected and processed the data that it received from the smartphones and smartwatches through
the Wi-Fi router, and it managed the EMA subsystem that ran on the laptop as well.
EMA subsystem. Ecological Momentary Assessment (EMA) is a data collection
technique in which one’s behavior is repeatedly sampled in the natural environment (Shiffman et
al., 2008). In this study, participants were assessed on a number of individual behaviors and

40
states via mobile questionnaires sent to their smartphone approximately every hour during
waking hours. Each smartphone had an application (app) developed by members of our research
team installed on it. The app acted as the mobile questionnaire platform (i.e., participants
answered the questionnaires within the app interface).
There were two types of EMAs that participants received: time-triggered mobile
questionnaires and eating event-triggered mobile questionnaires.
A time-triggered mobile questionnaire was sent to participants’ smartphones every
hour at the top of the hour (e.g., 10am, 11am, 12pm, etc.) (Figure 2a). The questionnaire
included a brief validated positive affect and negative affect survey.
Shortly after an eating event was detected for any given participant, an eating event-
triggered mobile questionnaire was sent to the corresponding participant’s smartphone asking
to confirm whether or not they had just eaten (Figure 2b). If they confirmed that they had just
eaten, then following this first question, they were asked a battery of survey items including
validated measures of hunger and satiety (Cardello, Schutz, Lesher, & Merrill, 2005), mindful
eating (Framson et al., 2009), and positive/negative affect (Cohen, Kamarck, & Mermelstein,
1983; Forrest et al., 2018; Laurent et al., 1999; Terry, Lane, Lane, & Keohane, 1999); and with
whom they were eating, if anyone. If the participant had not finished eating, then they were
given the option to request more time before filling out the questionnaire.
If they responded to the first question indicating that they had not just eaten, then they
were asked to report what activity they had just completed. They were then asked validated
measures of positive and negative affect. Figure 3 illustrates the full eating event-triggered EMA
question logic. The full list the questions for the time-triggered and the eating event-triggered
mobile questionnaires can be found in Table 1.

41
Participation windows. Before a family’s deployment started, all participants were
individually asked about the time at which they normally woke up, and the time at which they
normally went to bed. Participants were limited to only one “personalized participation window”
for the study. Therefore, they could not have different windows for Monday vs. Tuesday,
weekday vs. weekend, etc. If the times at which they woke up and/or went to bed varied
extensively between days, then they were asked to provide a time frame that generally worked
for all days.
The purpose was to create “personalized participation windows” in order to account for
variation in participants’ daily routines and sleeping patterns. For the duration of the study,
participants only received EMAs during their personalized participation window. For example, if
a participant’s window was from 6:30am to 11:00pm, then they only received EMAs during that
time period.
Remote monitoring subsystem. The monitoring subsystem (named M
2
G) was used to
monitor the status of the M2FED system in real-time (Ma et al., 2017). The subsystem monitored
a number of things, including the battery status and network connection of the smartwatches, the
smartphones, and the base station; the processes running on the base station; the detected eating
events; and whether or not participants responded to any given EMA sent to their smartphones.
When the monitoring system detected an issue (e.g., the base station was no longer connected to
the router), an email was sent to the research team to alert them of the issue.
Procedures
Following enrollment, two members of the research team visited the homes of all
recruited families a total of two times.

42
Visit 1. For the first home visit, the team went to the participants’ home to consent all
participating family members, took participants’ body measurements using a research-grade
Tanita  scale (Model TBF 300) and stadiometer, administered baseline surveys, and installed
the components of the cyberphysical system around the home (all “living spaces”; not including
bedrooms or bathrooms).
The system’s base station, Wi-Fi router, and Bluetooth beacons were placed in a discrete
location in the family’s home so they could run without interference for the duration of the study.
Samsung smartphones and Sony smartwatches were provided to all participating family
members for the duration of the study (all features except for answering questionnaires were
turned off). Each phone and watch were designated to a specific participant, and labeled with
their name so that they knew which devices were theirs. The team instructed the family how to
properly wear, charge, and care for the smartwatches, and how to answer an EMA on the
smartphones. The family was instructed to wear the watch at all times when they were at home,
and to answer all EMA questionnaires they received when they were at home. They were also
instructed to leave their designed phone and watch at home when they left in order to prevent the
devices from getting damaged or lost while outside of the home.
Upon leaving the visit, family members underwent approximately fourteen consecutive
days of (1) use of a smartphone to complete hourly time-triggered and eating event-triggered
mobile questionnaires, up to once every hour during waking hours; and (2) eating event
monitoring, in the form of a wrist-worn smartwatch during waking hours.
Visit 2. At the final home visit, approximately two weeks following the first home visit,
the research team terminated data collection, and uninstalled and removed all equipment from

43
the home. Each participant received $100 in a Visa gift card format as compensation for the two-
week study.
Measures
Eating events. During the two-week assessment period, participants were asked to wear
their dedicated smartwatch on their dominant wrist at all times while they were home during
waking hours. Automatic eating event detection software on the smartwatches developed by our
research team (Mondol et al., 2020) collected the timestamps (approximate start and end times,
format: mm/dd/yyyy, hh:mm:ss) for all detected eating events that occurred while the watch was
worn. After an eating event was detected, participants received a brief mobile questionnaire on
their study phones to confirm whether the detected eating event was a true event. The first
question on the questionnaire was “Were you eating or drinking just now?” If the participant
responded “No,” then they were asked to report what they were doing. Options included “Using
my phone,” “Smoking,” “Fixing my hair,” “Putting on sunscreen or lotion,” or “Other” with an
open-text field. If the participant responded “Yes,” then they were asked to report on a range of
momentary measures, such as hunger level before the eating event and with whom they were
eating. The full list the questions for the time-triggered and the eating event-triggered mobile
questionnaires can be found in Table 1.
EMAs. Timestamps (format: mm/dd/yyyy, hh:mm:ss) of when the hourly time-triggered
and eating event-triggered mobile questionnaires were sent to and received by participants’
smartphones were obtained from the monitoring system. Additionally, whether or not the
participant completed the questionnaire was obtained.
Timing. “Time of day” at which and “day of week” on which an eating event occurred
was calculated using the timestamp of the detected eating events. Time of day at which the

44
eating event occurred was stored in hh:mm:ss format. The ‘lubridate’ R package (Grolemund &
Wickham, 2011) was used to convert the date on which the eating event occurred (format:
mm/dd/yyyy) to the day of corresponding week (e.g., Monday, Tuesday, etc.), which was then
converted to weekday (Monday, Tuesday, etc.) and weekend (Saturday, Sunday).
Anthropometrics. During Home Visit 1, height (cm), weight (lbs), and body fat
percentage (%) were measured in all participants in a private section of the home, using a
portable stadiometer and a research grade Tanita  scale (model TBF 300).
Demographics. During Home Visit 1, participants were asked to provide basic
demographic information via a paper questionnaire, including their current age (in years), gender
(Female, Male), race (Hispanic or Latino, Asian or Pacific Islander, White or Caucasian, Black
or African-American, American Indian or Native American, Mixed, Other), Hispanic/Latino
ethnicity (Yes, No, Don’t know), and family role (Mother, Father, Child, Grandparent, Aunt,
Uncle, Other).
Analytic Approach
Data processing
A limitation of the M2FED study’s EMA sampling protocol was that the participants’
study phones (which were instructed to be kept in the home at all times) received the hourly,
time-triggered surveys regardless of whether the participants themselves were at home or not
(e.g., at school/work, running errands, etc.). This means that time frame in which any given
participant was at home and participating in the study is not necessarily continuous. Although we
do not possess the ground-truth for participants’ presence in the home (e.g., no cameras, no self-
report diaries), our research team generated an “participation algorithm” using the EMA system,
the proximity sensors, and the accelerometer in the watch to identify time intervals in which we

45
are confident participants were both (i) at home and (ii) actively participating in the study
(i.e., answering EMAs and/or wearing the smartwatch) (Figure 4).
If participants had answered an EMA at time t, then we assigned their status as
“participating” for the 30-minute interval surrounding time t, i.e. from t-15 minutes to t+15
minutes. For times outside the EMA interaction windows, we used data from the sensors
(smartwatch accelerometer and Bluetooth beacons) to determine the participants’ status. For
every minute, if the smartwatch’s accelerometer data was both available (i.e., not missing for that
minute) AND indicated movement (i.e. the frequency and instantaneous changes of the sensor
signal was above a threshold, representing change in the signal due to movement) AND beacon
data were available, then they were classified as “participating” for that 1-minute interval.
Contiguous minute intervals with “participating” status were merged to acquire larger time
intervals. For each participant, these “participation” time intervals were calculated, and the union
of all intervals (Figure 5) were used as the valid time intervals in the analyses.
Data analyses
Individual- and family-level characteristics. The mean/standard deviation or the
count/proportion of the analytic sample’s age, body mass index, gender, and race/ethnicity were
calculated and reported by family role (child, parent). At the family-level, the count and
proportion of families’ type of household (1- or 2-parent household), number of children living
in the home, and average length of family deployment were reported.
EMA descriptives. The mean and standard deviation of EMAs received per family,
received per person, and received per person per day, were calculated. After applying the
participation algorithm to the EMA data, these metrics were calculated again and reported in a

46
separate table. The frequency distribution of EMAs by family role and by time of day were
calculated.
Aim 1(a): Evaluate participant compliance with the EMA protocol, (i) overall, (ii) for
hourly time-triggered survey assessments, and (iii) for eating event-triggered survey
assessments.
To test Study Aim 1(a), EMA compliance was calculated as follows (where i = 1, 2, …, n
number of participants in the study):
i. Overall compliance to EMAs for participanti = total number of EMAs answered by
participanti / total number of EMAs received at home by participanti
ii. Compliance to time-triggered EMAs for participanti: total number of time-triggered
EMAs answered / total number of time-triggered EMAs received at home by participanti
iii. Compliance to eating event-triggered EMAs for participanti: total number of eating
event-triggered EMAs answered / total number of eating event-triggered EMAs received
at home by participanti
Means and standard deviations of overall compliance to EMAs, compliance to time-triggered
EMAs, compliance to eating event-triggered EMAs were also calculated across all participants.
Aim 1(b): Evaluate the impact of time (time of day, day of week, difference between
week 1 and 2 of assessment), age, gender, family role, and other family members’
compliance (whether another participating family member j had answered a survey that
had been received within 15 minutes of focal person i’s survey) on compliance.
To test Study Aim 1(b), the unit of analysis was every EMA that was sent to and received by
participants’ smartphones throughout the span of the two-week data collection period.
Compliance (dependent variable) was calculated as “1” if the questionnaire was answered, and as

47
“0” if the survey was not answered. A logistic regression model was fit with the following
independent variables: type of EMA (time-triggered, eating event-triggered), time of day
(morning, afternoon, evening), day of week (weekday, weekend), gender (male, female), family
role (parent, child, other), and social factors (whether another participating family member j had
answered a survey that had been received within 15 minutes of focal person i’s questionnaire).
Aim 2(a): Evaluate the performance of the wrist-worn smartwatch to automatically
detect participants’ eating events in the home.
To test Study Aim 2(a), we evaluated the performance of the smartwatch by computing
the following metrics for all eating events automatically detected during the deployments:
i. True positives (TP) = cases in which an eating event actually did occur, and that eating
event was correctly detected by the smartwatch algorithm
ii. False positives (FP) = cases in which an eating event actually did not occur, but an eating
event was erroneously detected by the smartwatch algorithm
iii. Precision = TP / (TP + FP)
Aim 2(b): Determine whether there were systematic differences in the detection of
eating events by age, gender, family role, and height.
To test Study Aim 2(b), non-parametric methods were used to determine whether there were
differences in the detection of eating events by participant age, gender, family role, and height.
The metric we used to compare across demographic groups was the proportion of correctly
detected eating events (=true positives / total number of detected eating events). If any
participant had received fewer than 3 eating event-triggered EMAs, then their data were removed
for this analysis.

48
For categorical variables with two groups (i.e., gender), the appropriate assumptions were
tested
7
, and then the Mann Whitney U Test was used to test for equality of central tendency of
the two distributions; for categorical variables with three or more categories (i.e., family role),
the Kruskal-Wallis Test was used. Lastly, for continuous variables (i.e., height (cm), age
(years)), the appropriate assumptions were tested
8
, and then Spearman’s Rank Correlation was
used to measure the strength and direction of the relationship between the continuous variable
and the proportion of correctly detected events.
Missing data
There were no missing data for the anthropometric and demographic variables. Similarly,
there were no missing data on detected eating events and corresponding variables including time
of eating event and day of eating event. However, there were missing data for the time-triggered
and eating event-triggered EMAs.
Missingness attributed to technical issues. Preliminary analyses indicated that not all
EMAs that were sent to participants’ study phones by the M2FED system were received by the
phone. The M2FED system ran independently on the base station regardless of network
connection, and therefore it sent EMAs regardless of network connection. However, network
connection was needed in order for the phone to successfully receive the EMA.
Although we do not have data that explain why this happened in every instance, we know
from in-the-field troubleshooting and from accounts given by participants that at least a portion

7
The assumptions of the Mann Whitney U Test and the Kruskal-Wallis Test are that (1) the
dependent variable is ordinal or continuous, (2) the independent variable consists of two or more
categorical, independent groups, (3) there is independence of observations, and (4) the two or
more distributions have the same shape (in order to compare medians of dependent variable).
8
The assumptions of the Spearman correlation are that (1) data must be at least ordinal, and (2)
the scores on one variable must be monotonically related to the other variable.

49
of the non-received EMAs resulted from (i) network connection issues in the home (i.e., the
router was not working and the EMAs could not be received on the phone) and (ii) EMA
application failure (i.e., the EMA application on the phone failed to work properly).
For these analyses, we removed any EMAs that were sent by the system but not received
by the phone.
Missingness attributed to participant non-response or partial response. A different type of
missing data that we encountered was due to participant non-response (i.e., participant did not
respond to any EMA questions) or partial response (i.e., participant did not respond to all EMA
questions).
For Aim 1 analyses, if participants did not respond to any questions on a given mobile
questionnaire, then this EMA was labeled as ‘received but not answered’. If participants did not
respond to all questions, then this EMA was labeled as ‘received and partially answered’. These
EMA observations were kept in the dataset in order to calculate EMA compliance.
For Aim 2 analyses, if participants did not respond to at least the first question on a given
eating event-triggered EMA (“Were you eating or drinking just now?”), then this EMA
observation was removed from the dataset.
Statistical software
R (version 4.0.2) was used to perform these analyses.
RESULTS
Individual- and Family-Level Characteristics
A total of 74 participants from 20 families enrolled in the M2FED study. Thirteen
participants dropped out of the study or were removed from the dataset if their participation (as

50
determined by the participation algorithm) was 0% (i.e., they did not answer any EMAs nor ever
wore the smartwatch) (Figure 6). Additionally, the data from three non-parent adult participants
made up approximately 1.5% of the EMAs received, so these participants were removed from
the analytic sample as well. The remaining 58 participants included in the analytic sample did
not significantly differ from the enrolled sample (N=74) by age, gender, or parent role (Table 2).
Of the 58 participants, 41.4% were parents (n=24) and 58.6% were children (n=34). On
average, children were 15.74 years old (SD=5.30 years) and parents were 44.04 years old
(SD=6.65 years). There were 13 female children (38.2% of children) and 17 female parents
(70.8% of parents). 61.8% of children and 66.7% of parents identified as Hispanic or Latino
(Table 3).
Of the 20 enrolled families, a majority (85.0%) were 2-parent households. Three families
(15.0%) had 1 child living in the home; fifteen families (75.0%) had 2 children; one family had 3
children; and one family had 4 children (Table 4). On average, family deployments lasted 14.90
days (SD=3.13).
EMA Descriptives
In total, 15,010 EMAs (14,348 time-triggered and 662 eating event-triggered) were sent
by the M2FED system and received by participants’ study phones. After filtering the data
through the participation algorithm, a total of 4,171 EMAs remained in the dataset: 3,710 time-
triggered and 461 eating event-triggered (Table 5).
On average, families received 209.0 EMAs (SD=89.4, Range=[86, 391]), and individuals
received 71.9 EMAs (SD=34.3, Range=[8, 176]) each. Participants received, on average, 64.0
time-triggered EMAs (SD=31.3, Range=[8, 147]) and 8.0 eating event-triggered EMAs (SD=8.9,
Range=[0, 40]) across the deployment. The daily average number of EMAs received per person

51
was 5.2 (SD=2.7, Range=[0.6, 11.7]) for all EMAs, 4.7 (SD=2.4, Range=[0.3, 10.2]) for time-
triggered EMAs, and 0.6 (SD=0.6, Range=[0, 2.7]) for eating event triggered EMAs (Table 5).
Of the 4171 total EMAs, 775 (18.6%) were received in the morning, 1270 (30.4%) in the
afternoon, and 2126 (51.0%) in the evening. Of the 461 eating event-triggered EMAs, the
majority (211; 45.8%) were sent in the evening (Figure 7). Children received 2399 (57.5%) of
the total EMAs, fathers received 447 (10.7%), and mothers received 1325 (31.8%). Of the 461
eating event-triggered EMAs, children received 230 (49.9%), fathers received 34 (7.4%), and
mothers received 197 (42.7%) (Figure 8).
Participant Compliance (Aim 1a)
The overall compliance rate across the 20 deployments was 89.3% for all EMAs, 89.7%
for time-triggered EMAs, and 85.7% for eating event-triggered EMAs (Table 6). The average
family-level compliance was 89.4% (SD=5.74%, Range=[75.7%, 98.1%]) for all EMAs, 89.8%
(SD=5.84%, Range=[75.8%, 98.7%]) for time-triggered EMAs, and 85.9% (SD=14.3%,
Range=[55.6%, 100%]) for eating event-triggered EMAs. At the individual-level, the average
compliance for all EMAs was 89.6% (SD=9.50%, Range=[53.8%, 100%]), for time-triggered
EMAs was 89.5% (SD=10.1%, Range=[50.0%, 100%]), and for eating event-triggered EMAs
was 88.0% (SD=17.5%, Range=[28.6%, 100%]).The distributions of individual- and family-
level compliance are shown in Figure 9.
Predictors of Compliance (Aim 1b)
Logistic regression models were fit for (i) all EMAs, (ii) time-triggered EMAs, and (iii)
eating event-triggered EMAs.

52
Results from the first model indicate that time of day and whether other family members
had also answered an EMA were significant predictors of compliance to all EMAs (all p<0.01)
(Table 7). Participants were 37% less likely (OR=0.63, 95% CI: 0.46, 0.86) to respond to an
EMA in the afternoon; and 39% less likely (OR=0.61, 95% CI: 0.45, 0.81) to respond to an
EMA in the evening, compared to in the morning (reference group). Participants were 91% more
likely (OR=1.91, 95% CI: 1.56, 2.34) to respond to an EMA if another family member had
responded to an EMA in the surrounding 30 minute time interval.
Results from the second model indicate that time of day (p<0.01) and whether other
family members had also answered an EMA (p<0.001) were significant predictors of compliance
to time-triggered EMAs (Table 7). Participants were 40% less likely (OR=0.60, 95% CI: 0.42,
0.85) to respond to a time-triggered EMA in the afternoon; and 47% less likely (OR=0.53, 95%
CI: 0.38, 0.74) to respond to a time-triggered EMA in the evening, compared to in the morning
(reference group). Participants were roughly two times as likely (OR=2.07, 95% CI: 1.66, 2.58)
to respond to a time-triggered EMA if another family member had responded to any EMA in the
surrounding 30 minute time interval.
Results from the third model indicate that weekend status (p<0.05) and deployment day
(p<0.01) were significant predictors of compliance to eating event-triggered EMAs (Table 7).
Participants were 2.4 times as likely (OR=2.40, 95% CI: 1.25, 4.91) to respond to an eating
event-triggered EMA on the weekend, compared to on a weekday. Participants were 8% less
likely (OR=0.92, 95% CI: 0.86, 0.97) to respond to an eating event-triggered EMA for every 1-
day increase in deployment day.

53
Smartwatch Algorithm Evaluation (Aim 2a)
For a subsample of participants (n=46), at least one eating event was automatically
detected during the deployments. This subsample did not significantly differ from the enrolled
sample (N=74) by age, gender, or parent role (Table 2).
A total of 461 eating events were automatically detected by the smartwatch algorithm
across these 46 participants. Participants responded to 395 (85.7%) of the corresponding eating
event-triggered EMAs. Participants confirmed that 302 of the 395 (76.5%) detected events were
true eating events (i.e., true positives); 93 of the 395 (23.5%) were not true eating events (i.e.,
false positives). The calculated precision measure, i.e., the number of true positives divided by
the sum of true positives and false negatives, was 0.765.
Differences in Eating Event Detection (Aim 2b)
Thirty-six participants received at least 3 eating event-triggered EMAs. This subsample
of participants did not significantly differ from the enrolled sample (N=74) by age, gender, or
parent role (Table 2).
For this subsample, the average individual-level proportion of correctly detected eating
events (true positives / total number of detected eating events) was 78.5% (SD=19.0%,
Range=[30.0%, 100%]. Neither age (in years) nor height (in inches) were significantly correlated
with the proportion of correctly detected eating events (rs=0.236 and p=0.167; rs=-0.116 and
p=0.519; respectively).
The average individual-level proportion of correctly detected eating events for females
was 82.1% (SD=20.4%, Range=[30.0%, 100%]) and for males was 74.0% (SD=16.6%,
Range=[50.0%, 100%]). The difference between the two groups was not significant (W=112,
p=0.126). The average individual-level proportion of correctly detected eating events for

54
children was 74.3% (SD=19.3%, Range=[30.0%, 100%]), for fathers was 76.1% (SD=21.5%,
Range=[58.3%, 100%]), and for mothers was 86.5% (SD=16.8%, Range=[54.5%, 100%]). The
difference among these three groups was not significant (K-W χ
2
=2.998, p=0.223).
DICUSSION
The M2FED study sought a dramatically different mHealth approach to obesity
prevention and intervention by not focusing directly on diet and activity, but rather on family
eating dynamics (FED). An in-home sensor system was developed and deployed in order to
monitor FED in real-time and in context.
Evaluating EMA Compliance
After applying our customized participation algorithm, we found that both individual-
and family-level compliance rates to the study’s EMA protocols were relatively high (both > the
recommended 80%) (Shiffman et al., 2008). Unsurprisingly, compliance was significantly higher
in the mornings overall and higher on the weekends for eating event-triggered EMAs. We also
saw that overall compliance decreased as the two-week study went on, most likely attributable to
participant fatigue.
One particularly interesting finding was that participants were significantly more likely to
answer an EMA if another family member had answered an EMA in a similar time frame. A
similar finding was reported in Dzubur (2018), in which mother-child dyads were more likely to
comply with prompts when they were together (Dzubur et al., 2018). Although the overarching
aims of the M2FED study were to measure the social influence of family members on eating
behavior, this finding also indicates that social influence came into play in other parts of the
study as well. Drawing from the social psychology field, a number of social mechanisms could

55
partially explain these findings. For instance, an expectation could have been set early on in
particular families to answer the EMA prompts, thus establishing a social norm for EMA
compliance (Cialdini & Goldstein, 2004; Cialdini & Trost, 1998). Similarly, some individuals
may have been inclined to answer EMA prompts to conform to the behavior of other family
members around the same time (Cialdini & Goldstein, 2004; Cialdini & Trost, 1998), especially
considering that family members received their time-triggered EMAs at approximately the same
time as each other.
Studies have used EMA to measure a variety of dietary outcomes, including frequency of
food intake, intake of specific types of foods (e.g., low glycemic index foods), and energy intake
(Schembre et al., 2018). It’s been suggested in a recent systematic review of mobile ecological
momentary diet assessment methods that EMA has the potential to be a novel dietary assessment
method, both on its own and as a supplement to other mHealth technologies (Schembre et al.,
2018). The use of EMA to assess diet and/or eating behavior provides some key advantages –
namely, the reduction of participant burden and recall bias, and the maximization of ecological
validity (Schembre et al., 2018). Taken together with the findings from Dzubur and Schembre,
our findings suggest that EMA can be used to sufficiently supplement automatic dietary
assessment (ADA) approaches, and may be a particularly useful approach to leverage social
relations and to maintain compliance in dyad- and group-based EMA studies.
Evaluating Automatic Dietary Assessment (ADA)
A variety of technologies have been used to passively measure eating activity in
naturalistic settings over long periods of time with minimal user interaction. One of the most
popular technologies to assess diet/eating in the field is the wrist-worn smartwatch and/or
accelerometer (Bell et al., 2020; Spruijt-Metz et al., 2018). The performance of automatic,

56
wearable-based, in-field eating detection approaches to date has been reviewed by Bell and
colleagues (2020). The smartwatch utilized in the M2FED study performed on par with other in-
field devices, although comparability is difficult due to the wide and varying metrics used by
other papers (Bell et al., 2020). Although some wearable devices included in this review
performed very well, the duration of the free-living deployment was 1 day (~24 hours) or shorter
for more than half of the studies, and another one-third were 1-week in length or shorter (Bell et
al., 2020).
Three studies had durations that lasted at least two weeks or longer (Navarathna,
Bequette, & Cameron, 2018; Thomaz, Essa, & Abowd, 2015; Ye et al., 2016), two of which had
sample sizes of only 1 participant each. Therefore, the M2FED study is one of the first studies to
extensively test the feasibility of deploying an ADA approach for a considerable amount of time
(2 weeks) and with a relatively large same size (> 50 participants). Part of this success stems
from the combined use of mobile devices (for EMA) and smartwatches, which were selected for
the M2FED study to maximize long-term usability. Although other technologies have been able
to perform better in the field, the usability of these technologies (EMG electrodes, ear and neck
sensors, wearable video cameras, etc.) may be lower compared to wrist-worn devices because of
the inconvenient location of sensor placement, the potential to interfere with the participant’s
behavior in real life (Rast & Labruyère, 2020), and the potential intrusiveness or discomfort
caused by the sensor (Fontana, Farooq, & Sazonov, 2021).
This study also demonstrates that EMA is a feasible tool to collect ground-truth eating
activity and thus evaluate the performance of wearable sensors in the field. Only two studies
(Gomes & Sousa, 2019; Ye et al., 2016) included in the Bell et al. review used a novel method
for obtaining ground-truth eating activity in the wild similar to the way EMA was used in the

57
M2FED study. In Ye et al., when an eating gesture was automatically detected via wrist-worn
sensor, participants were sent a short message on their smartwatch to confirm or reject in real-
time whether they were eating (Ye et al., 2016). Similarly, in Gomes & Sousa, when drinking
activity was detected via wearable sensor, participants were sent an alert on their smartphone and
could then confirm or reject whether they were drinking via EMA (Gomes & Sousa, 2019).
Although EMA and similar self-report methods have their own limitations (McClung et al.,
2018; Spruijt-Metz et al., 2018), they offer the ability to capture and validate ground-truth eating
activity near the time of eating and thus improve research scalability and participant acceptability
(Schembre et al., 2018).
Another key feature of the M2FED study was the ability to capture intra-personal
(individual) and inter-personal (social) contexts with our combined event- and signal-contingent
protocols. A systematic review notes that fewer than 7% of EMA studies assessing diet use a
combined approach (Maugeri & Barchitta, 2019). EMA is a powerful tool that can be used to
validate automatically detected eating behavior in field and to easily collect information about
meaningful contexts, but few studies have used this approach and still rely on paper-pen to
validate their findings (Bell et al., 2020).
Limitations and Strengths
The M2FED study design is not without notable limitations. Firstly, our method of
obtaining ground-truth eating was only deployed via eating-event-triggered EMA after an eating
event was detected by the smartwatch. Thus, we could only verify true positive eating events and
false positive eating events. The M2FED system was not designed to verify true negative or false
negative eating events, which limits our ability to calculate common evaluation metrics (namely,
accuracy and F1-score) and compare our results to other in-field studies described in the

58
literature. Secondly, the false positive eating events were self-reported validation, which might
be subject to social desirability in under-reporting an eating event. This could potentially bias the
validity of the results. Thirdly, we encountered various difficulties with the deployed
technologies, including the smartwatches (i.e., limited battery), the mobile phones (i.e., limited
battery, app crashes), and the Wi-Fi router (i.e., wireless connection dropped). Although these
challenges were anticipated and addressed in a timely manner on all occasions, some data were
lost during the data collection process.
On the other hand, this study also possesses a number of strengths. Firstly, we recruited a
large and ethnically diverse sample of Los Angeles families. It has been previously noted that the
lack of diverse samples in eating-related mHealth and EMA studies is a major limitation of past
research (Smith & Juarascio, 2019). Secondly, as noted above, the M2FED study facilitated one
of the longest in-field deployments found in the literature so far. The majority of ADA research
has taken place in the lab. By deploying in the field, we are able to better understand real-life
eating behavior (vs. eating behavior in a lab) and gain a better understanding of the challenges
that come with deploying wearable sensors outside of the lab. Thirdly, because the deployment
process was across a two-year period, we were able to iteratively improve our automatic eating
event detection algorithm and then utilize the newest version in the following deployments.
Future Directions
The mHealth field is converging toward the use of a combination of user-friendly devices
to assess eating behavior in the wild (e.g., mobile phones, wrist-worn devices) (Bell et al., 2020;
Heydarian et al., 2019). Implementing user-friendly technologies for in-field dietary assessment
or eating behavior interventions offers at least two substantial advantages – people are generally
familiar with them (Heydarian et al., 2019) and may be willing to use them for longer periods of

59
time compared to more intrusive devices. Although early studies experimented with less familiar,
often not off-the-shelf technologies (e.g., piezoelectric strain gauge sensors), the majority of
recent studies have opted for accelerometers and/or gyroscopes that are embedded within a wrist-
worn smartwatch (Bell et al., 2020). Furthermore, the combination of a wrist-worn smartwatch to
automatically detect eating and a mobile or wearable device to capture ground-truth eating has
been featured in a few studies published in just the past year (Bin Morshed et al., 2020;
Goldstein, Hoover, Evans, & Thomas, 2021; Sen, Subbaraju, Misra, Balan, & Lee, 2020). This
approach is becoming more commonplace and these type of devices offer advantages for the user
(participant) and make the use of mHealth technologies more accessible to non-engineering
behavioral researchers. However, a number of related challenges have emerged. Future research
will need to address comparability between newer technology-assisted measures vs. more
traditional self-report measures of eating (Fowler et al., 2021) and vs. other similar technology-
assisted measures (Bell et al., 2020).
These user-friendly technologies also allow for the passive measurement or low-effort
reporting of various contexts and environments with relative ease. For example, fine-grained
real-time GPS data can be scraped from both mobile devices and smartwatches to determine an
individual’s location, and potentially assess the external influences on behavior (Cetateanu &
Jones, 2016; Yang, Wang, Nakandala, Kumar, & Jankowska, 2019). Similarly, social
environment can be gleaned from wearable cameras (Gemming et al., 2015a), self-report EMA
(O’Connor et al., 2019), or proximity Bluetooth sensors (Mundnich et al., 2020).
The ability to determine one’s context or environment is a necessary component of
ecological momentary interventions (Heron & Smyth, 2010) or just-in-time interventions
(Spruijt-Metz & Nilsen, 2014). These types of intervention designs aim to provide the right

60
amount of support at the right time and in the right contexts to promote behavior change
(Nahum-Shani, Hekler, & Spruijt-Metz, 2015; Nahum-Shani et al., 2017; Spruijt-Metz & Nilsen,
2014). These types of designs are well-suited for and offer unique opportunities for family-based
settings (Heron, Miadich, Everhart, & Smyth, 2019). They offer the ability to intervene on
children and adolescents and/or can be designed to target the behavior of multiple family
members at once (Heron et al., 2019). Because family members have shared genetic,
environmental, and behavioral risks, family units are especially important targets for intervention
and prevention (Kral & Rauh, 2010) and have the potential to halt the intergenerational
transmission of obesity and other chronic diseases.
CONCLUSION
This paper demonstrates that EMA is a feasible tool to collect ground-truth eating activity
and thus evaluate the performance of wearable sensors in the field. The combination of a wrist-
worn smartwatch to automatically detect eating and a mobile or wearable device to capture
ground-truth eating devices offers key advantages for the user (participant) and make the use of
mHealth technologies more accessible to non-engineering behavioral researchers.

61
TABLES
Table 1. M2FED Ecological Momentary Assessment (EMA) items

Variable
(subscale)
Item(s) Response
options
Format
Time-Triggered EMA
Eating
ground truth
Did you eat within the last hour? Yes
No

Positive and
negative
affect
How were you feeling right before
the phone signal went off? [Upset,
Nervous, Stressed, Couldn’t Cope,
Happy, Great, Cheerful, Joyful]
Not at all
A little
Some
Very
Separate screen for
each item
Eating Event-Triggered EMA
Eating
confirmation
Were you eating or drinking just
now?
Yes
No

Eating type What did you just eat? Meal
Snack
Drink only

Social
context
Who was eating with you? (Check all
that apply)
Nobody
Spouse/partner
Child(ren)
Mother
Father
Sister(s)
Brother(s)
Grandparent
Other family
Friend(s)
Other people

Eating in the
absence of
hunger -
started
eating
I started eating because [Food
looked, tasted, or smelled so good;
Others were eating; Feeling sad or
depressed; Feeling bored; Feeling
angry or frustrated; Feeling tired;
Feeling anxious or nervous; My
family or parents wanted me to eat].
Not at all
A little
Some
Very
Separate screen for
each item

62
Eating in the
absence of
hunger -
kept eating
I kept eating because [Food looked,
tasted, or smelled so good; Others
were eating; Feeling sad or
depressed; Feeling bored; Feeling
angry or frustrated; Feeling tired;
Feeling anxious or nervous; I wanted
to finish the food on my plate].
Not at all
A little
Some
Very
Separate screen for
each item
Hunger
level prior to
eating
How hungry were you right before
you ate?
0 = Not at all
hungry
100 = Greatest
imaginable
hunger
Sliding scale 0 to 100
Satiation
level after
eating
How full were you right after you
ate?
0 = Not at all
full
100 = Greatest
imaginable
fullness
Sliding scale 0 to 100
Mindful
eating
Before the beep, while I was eating
[My thoughts were wandering while
I ate; I was thinking about things I
need to do while I ate; I ate so
quickly that I didn’t taste anything].
Very true
Somewhat true
A little true
Not true
Separate screen for
each item
Positive and
negative
affect
How were you feeling right before
the phone signal went off? [Upset,
Nervous, Stressed, Couldn’t Cope,
Happy, Great, Cheerful, Joyful]
Not at all
A little
Some
Very
Separate screen for
each item

63
Table 2. Comparison of recruited sample and analytic samples

Characteristic
Recruited
Sample (N=74)
Analytic
Sample for
Study 1 (Aim
1a and 1b) and
Study 3 (n=58)
Analytic sample
for Study 1
(Aim 2a) and
Study 2 (n=46)
Analytic sample
for Study 1
(Aim 2b) (n=36)
Age (years) 28.91 (15.79) 27.45 (15.23) 28.76 (15.51) 26.67 (14.83)
Gender, Female
(%)
37 (50.0%) 30 (51.7%) 24 (52.2%) 20 (55.6%)
Parent, yes (%) 31 (41.9%) 24 (41.4%) 21 (45.7%) 15 (41.7%)
Note: P-values indicate a significant difference compared to the recruited sample (N=74).
*
p<0.05

64
Table 3. Individual-level characteristics of M2FED analytic sample (n=58), by family
member role (Dissertation Studies 1 and 3)

Note: The percentages

presented are column percentages.
1
n=53

Characteristic Child (n=34) Parent (n=24)
M (SD) / n (%)
Age (years) 15.74 (5.30) 44.04 (6.65)
Gender, Female (%) 13 (38.2%) 17 (70.8%)
Race/Ethnicity (%)

Asian or Pacific Islander 1 (2.9%) 1 (4.2%)
Black or African-American 2 (5.9%) 1 (4.2%)
Hispanic or Latino 21 (61.8%) 16 (66.7%)
White 4 (11.8%) 4 (16.7%)
Mixed 6 (17.6%) 1 (4.2%)
Other 0 (0.0%) 1 (4.2%)
BMI Percentile
1
22.56 (4.72) 33.10 (7.49)

65
Table 4. Family-level and deployment-level characteristics of M2FED study families (N=20)

Characteristic Family (N=20)
M (SD) / N (%)
Number of Parents Living in
Home (%)

1-Parent Household 3 (15.0%)
2-Parent Household 17 (85.0%)
Number of Children Living in
the Home (%)

1 Child 3 (15.0%)
2 Children 15 (75.0%)
3 Children 1 (5.0%)
4 Children 1 (5.0%)
Deployment length (days) 14.90 (3.13)

66
Table 5. EMA summary statistics after applying participation algorithm, by prompt type

Total
EMAs
Received
EMAs Received, Per
Family
EMAs Received, Per
Person
EMAs Received, Per
Person, Per Day
M (SD)
[Min,
Max]
M (SD)
[Min,
Max]
M (SD)
[Min,
Max]
All EMA 4,171
209.0
(89.4)
[86, 391]
71.9
(34.3)
[8, 176] 5.2 (2.7)
[0.6,
11.7]
Time-
Triggered
EMA
3,710
186.0
(84.3)
[77, 356]
64.0
(31.3)
[8, 147] 4.7 (2.4)
[0.3,
10.2]
Eating
Event-
Triggered
EMA
461
23.0
(17.2)
[3, 69] 8.0 (8.9) [0, 40] 0.6 (0.6) [0, 2.7]

67
Table 6. EMA compliance rates after applying participation algorithm, by prompt type

Total
EMAs
Received
Total
EMAs
Answered
Overall
Compliance
Family-Level
Compliance
Individual-Level
Compliance
M (SD)
[Min,
Max]
M (SD)
[Min,
Max]
All EMA 4,171 3,723 89.3%
89.4%
(5.74%)
[75.7%,
98.1%]
89.6%
(9.50%)
[53.8%,
100%]
Time-
Triggered
EMA
3,710 3,328 89.7%
89.8%
(5.84%)
[75.8%,
98.7%]
89.5%
(10.1%)
[50.0%,
100%]
Eating
Event-
Triggered
EMA
461 395 85.7%
85.9%
(14.3%)
[55.6%,
100%]
88.0%
(17.5%)
[28.6%,
100%]

68
Table 7. Logistic regression model results, examining predictors of compliance

All EMA Time-Triggered EMA
Eating Event-
Triggered EMA

β
(SE)
OR
(CI)
β
(SE)
OR
(CI)
β
(SE)
OR
(CI)
(Intercept) 2.17
***
8.75 2.22
***
9.22 2.41
**
11.15

(0.27) (5.20, 14.82) (0.29) (5.24, 16.36) (0.74) (2.65, 48.64)
Age (years) 0.00 1.00 0.00 1.00 0.02 1.02

(0.01) (0.98, 1.03) (0.01) (0.98, 1.02) (0.03) (0.96, 1.08)
Afternoon -0.47
**
0.63 -0.51
**
0.60 -0.35 0.71

(0.16) (0.46, 0.86) (0.18) (0.42, 0.85) (0.38) (0.33, 1.46)
Evening -0.50
**
0.61 -0.63
***
0.53 0.28 1.32

(0.15) (0.45, 0.81) (0.17) (0.38, 0.74) (0.38) (0.62, 2.75)
Weekend,
yes
0.06 1.06 -0.06 0.95 0.87
*
2.40

(0.11) (0.86, 1.31) (0.12) (0.75, 1.19) (0.35) (1.25, 4.91)
Deployment
day
-0.02 0.98 -0.01 0.99 -0.09
**
0.92

(0.01) (0.96, 1.01) (0.01) (0.97, 1.01) (0.03) (0.86, 0.97)
Female, yes 0.19 1.21 0.31 1.37 -0.65 0.52

(0.15) (0.90, 1.65) (0.17) (0.98, 1.92) (0.43) (0.22, 1.22)
Father -0.01 0.99 0.06 1.06 -0.65 0.52

(0.34) (0.51, 1.93) (0.36) (0.53, 2.16) (1.07) (0.06, 4.56)
Mother -0.42 0.66 -0.37 0.69 -0.64 0.53

(0.35) (0.33, 1.30) (0.38) (0.33, 1.47) (0.93) (0.08, 3.26)
Others
answered,
yes
0.65
***
1.91 0.73
***
2.07 -0.02 0.99

(0.10) (1.56, 2.34) (0.11) (1.66, 2.58) (0.30) (0.54, 1.76)
AIC 2805.16 2417.32 375.57
BIC 2868.52 2479.50 416.91
Num. obs. 4171 3710 461
a
OR=Odds ratio
***
p<0.001,
**
p<0.01,
*
p<0.05

69
FIGURES
Figure 1. Overview of M2FED cyberphysical system

Bluetooth proximity
beacons
Sensors
Smartwatches
Smartphones
Base
Station
Wi-Fi router
Laptop
(M2FED Controller)
Sensor data &
EMA data
Sensor data
Ecological
Momentary
Assessment
Cloud
Computing
& Storage
Amazon Web
Services
Monitor
Dashboard

70
Figure 2. Examples of a (a) time-triggered and (b) eating event-triggered mobile
questionnaire received on a participant’s phone

(a) Time-triggered EMA

This figure (a) is an example of a time-triggered mobile questionnaire that the
participants received on their phone during the study. It contains the first four questions of the
questionnaire that measure negative affect.

How were you feeling right
before the phone signal
went off?
Upset
o Very true
o Somewhat true
o A little true
o Not true
Nervous
o Very true
o Somewhat true
o A little true
o Not true
Stressed
o Very true
o Somewhat true
o A little true
o Not true
Couldn ’t Cope
o Very true
o Somewhat true
o A little true
o Not true
Next >>

71
(b) Eating event-triggered EMA

This figure (b) is an example of a eating event-triggered mobile questionnaire that the
participants received on their phone during the study. It contains the first question of the
questionnaire that measures whether the participant had just eaten or drank.

Were you eating or drinking
just now?
o Yes
o No
Next >>

72
Figure 3. Eating event-triggered EMA question logic

*See Table 1 for full list of questions and response options

Were you eating or
drinking just right
now?
Yes
Are you done eating?
Yes
What did you just
eat?
Who was eating with
you?
Eating in the absence
of hunger questions
Hunger/satiety
questions
Mindful eating
questions
Positive/negative
affect questions
End of Survey
No
Please tell us when
you are done by
pressing the "Done"
button on your screen
User presses "DONE"
button
No
Could you tell me
what you were just
doing?
Positive/negative
affect questions
End of Survey

73
Figure 4. Decision tree to determine when study participants are participating at home

Was a EMA answered in
the past 15 minutes?
No
Are accelerometer data
available?
Yes
Do the accelerometer
data indicate movement?
Yes
Are beacon data
available?
Yes
Participant is classified as
"participating" for the
past 1 minute.
No
The participant is outside
the range of the bluetooth
beacon transmission. No
conclusion can be made.
No
The participant is not
wearing the watch or is
not moving. No
concluion can be made.
No
The watch is off, not
transmitting data, or the
participant is not wearing
the watch. No conlusion
can be made.
Yes
Participant is classified as
"participating" for past 15
minutes and next 15
minutes.

74
Figure 5. Example of ‘participation’ time intervals for a participant

In this example, the shaded grey regions indicate the valid ‘participation’ time intervals for this
participant. In the first interval, we see that the participant answered an EMA, and there were
available data from the accelerometer and beacon. In the second interval, the participant did not
answer an EMA, but there were available data from the accelerometer and beacon. In the third
interval, the participant answered an EMA and there were some available data from the
accelerometer.

Time
EMA
Accelerometer
Data Streams:
Beacon
At-Home
Interval #1
At-Home
Interval #2
At-Home
Interval #3

75
Figure 6. Flow of participants in the M2FED study

Recruited sample (N=74)
n=13 dropped out of study
or were removed by
'participation' algorithm
n=3 non-parent adult
participants were removed
Analytic sample for Study
1 (Aim 1a and 1b) and
Study 3 (n=58)
n=12 did not receive and/or
answer at least 1 eating
event-triggered EMA
Analytic sample for Study
1 (Aim 2a) and Study 2
(n=46)
n=10 did not receive and/or
answer at least three eating
event-triggered EMA
Analytic sample for Study
1 (Aim 2b) (n=36)

76
Figure 7. Distribution of EMAs received across the time of day (hour), by EMA survey type

77
Figure 8. Distribution of EMAs received, by family role, by EMA survey type

78
Figure 9. Average (a) family-level and (b) individual-level compliance

(a)

(b)

79
CHAPTER 3: UTILIZING WEARABLE SENSORS AND ECOLOGICAL
MOMENTARY ASSESSMENT (EMA) TO AUTOMATICALLY DETECT
CONTEXTUALIZED EATING EVENTS IN THE HOME: A MULTI-LEVEL LATENT
CLASS ANALYSIS
ABSTRACT
Introduction: Eating behaviors and patterns and the various contexts in which they occur (i.e.,
individual contexts, social contexts) have been shown to play a significant role in the
development of obesity and other chronic diseases. Emerging technologies offer the ability to
investigate the role of social context in naturalistic settings, and how social context interacts with
intra-personal context to influence eating behavior together in real-time. The purpose of this
paper is to identify latent classes of real-time eating events (“contextualized eating events”) that
were detected via automatic eating detection software and Ecological Momentary Assessment
(EMA) in family homes.
Methods: The Monitoring and Modeling Family Eating Dynamics (M2FED) system measured
in-home family eating behavior using a suite of wearable sensors and mobile devices. In a
sample of 20 families, eating events were automatically detected in real-time with a wrist-worn
smartwatch. Eating event-triggered EMA was deployed to collect ground-truth eating activity as
well as momentary intra-personal contexts, including hunger/satiety, negative affect, mindful
eating, eating in the absence of hunger, and inter-personal contexts (i.e., eating companionship).
Multi-level latent class analysis was used to identify latent classes of eating events using intra-
and inter-personal features characterized by these seven indicators.
Results: A 4-class solution was produced by the latent class analysis. Class 1 (18.6%) was
characterized by high levels of negative affect, hunger, and eating in the absence of hunger. The

80
eating events in this class were significantly more likely to be reported as a snack (vs. meal).
Class 2 (18.3%) was primarily characterized by a higher rate of mindfulness. Class 3 (31.7%)
was characterized by a high rate of satiety and lower rates of eating in the absence of hunger.
Lastly, Class 4 (31.4%) was primarily characterized by a high level of negative affect and lower
rates of satiety and eating in the absence of hunger.
Conclusion: Results from this study identify the intra- and inter-personal contexts in which
various eating events and activities co-occur. Findings can inform future real-time interventions
on the optimal contexts during which promoting healthy eating behaviors may be most necessary
and effective. Furthermore, these findings indicate the feasibility of using latent class analysis to
categorize event-level data. As intensive longitudinal data become more prominent in behavioral
research, this “event-centric” approach can address a variety of research questions at the
intersection of behavior and contexts.
INTRODUCTION
Global Obesity Epidemic
In the past few decades, there has been a substantial upward trend of overweight and
obesity prevalence in youth and adults, both in the United States (U.S.) and globally (Han et al.,
2010; Ng et al., 2014; Skinner et al., 2018; The GBD 2015 Obesity Collaborators, 2017), thus
making the global obesity epidemic a major public health concern (World Health Organization,
2011). The obesity prevalence for U.S. youth is 18.5%, while the prevalence for U.S. adults is
39.6%. Moreover, large racial disparities in obesity prevalence exist in the U.S. – on average,
25.8% of Hispanic and 22.0% of Black children and adolescents have obesity, compared to

81
14.1% of white youth; and 47.0% of Hispanic and 46.8% of Black adults have obesity, compared
to 37.9% of white adults (Hales et al., 2017).
Children who experience overweight and obesity are at an increased risk of remaining
overweight into adulthood (A. S. Singh et al., 2008). Additionally, a family history of obesity
strongly predicts childhood and adult obesity (also referred to as the “intergenerational
transmission of obesity”) due to family members’ shared genetic, environmental, and behavioral
risks (Kral & Rauh, 2010). Consequently, both children and adults who have obesity are at an
increased risk of developing related adverse health outcomes, including diabetes, stroke,
cardiovascular disease, hypertension, certain cancers, and premature mortality (Lauby-Secretan
et al., 2016; Reilly & Kelly, 2011). The association between obesity and major health risks for
both children and adults highlights the urgent need for effective overweight and obesity
prevention and intervention efforts in order to improve population-level health.
Dietary Intake, Eating Behavior, and Health
The role of dietary intake (i.e., what and how much is consumed) on human health and
disease such as obesity has been extensively studied (Willett, 2013), with findings that indicate
dietary intake is a critical component of chronic disease prevention (Nishida et al., 2004). For
instance, intake of certain types of foods, such as fast food (Rosenheck, 2008), sugar-sweetened
beverages (Malik, Schulze, & Hu, 2006; Ruanpeng, Thongprayoon, Cheungpasitporn, &
Harindhanavudhi, 2017), and ultra-processed foods (Askari, Heshmati, Shahinfar, Tripathi, &
Daneshzad, 2020) have been linked to increased obesity risk in children and adults. It is
estimated that poor dietary factors (i.e., over-consumption of salt, under-consumption of fruits,
vegetables, and whole grains) contributed to 11 million deaths globally in 2017 (Afshin et al.,
2019).

82
A large portion of the nutritional epidemiology field has looked exclusively at dietary
intake (i.e.,, intake of certain types of foods). Yet, eating behaviors and patterns (i.e., food
choices and motives, feeding practices) and the various contexts in which they occur (i.e.,
individual contexts, social contexts, physical contexts, etc.) have also been shown to play a
significant role in the development of obesity and other chronic diseases (e.g., type 2 diabetes,
heart disease) (Higgs & Thomas, 2016; Jannasch et al., 2017; Neuhouser, 2018; Reicks et al.,
2015; Robinson et al., 2014; Tourlouki et al., 2009). Eating behaviors and contexts have received
less focus until recently, and may provide a fuller understanding of the relationship between
eating and health and serve as points of intervention to improve health outcomes. For example,
research has shown social contexts like eating with family (Neumark-Sztainer et al., 2004; Suggs
et al., 2018; Videon & Manning, 2003), physical contexts like living in a rural food environment
(Lenardson et al., 2015), psychological contexts like chronic stress levels (Isasi et al., 2015;
Torres & Nowson, 2007), and timing of eating (Wilkinson et al., 2020) all have important
impacts on eating patterns and health outcomes.
Despite this strong association between eating behavior, health outcomes, and
morbidity/mortality, our ability to assess eating behaviors and the contexts in which they occur
accurately and in real-time is an existing challenge due to self-reported and retrospective
assessment methods. These commonly used eating assessment methods include food records
(food diaries), food frequency questionnaires (FFQ), and 24-hour dietary recalls, which typically
summarize information at the day-, week-, or even year-level (Magarey et al., 2011; Shim et al.,
2014; Willett, 1998), can have high respondent burden, are prone to self-report biases (e.g.,
social desirability bias, recall/memory bias) (Althubaiti, 2016; Livingstone et al., 2004;

83
Westerterp & Goris, 2002), and typically do not collect information on potentially relevant
contexts (e.g., individual, social, physical, etc.).
As a result, the validity of data collected via self-report methods and subsequent research
findings can come under question. These biases may result in the under- and over-reporting of
dietary intake, thereby giving an inaccurate picture of dietary intake patterns, potentially skewing
research findings, and providing inadequate scientific conclusions (Archer & Blair, 2015;
Schoeller, 1995; Schoeller et al., 2013).
Emerging Technologies
Emerging technologies can improve these traditional assessment methods by using
innovations that improve the quality and validity of data that are collected. Technologies, such as
wearable sensors, mobile phones, and cameras, offer many advantages by measuring eating
behavior at the time and location the behavior is actually occurring, thereby reducing recall bias,
increasing ecological validity, and collecting potentially-relevant contextual information
(Gemming et al., 2015a; Spruijt-Metz et al., 2018). Additionally, technologies offer the ability to
passively capture frequent measurements of the same variable over long periods of time, thus
producing temporally dense data on both the behavior and the context and improving our ability
to examine how they vary over time.
Emerging evidence suggests that momentary (i.e., near or at the moment the behavior is
occurring) contextual states (i.e., intra-personal states such as stress, inter-personal states such as
eating companionship) play an important role in eating behavior (Anestis et al., 2010; Elliston,
Ferguson, Schüz, & Schüz, 2017; Engel et al., 2007; Goldschmidt et al., 2012; Goldschmidt et
al., 2018; Liao et al., 2018; Mason et al., 2019; Smith et al., 2018; Smyth et al., 2007). Emerging
technologies present the opportunity to examine these momentary eating behavior and

84
contextual factors in real-time, and to apply new analytic methods for modeling the multiple
signals from these different contextual states.
Intra-personal (Individual) Context
A number of studies have used Ecological Momentary Assessment (EMA), a data
collection technique in which one’s self-reported behaviors and experiences are repeatedly
sampled in the natural environment (Shiffman et al., 2008), to assess the relationship between
momentary intra-personal context (e.g., negative affect, stress) and eating behavior (Engel et al.,
2016). Studies have found that momentary anger, negative affect, and stress are significantly
associated with binge eating episodes in women with eating disorders (Anestis et al., 2010; Engel
et al., 2007; Mason et al., 2018); and fruit and vegetable consumption in the preceding 2 hours is
associated with happiness in women (Liao et al., 2018). Many of these studies have been
conducted in adult women, typically with an eating disorder; the association between momentary
intra-personal context and eating behavior has yet to be explored in more diverse populations.
Other intra-personal contextual states, such as hunger level, level of eating in the absence
of hunger, and level of mindful eating, have been established as associates of eating behavior in
cross-sectional studies (Beshara, Hutchinson, & Wilson, 2013; Fogel et al., 2018; Gilbert &
Waltz, 2010; Lansigan, Emond, & Gilbert-Diamond, 2015; Lavender, Gratz, & Tull, 2011), but
are underexplored as potentially important momentary contextual states.
Inter-personal (Social) Context
The study of social influences on eating behavior has rapidly expanded in the past
decade. The evidence reviewed above indicates that eating behavior is influenced by intra-
personal factors such as internal motivations and hunger levels, but the types and the quantities

85
of food that one eats cannot be entirely explained by these intra-personal factors. Social context
may play an especially important role in eating behavior because eating is an inherently social
practice (Delormier et al., 2009) – the majority of eating and drinking episodes take place with
others (i.e., not alone) (Oh et al., 2014) and shared meals have been and continue to be the
centerpieces of many traditions, celebrations, and other cultural occasions around the world.
Social contexts can have a strong impact on food intake through several mechanisms,
including mindless imitation of others’ eating behaviors (Bell et al., 2019; Hermans et al., 2012;
Sharps et al., 2015), food matching and modeling (Salvy et al., 2012), and situational and
cultural food norms (Herman & Polivy, 2005). Relatedly, eating with others leads us to eat more
than we normally would have without others (social facilitation) – and this effect becomes
stronger the as the number of people we eat with grows (Herman et al., 2019).
Strong evidence suggests that a particular type of social influence – family networks and
systems – can impact eating behaviors and, subsequently, obesity risk. Studies have found that
some family features are predictive of children’s and parents’ dietary intake quality, including
parental behaviors, attitudes, and feeding styles (Lytle et al., 2011; Savage et al., 2007; Vollmer
& Mobley, 2013; Yee et al., 2017); certain family features and structures (e.g., parents’ level of
education and dual-parent families, respectively) (Patrick & Nicklas, 2005; Pearson et al., 2010);
and family meal features (e.g., turning off the television during a meal, longer meal duration)
(Dallacker et al., 2019). There are some family features, such as parental stress, that have been
associated with poorer dietary quality (Bauer et al., 2012; Parks et al., 2012), whereas other
features, such as higher frequency of family meals, have been associated with better dietary
quality (Robson, McCullough, Rex, Munafò, & Taylor, 2020). These findings indicate that some

86
family features may be able to serve as points of intervention to promote healthier eating
behaviors within families.
Many of the studies examining the influence of family context on eating behaviors,
however, (i) have not examined multi-level features of family members or their relationships; (ii)
have relied on cross-sectional data that only provide brief snapshots of context and behaviors,
thereby introducing recall bias; (iii) have primarily taken place in laboratory settings (vs. in-field
settings), thereby threatening ecological validity of the findings (Herman et al., 2019); and/or (iv)
have examined important features and contexts independent of one another. With the
emergence of data collection methods utilizing various technologies, we can begin to
examine the role of social context in naturalistic settings, and how social context interacts
with intra-personal context to influence eating behavior together in real-time.
Latent Class Analysis
As reviewed above, many intra-personal states and features of family contexts have been
separately linked to dietary intake and quality. A common approach to analyzing these types of
data has been to use multi-level regression models (e.g., to examine whether emotions relate to
eating behaviors) or similar methodological approaches that could be considered “variable
centric” (i.e., examining which variables are associated with eating).
However, when examining these intra-personal features in isolation of inter-personal
features, we are unable to examine how these two types of features co-occur in specific moments
and during specific eating events. We posit that both intra- personal features (affective states,
hunger/satiety, etc.) and inter-personal features (eating with others) could be important features
of the family eating system.

87
We use Latent Class Analysis (LCA) to understand the interdependence among these
features during eating events to ultimately understand their role in the broader family eating
system. LCA is a statistical method for identifying unobservable, or latent, subgroups within a
population (or, in this case, the unit of analysis will be eating events) (Goodman, 1974;
Lazarsfeld & Henry, 1968). This “event centric” approach (i.e., distinguishing different
contextual features that characterize types of eating events) provides us with a richer
understanding of eating events when informed by both intra- and inter-personal features.
Study Aims
Aim 1: Identify and characterize the latent classes of participants’ eating events
automatically detected in the home, using self-reported data on an individual’s hunger, satiety,
positive and negative affect, mindful eating, eating in the absence of hunger, and eating
companionship (social context), collected immediately following an eating event via EMA.
Aim 2: Examine eating event-level predictors of the latent classes of eating events.
METHODS
Participants and Recruitment
Eligibility. Families were eligible to enroll in the Monitoring and Modeling Family
Eating Dynamics (M2FED) study if they met the following inclusion criteria:
i. consisted of at least two related family members;
ii. included at least one adult parent and one child between the ages of 11 and 18 years old;
and
iii. lived within Los Angeles County.

88
Children who were aged 10 or younger were ineligible to participate, but families with a
child or children in that age range were still eligible to enroll (excluding the ineligible child or
children). Exclusion criteria for the study included: families with one or more members living in
the home who did not primarily speak English. There were no demographic- or disease-related
exclusion criteria.
Family members that were eligible to participate in the study will be referred to as
‘participant’ hereafter.
Method of recruitment. Our recruitment team visited public spaces, such as train and bus
stops, and public events, such as football games and book fairs, within Los Angeles County from
May 2017 to August 2019 to recruit families for the M2FED study. Flyers containing the study’s
website information and the research team’s contact information were handed out at these
locations. Additionally, when any family successfully completed the study, they were offered an
additional $20 if they referred other eligible families that successfully enrolled in the study (i.e.,
snowball sampling).
Any family that expressed interest in the study and met the eligibility requirements were
invited to participate. Once one or both parents agreed that the family would participate, the
recruitment staff administered an intake screening tool over the phone before officially enrolling
the family in the study.
This study was approved by the Institutional Review Board of the University of Southern
California. All parents provided informed written consent, and all children provided assent.
M2FED System Overview
The M2FED cyberphysical system (Figure 1) is a novel system for measuring and
monitoring in-home family eating behavior. Broadly, cyberphysical systems can be defined as

89
“physical and engineered systems whose operations are monitored, coordinated, controlled and
integrated by a computing and communication core” (Rajkumar et al., 2010). The primary
components of the M2FED system include sensors (smartphones, smartwatches, Bluetooth
proximity beacons), a base station and Wi-Fi router, an EMA subsystem, and a remote
monitoring subsystem. The technical details of this system are provided elsewhere (Mondol et
al., 2020). Because the scope of the project was to examine in-home eating behavior, all data
collected by the system were measured in the home (i.e., no data were collected outside of
the home).
Eating event detection algorithms. Wrist-worn smartwatches (Sony Smartwatch 3,
Android Wear operating system) were used to detect in real-time when eating events were
occurring in the home for each participating family member. Data on eating events that occurred
outside of the home were not collected. Inertial sensors inside the smartwatch (accelerometer and
gyroscope) provided motion data that were used to detect arm movements and hand gestures
associated with eating (hand-to-mouth gestures). If a set of hand-to-mouth gestures were
detected, then the data were processed with a more sophisticated algorithm to automatically
detect an eating event (Mondol et al., 2020).
An eating event can be defined as a set of hand-to-mouth gestures, representing
phenomenon such as consuming a meal, a snack, a drink or a combination of these consumption
behaviors in which hand-to-mouth gestures are clustered temporally. Further details on the eating
event detection algorithm can be found here (Mondol et al., 2020).
EMA protocol. Once an eating event in the home was detected, an eating event-
triggered mobile questionnaire immediately was sent to the corresponding family member’s
smartphone (Samsung Galaxy S7, Android operating system) asking to confirm whether or not

90
they were just eating (Figure 3). They were also asked a battery of survey items that included
validated measures of hunger, satiety, mindful eating, and negative affect; and with whom they
were eating, if anyone. If the participant had not finished eating, then they were given the option
to request more time before filling out the questionnaire.
If an eating event was not detected within an hour, then an hourly time-triggered mobile
questionnaire was sent to the smartphone, which included a brief validated negative affect
survey. Figure 10 illustrates the sampling protocols for time-triggered and eating event-triggered
EMAs in detail.
To manage participant burden, the system was set to send a mobile questionnaire to
participants no more than once per hour. If they did not answer the first prompt within three
minutes, then a reminder was sent to the smartphone. A total of three reminders were sent if their
questionnaire was not answered; the questionnaire timed out after 15 minutes had passed since
the initial questionnaire had been sent. Also, EMAs were only sent to participants during their
“personalized participation window,” the period in which they indicated they normally woke up
and went to bed.
Monitoring system. A remoting monitoring system monitored the operation and
connectivity of all technological components of the system (e.g., base station, smartwatches,
smartphones, etc.) in real-time (Ma et al., 2017). It allowed the research team to ensure that the
system was running properly throughout the in-home deployment.
Procedures
After a family officially enrolled in the study, a date and time was confirmed for two
members of the research team to visit the home. There were a total of two home visits.

91
Visit 1. During the first home visit, the research team provided consent forms to all
participants and obtained informed consent and assent before proceeding. Then, the team
measured all participants’ body measurements (height, weight, body fat percentage) using a
Tanita scale and a stadiometer. While participants filled out baseline surveys, the research team
installed the various components of the M2FED system around the home (all “living spaces”; not
including bedrooms or bathrooms).
To ensure that the system would run without interference, the system’s base station and
Wi-Fi router were placed in discrete locations in the home. Similarly, Bluetooth beacons were
placed on the walls of all “living spaces” in discrete locations. Then, each participant was
provided with their own designated Samsung Galaxy S7 smartphone (Android operating system)
and Sony Smartwatch 3 (Android Wear operating system). For the smartwatch, all features were
turned off; and for the smartphone, all features except for answering questionnaires via the EMA
app were turned off. The research team labeled each participant’s phone and watch with their
name so they would know which devices were theirs. After the participants received their two
devices, they were instructed on how to properly wear, charge, and care for them. They also were
given a demonstration on how to answer an EMA on the smartphone.
The research team instructed the participants to not bring their designated phone and
watch outside of the home to prevent the devices from getting damaged or lost.
9
They were also
instructed to, while at home, wear the watch at all times and answer all EMA questionnaires that
their phone received.

9
Also, the M2FED system was not designed to collect data outside of the home. Because the
data were transferred via the Wi-Fi router that was installed in the home, once the devices were
outside of the reach of the router, they could not send or receive any data.

92
After the system was successfully set up in the home and all participants were equipped
with their devices, the family’s “deployment” began and lasted for approximately fourteen
consecutive days. During this period, participants used their smartphones to complete hourly
time-triggered and eating event-triggered mobile questionnaires (up to once every hour during
waking hours); and eating event monitoring in the form of a wrist-worn smartwatch during
waking hours.
Visit 2. After the family’s deployment ended approximately two weeks later, the research
team visited the home one final time in order to terminate the data collection. They uninstalled
all of the equipment and removed it from the home. All participating family members were given
a $100 Visa gift card as compensation for the two-week study.
Measures
Eating events. During the two-week assessment period, participants were asked to wear
their dedicated smartwatch on their dominant wrist at all times while they were home during
waking hours. Automatic eating event detection software on the watches collected timestamps
(approximate start and end times, format: mm/dd/yyyy, hh:mm:ss) for all detected eating events
that occurred while the watch was worn. After an eating event was detected, participants
received a brief mobile questionnaire on their study phones to confirm whether the detected
eating event was a true event. The first question on the questionnaire was “Were you eating or
drinking just now?” If the participant responded “No,” then they were asked to report what they
were doing. Options included “Using my phone,” “Smoking,” “Fixing my hair,” “Putting on
sunscreen or lotion,” or “Other” with an open-text field. If the participant responded “Yes,” then
they were asked to report on the following momentary measures:

93
Hunger and satiety. To measure perceived hunger, participants were asked to rate how
hungry they were right before they ate, on a scale ranging from 0 (“Not hungry at all”) to 100
(“Greatest imaginable hunger”). To measure perceived satiety (i.e., feeling full after a meal),
participants were asked to rate how full they were right after they ate, on a scale ranging from 0
(“Not full at all”) to 100 (“Greatest imaginable fullness”). These measures come from the Satiety
Labeled Intensity Magnitude (SLIM) scale, which was developed for the purpose of measuring
responses over time with the same individual, and is a sensitive and reliable scale for measuring
perceived hunger/satiety (Cardello et al., 2005).
For the current analyses, the hunger scores were categorized into a binary variable the
following way: if any hunger score was greater than the median hunger score, it was coded as
hunger=1. Otherwise, any hunger score that was less than or equal to the median hunger score
was coded as hunger=0. The same approach was taken for the satiety scores.
Negative affect. Four items from three different validated scales were used to measure
negative affect via EMA: the Positive and Negative Affect Schedule for children (PANAS-c)
(Laurent et al., 1999); the Perceived Stress Scale (PSS) (Cohen et al., 1983); and the Profile of
Mood States-Adolescents (POMS-A) (Terry et al., 1999). Participants were asked to rate how
they were feeling right before they received the EMA mobile questionnaire on the following four
negative affective states: (1) Upset (from PANAS-c, PSS), (2) Nervous (from PANAS-c, PSS,
POMS-A), (3) Stressed (from PSS), and (4) Couldn’t cope (from PSS). Participants could choose
from the following ratings for each item: 1 = “Not at all,” 2 = “A little,” 3 = “Some,” and 4 =
“Very.” Mean scores of the four negative affect (NA) items were calculated, with a possible
range of 1.00 to 4.00 (lower average score indicates low NA, higher mean score indicates high
NA).

94
For the current analyses, because the distribution of the mean NA scores was highly right
skewed, negative affect was categorized into a binary variable the following way: if any mean
score was greater than the minimum (1.00), it was coded as negative affect=1. Otherwise, any
mean score that was the minimum (i.e., a participant answered “1” for all four NA items) was
coded as negative affect=0.
Mindful eating. Mindful eating is defined as the “nonjudgement awareness of physical
and emotional sensations while eating.” (Framson et al., 2009) To measure mindful eating,
participants were asked to report on their agreement with the three following items: (1) “My
thoughts were wandering while I ate,” (2) “I was thinking about things I need to do while I ate,”
and (3) “I ate so quickly that I didn’t taste anything.” Response options were: 1 = “Very true,” 2
= “Somewhat true,” 3 = “A little true,” 4 = “Not true.” An average mindful eating score was
calculated, with a possible range of 1.00 to 4.00 (lower average score indicates low mindful
eating, higher average score indicates high mindful eating).
The three items were taken from the Mindful Eating Questionnaire (MEQ), which
originally contains 28 items and spans five domains (disinhibition, awareness, external cues,
emotional response, and distraction) (Framson et al., 2009). In order to keep the mobile
questionnaire short, only the three items from the “distraction” domain were used. Of note, the
MEQ was validated in a sample of primarily white women (Framson et al., 2009).
For the current analyses, the mean mindfulness scores were categorized into a binary
variable the following way: if any mean mindfulness score was greater than the median mean
mindfulness score, it was coded as mindfulness=1. Otherwise, any mean mindfulness score that
was less than or equal to the median mean mindfulness score was coded as mindfulness=0.

95
Eating in the absence of hunger. The 14-item Eating in the Absence of Hunger
Questionnaire for Children and Adolescents (EAH-c) was used to measure participants’ level of
eating in the absence of physiological hunger (Tanofsky-Kraff et al., 2008). Two additional items
were added in order to assess social influence on eating in the absence of hunger, resulting in 16
items total. Participants rated the extent to which the following eight environmental or internal
events led to them start eating: (1) “Food looked, tasted or smelled so good,” (2) “Others were
eating,” (3) “Feeling sad or depressed,” (4) “Feeling bored,” (5) “Feeling angry or frustrated,”
(6) “Feeling tired,” (7) “Feeling anxious or nervous,” and (8) “My family or parents wanted me
to eat” (added item). Participants then rated the extent to which those first seven events led them
to keep eating; they were also asked to rate one additional item: “I wanted to finish the food on
my plate” (added item). Response options were: 1 = “Not at all,” 2 = “A little,” 3 = “Some,” and
4 = “Very.” The original EAH-c provided options on a 5-point Likert scale, however we reduced
the scale to 4 points for further simplicity) (Tanofsky-Kraff et al., 2008).
Cumulative scores of eating in the absence of hunger (EAH) for starting to eat and
continuing to eat were separately calculated, each with a possible range of 8 to 32 (lower
cumulative score indicates low EAH, higher cumulative score indicates high EAH).
For the current analyses, because the distributions of the cumulative EAH scores were
highly right skewed, the two EAH scores were categorized into binary variables the following
way: if any cumulative score was greater than the minimum (8.00), it was coded as EAH=1.
Otherwise, any cumulative score that was the minimum (i.e., a participant answered “1” for all
eight EAH items) was coded as EAH=0.
Eating companionship (social context). Participants were asked to report who was eating
with them. They could select from one or more of the following options: “Nobody,”

96
“Spouse/partner,” “Child(ren),” “Mother,” “Father,” “Sister(s),” “Brother(s),” “Grandparent(s),”
“Other family,” “Friend(s),” and “Other people.”
For the current analyses, the response values were transformed into a dummy variable,
such that 1 = they were eating with at least one other person and 0 = they were eating alone.
Timing. The“time of day” at which an eating event occurred and “type of day” were
calculated using the stored date and timestamp of the detected eating events. Time of day at
which the eating event occurred was stored in hh:mm:ss format, and then transformed into three
binary variables: Morning=1 if the timestamp was between 00:00:00 and 11:59:59, Afternoon=1
if the timestamp was between 12:00:00 to 16:59:59, and Evening=1 if the timestamp was
between 17:00:00 to 23:59:59.
The ‘lubridate’ R package (Grolemund & Wickham, 2011) was used to convert the date
on which the eating event occurred (format: mm/dd/yyyy) to the type of day (weekend vs.
weekday). Subsequently, this was transformed into a “weekend status” dummy variable:
1=weekend and 0=weekday.
Anthropometrics. During Home Visit 1, height (cm), weight (lbs), and body fat
percentage (%) were measured in all participants in a private section of the home, using a
portable stadiometer and a research grade Tanita  scale (model TBF 300). Body mass index
(BMI) for participants 20+ years old was calculated from their measured height and weight
(kg/m
2
). For participants under the age of 20, age- and sex-specific BMI percentiles were using
the Centers for Disease Control and Prevention online BMI tool (CDC Division of Nutrition,
2018).
Demographics. During Home Visit 1, participants were asked to provide basic
demographic information via a paper questionnaire, including their current age (in years), gender

97
(Female, Male), race (Hispanic or Latino, Asian or Pacific Islander, White or Caucasian, Black
or African-American, American Indian or Native American, Mixed, Other), Hispanic/Latino
ethnicity (Yes, No, Don’t know), and family role (Mother, Father, Child, Grandparent, Aunt,
Uncle, Other).
Analytic Approach
Individual-level characteristics. The mean/standard deviation or the count/proportion of
the analytic sample’s age, body mass index, gender, and race/ethnicity were calculated and
reported overall and by family role (child, parent).
Eating event descriptives. The mean and standard deviation were calculated for the
continuous responses to the eating event-triggered EMA questions about negative affect, mindful
eating, eating in the absence of hunger (start eating), eating in the absence of hunger (kept
eating), hunger level before eating, and satiety level after eating. Counts and proportions were
calculated for the categorical responses to the EMA question about eating companionship
(yes/no) and system-defined time of day (Morning, Afternoon, Evening) and weekend status
(yes/no).
Data analyses
To identify latent classes of eating events using intra- and inter-personal features (Study
Aim 1), multi-level latent class analysis (MLCA) using the one-step approach was performed
to categorize the detected eating events into different types of commonly observed eating events
(latent classes) (Bolck, Croon, & Hagenaars, 2004). Latent class analysis (LCA) is a statistical
method for identifying unobservable, or latent, subgroups within a population (or, in this case,
the unit of analysis will be eating events) (Goodman, 1974; Lazarsfeld & Henry, 1968).

98
One of the assumptions of these latent class models is that the observations are
independent. It has been shown that if this assumption is ignored or violated, the dependent
nature of the data will bias the results (Park & Yu, 2015). The structure of the data that were
used in these analyses are multi-level in nature – eating events (Level 1) are nested in individuals
(Level 2) who are nested within families (Level 3). We were interested in identifying latent
classes at the eating event level (Level 1) and examining predictors of the latent classes also at
Level 1. We corrected the standard errors for the nesting of multiple individuals in families using
a cluster-robust (Huber-White) sandwich estimator
10
. The results produce a 2-level model and
the standard errors of the models appropriately correct for the nesting of individuals in families.
Seven indicators (measures) that represent features of each eating event, captured via
EMA immediately after the event, were used to inform latent class membership: (i) the
individual’s hunger level preceding the eating event (binary; 1=high, 0=low), (ii) satiety level
following the eating event (binary; 1=high, 0=low), (iii) negative affect (binary; 1=high, 0=low),
(iv) mindful eating (binary; 1=high, 0=low), (v) eating in the absence of hunger (start eating)
(binary; 1=high, 0=low), (vi) eating in the absence of hunger (kept eating) (binary; 1=high,
0=low), and (vii) eating companionship (binary; 1=eating with at least one other person,
0=eating alone).
Regarding the Level 1 unit of analysis, only the eating events that were detected by the
smartwatch algorithm and confirmed by the participant were included in the analyses.
Mplus (version 8.4) (Muthén & Muthén, 1998-2017) was used to perform the following
analytic approach:

10
In the Mplus software, this is implemented by using TYPE = COMPLEX in the ANALYSIS
command.

99
Multi-level models with an increasing number of latent classes (2-class model, 3-class
model, 4-class model, etc.) without covariates were tested, and multiple model fit indices,
including Bayesian information criterion (BIC), Akaike information criterion (AIC), and
likelihood ratio tests (LRT), were used to determine the adequate number of latent classes to be
included in the final model (Nylund, Asparouhov, & Muthén, 2007). A smaller BIC and AIC is
associated with a better model fit; and the LRT is used to test the improvement in the models as
more classes are added.
To test Study Aim 2, the one-step approach was used: this selected model was then
further analyzed with event-level predictors.
Eating event-level (Level 1) predictors included type of meal (1=Meal, 0=Snack), the
time of day the eating event took place (three dummy variables were created to indicate three
times including Morning, Afternoon, and Evening [reference group]), and the type of day the
eating event took place (1=Weekday, 0=Weekend).
Missing data
The latent class models were estimated with “full information maximum likelihood”
(FIML) parameter estimation. Observations with partially missing responses for any of the
variables that determine latent class membership (hunger, satiety, negative affect, mindfulness,
EAH (started eating), EAH (kept eating), and eating companionship) were not removed. Instead,
all available data points were used and missing data were imputed using FIML.
Any observations that were missing all responses to the variables that determine latent
class membership (hunger, satiety, negative affect, mindfulness, EAH (started eating), EAH
(kept eating), and eating companionship) were removed.

100
RESULTS
Sample Characteristics
A total of 74 participants from 20 families enrolled in the M2FED study. Thirteen
participants dropped out of the study or were removed from the dataset if their participation (as
determined by the participation algorithm) was 0% (i.e., they did not answer any EMAs nor ever
wore the smartwatch) (Figure 6). Additionally, the data from three non-parent adult participants
made up approximately 1.5% of the EMAs received, so these participants were removed from
the analytic sample, resulting in a remaining 58 participants. Of these 58, 46 participants
received and answered at least one eating event-triggered EMA. This subsample did not
significantly differ from the enrolled sample (N=74) by age, gender, or parent role (Table 2).
The final analytic sample for these analyses was 46 participants. 45.7% were parents
(n=21) and 54.3% were children (n=25). On average, children were 15.48 years old (SD=3.63
years) and parents were 44.57 years old (SD=6.53 years). There were 9 female children (36.0%
of children) and 15 female parents (71.4% of parents). 60.0% of children and 71.4% of parents
identified as Hispanic or Latino (Table 8).

Eating Event Descriptives
A total of 461 eating event-triggered EMA surveys were received by the participants. Of
those 461, 290 (62.9%) were (i) answered by participants, (ii) confirmed as true eating events
(i.e., participants responded “Yes” to the initial question “Were you eating or drinking just
now?”, and (iii) contained at least partial responses for the seven indicator variables that
determined latent class membership. Focusing on these 290 true eating events, 57 (19.66%) of
these took place in the morning, 98 (33.79%) in the afternoon, and 135 (46.55%) in the evening

101
(Table 9). About one-third (32.76%) of reported eating events took place on the weekend.
Participants indicated that they were eating with at least one other person during 186 (64.14%) of
the eating events.
The binary features linked to each of the 290 eating events are summarized in Table 9.
One-third (31.03%) of eating events involved negative affect. Almost half (n=141, 48.62%)
involved hunger before eating, 143 (49.31%) reported fullness after eating, 241 (83.10%)
reported eating in the absence of hunger when they started eating, and 234 (80.69%) reported
eating in the absence of hunger (kept eating). Less than half (n=127, 43.79%) involved mindful
eating.
Multi-Level Latent Class Analysis (Aim 1)
Multi-level models with one to six latent classes without covariates were tested (Table
10). None of the p-values from the LRTs were statistically significant (<0.05) for any of the six
class solutions. However, using other model fit indices, the 4-class solution provided the best fit.
This solution produced the lowest BIC value (2256.743) and AIC (2131.967). Entropy for this 4-
class solution was 0.943. The 4 classes were distributed as follows: 54 eating events (18.6%)
were assigned to Class 1, 53 eating events (18.3%) to Class 2, 92 eating events (31.7%) to Class
3, and 91 eating events (31.4%) to Class 4 (Table 11).
After identifying the best fitting model without covariates, as described above, a 4-class
model with Level 1 covariates (meal, time of day, day of week) produced a BIC of 2277.372 and
an AIC of 2121.984; entropy was 0.871 (Table 12).
The latent class model results for this 4-class solution (including covariates) are presented
in Table 13.

102
Class 1 (“High NA, Hunger & EAH”): Eating events represented by Class 1 had
significantly higher rates of negative affect (β=2.965, SE=0.923), hunger before eating (β=0.939,
SE=0.309) and eating in the absence of hunger (start eating) (β=0.959, SE=0.370) (Table 13).
Class 2 (“Mindful”): Eating events represented by Class 2 had a significantly higher rate
for mindfulness (β=3.418, SE=0.969).
Class 3 (“Low EAH & High Satiety”): Eating events represented by Class 3 had a
significantly higher rate of satiety (β=1.467, SE=0.374) and significantly lower rates of eating in
the absence of hunger (start eating) (β=-3.410, SE=1.264) and (kept eating) (β=-2.527,
SE=0.457).
Class 4 (“High NA; Low Satiety & EAH; Eat Alone”): Eating events represented by Class
4 had a significantly higher rate of negative affect (β =1.613, SE=0.498) and significantly lower
rates of satiety (β =-1.115, SE=0.502), eating in the absence of hunger (kept eating) (β=-3.510,
SE=1.095), and eating with others (β=-1.501, SE=0.316).
Predictors of Latent Classes (Aim 2)
The odds of eating events in Class 1 being described as a meal (vs. snack/drink) were
0.052 times as likely compared to Class 4 (p<0.001). In other words, the eating events in Class 1
(High NA, Hunger & EAH) were significantly less likely to be described as meals vs. snack,
compared to the eating events in Class 4 (High NA; Low Satiety & EAH; Eat Alone). No other
Level 1 contextual factors (i.e., time of day, day of week) significant predicted the eating event
classes.

103
DISCUSSION
General Discussion
This study used multi-level latent class analysis (LCA) to understand participants’ eating
events automatically detected in the home across a two-week observation period. Four distinct
eating event classes were identified from the data. Class 1 (High NA, Hunger & EAH) and Class
4 (High NA; Low Satiety & EAH; Eat Alone) were primarily characterized by a high level of
negative affect, although Class 1 was significantly more likely to be a snack/drink (vs. a full
meal) compared to Class 4. Another notable difference between these two classes is that the
event events within Class 1—which were more likely to be snacks—had significantly higher
levels of hunger and eating in the absence of hunger, whereas eating events within Class 4 had
lower levels. Eating events within Class 4 were also significantly more likely to take place
without others around (i.e., alone). In contrast, eating events in Class 2 (Mindful) were primarily
characterized by a higher level of mindfulness. Lastly, events within Class 3 (Low EAH & High
Satiety) were mostly characterized by significantly lower levels of eating in the absence of
hunger and a high level of satiety. Overall, the features that characterize Classes 1 and 4, such as
negative affect and eating in the absence of hunger, are linked to unhealthy/undesirable eating
behaviors in children and adults (Anestis et al., 2010; Engel et al., 2007; Goldschmidt et al.,
2017; Lansigan et al., 2015; Mason et al., 2018); and the features that characterize Classes 2 and
3, such as mindfulness and satiety, are linked to healthy/desirable eating behaviors (French,
Epstein, Jeffery, Blundell, & Wardle, 2012; Warren, Smith, & Ashwell, 2017).
Class 4 (High NA; Low Satiety & EAH; Eat Alone) was the only class containing a
significant indicator for ‘inter-personal context’ (i.e., eating with others). Eating events within
this class were during stressful, yet low EAH intra-personal contexts – and were more likely to

104
take place alone. One’s social environment can play a large role in eating behavior, and in this
study, we see the interesting co-occurrence of (a lack of) a social environment and multiple,
momentary, intra-personal contexts (i.e., NA, satiety, EAH).
Some research has shown that individuals exhibit poor eating behaviors while alone. For
example, one study of Korean adults found that the diet quality of people who eat alone was
lower than those who ate together (Chae, Ju, Shin, Jang, & Park, 2018). Similarly, another study
found that eating alone was associated with unhealthy eating behaviors in a sample of Japanese
men (Tani et al., 2015). Both of these studies use retrospective, self-report measures of dietary
intake. In contrast, we identified this association in real-time, i.e., real-time social context co-
occurs with real-time eating behavior. The findings from this study indicate that some real-time
intra-personal (e.g., high negative affect) and inter-personal features (e.g., eating alone) may be
able to serve as points of intervention to promote healthier eating behaviors within families.
In line with the findings from this study, a recent systematic review reports that cognitive
and social factors consistently have an effect on eating and dietary intake behaviors (Mason, Do,
Wang, & Dunton, 2020). However, the association between affect and eating behaviors still
remains mixed (Mason et al., 2020). In Class 1 (High NA, Hunger & EAH) and Class 4 (High
NA; Low Satiety & EAH; Eat Alone), we see that negative affect is a significant and positive
indicator of membership. This finding extends the literature by showing that momentary negative
affect played a large role in determining membership for approximately half of the eating events
from the M2FED study.
Although one of our classes, Class 3 (Mindful), is primarily characterized by just one
indicator (mindfulness), multiple indicators significantly predict membership of Classes 1, 2, and
4. The approach of using multiple indicators to classify eating behaviors is uncommon compared

105
to other papers in the literature that have used LCA to classify various eating behaviors. For
example, one study used self-report dietary data from 24-hour recalls to identify distinct
temporal eating patterns; patterns such as “grazing” and “conventional” were identified (Leech,
Worsley, Timperio, & McNaughton, 2017). In another study, negative emotional eating patterns
were classified into four groups: “non-emotional eating”, “emotional over- and under-eating”,
“emotional over-eating”, and “emotional under-eating" (He, Chen, Wu, Niu, & Fan, 2020). In
both of these approaches, only one indicator variable was used for classification (“time” in Leech
et al.; “emotional eating” in He et al.). In contrast, we used multiple indicators (affect, eating in
the absence of hunger, mindfulness, etc.) from multiple contexts (intra- and inter-personal) to
generate a fuller picture of the eating events in our dataset.
This type of application of latent class analysis to identify subgroups of events – rather
than people – from multiple indicators is nascent. Only one study was found in the literature that
used the methodological approach in this manner. Similar to our analyses, Goldschmidt et al.
(2014) applied the latent class analysis framework to identify classes of eating episodes that were
collected via ecological momentary assessment in women with anorexia nervosa (Goldschmidt et
al., 2014). Eating episodes and other contextual factors were collected via an event-contingent
assessment, in which they reported eating episodes after their occurrence. The multiple indicators
used to inform latent class membership included: loss of control eating, overeating, eating alone,
food avoidance, and dietary restraint. Five classes of eating episodes were identified from the
latent class analysis: (1) avoidant eating, (2) solitary eating, (3) binge eating, (4) restrictive
eating, and (5) loss of control eating.
In Goldschmidt et al., all of the latent classes were characterized by features that are
associated with unhealthy/undesirable eating behaviors (e.g., binge eating, loss of control eating).

106
In contrast, we identified two classes primarily defined by features of healthy/desirable eating
behavior and two classes defined by features of unhealthy/undesirable behavior. One key
difference between their and our approach is that the indicators used in Goldschmidt et al. can all
be classified as “unhealthy” features (e.g., binging, dietary restraint, etc.), and may explain why
all of the identified classes can also be classified as “unhealthy” eating classes. By incorporating
“healthy” indicators in our analyses (e.g., mindfulness), we are able to extend the literature by
identifying “healthy” classes in additional to “unhealthy” classes.
From a methodological standpoint, this study sought to utilize in-field data from
ecological momentary assessment (EMA) and wearable sensors in order to provide insight into
how intra- and inter-personal contexts interact to influence eating behavior together in real-time.
The methodological approach of LCA is typically used to organize participants (people) into two
or more meaningful homogenous subgroups (Collins & Lanza, 2009), and is typically used in
cross-sectional studies in which there is only one observation per participant (Asparouhov,
Hamaker, & Muthén, 2017).
In contrast, the data collected from the M2FED study are intensive longitudinal data
(ILD). ILD are data with many measurements over time (Schafer, 2006), and are typically
generated from technologies such as smartphones, wearable sensors, and the Internet of Things
(IoT) (Collins & Lanza, 2009). Therefore, in contrast to cross-sectional data, the smallest unit of
measure is the behavior, state, event, etc. that is repeatedly measured across time (in this case, it
is an eating event). This study and its findings indicate that the using LCA to categorize event-
level data is feasible, and this application of LCA could be a future avenue of research in the
behavioral science field. As ILD become more prominently used in research settings, a variety of
research questions could be addressed using this approach.

107
Limitations and Advantages
Limitations. One of the primary limitations of the study is the way the indicator variables
were coded. Due to low variability, all indicator variables were dummy coded into “low” and
“high” categories. By aggregating the variables in this way, some of the complexity was
removed. Another limitation of the study is that the findings can only be generalized to in-home
eating behavior. The scope of this study did not include out-of-home eating. Lastly, many fathers
either dropped out of the study before its completion or had low participation (i.e., answered few
EMAs).
Advantages. Firstly, this study was conducted with a sample of ethnically diverse families
(majority Hispanic/Latino) living in Los Angeles County, who are typically underrepresented in
mHealth studies. Secondly, we implemented an “event centric” approach in order to provide us
with a richer understanding of eating events when informed by both intra- and inter-personal
features. Lastly, collecting measurements of eating behavior in context (where the behavior
occurs) and in or near real-time (when the behavior occurs) offers a number of methodological
advantages. This type of in-context data collection approach can improve the ecological validity
of research findings in comparison to findings from retrospective measures (Shiffman et al.,
2008); and it can reduce or eliminate the recall bias that typically plagues retrospective self-
report measures. In summary, real-time in-context data collection approaches, such as the one
used in the M2FED study, can improve the overall validity of eating behavior and contextual
data. These data can then reduce the level of measurement error (the difference between the
measured value and the true value) and subsequently provide us with a more accurate picture of
eating behavior (Archer & Blair, 2015; Schoeller, 1995; Schoeller et al., 2013).

108
Implications
There are a number of ways in which these findings can help shape future obesity
prevention strategies. Understanding the contexts in which certain types of eating behaviors
occur, including intra-personal and inter-personal contexts, is crucial to designing and
developing effective interventions. We categorized participants’ eating events into different
“contextualized” subgroups, based on various intra-personal (hunger level, satiety level, positive
and negative affect, mindful eating, and eating in the absence of hunger) and inter-personal
(eating companionship) contexts. These findings help us better understand the various
intersecting contexts in which “unhealthy” and “healthy” features of eating events occur, and
may provide important points of intervention to promote healthier eating behaviors within
families.
Many of the intra- and inter- personal contexts that we measured in the M2FED study—
and that are used as features to characterize eating event classes—have been separately linked to
unhealthy and healthy eating behaviors in the literature. The resulting problem is two-fold: (i)
when examining intra-personal contexts in isolation of other intra-personal contexts, we are
unable to examine how two or more intra-personal contexts co-occur in specific moments and
during specific eating events; and (ii) when examining intra-personal contexts in isolation of
inter-personal contexts, we are unable to examine how these two types of contexts co-occur in
specific moments and during specific eating events.
By utilizing multi-level latent class analysis, we gain a richer understanding of the
interdependence among these contexts, and how they vary across and event-level factors (i.e.,
meal status). Understanding when and for whom these intersecting contexts produce unhealthy

109
vs. healthy eating behaviors can inform obesity interventions by elucidating points of
intervention that may be most effective at changing behavior.
Future Directions and Conclusion
A novel feature of technological tools, including EMA and wearable sensors, is their ability
to passively measure important contextual features of eating behavior, such as physical context
(e.g., eating at home vs. at work), social context (e.g., social interaction), and more (Gemming et
al., 2015a; Spruijt-Metz et al., 2016). For example, wearable proximity sensors, like those
described in (Booth et al., 2019; L'Hommedieu et al., 2019), could be used to co-locate
individuals and thus automatically provide information on who is eating together or not without
user input, which could be useful in decreasing participant burden.
Additionally, these technological tools improve upon current dietary assessment methods by
(i) collecting real-time data in-context, frequently, and at a high level of granularity; and thus (ii)
improving the quality and validity of data that are collected. These methodological advantages
allow for a much fuller picture of individuals’ eating behaviors.
Technology-assisted dietary assessment tools also offer the ability to explore micro-level
eating activities, such as meal microstructure (the dynamic process of eating, including meal
duration, changes in eating rate, chewing frequency, etc.) (Doulah et al., 2017), food choices
(Marcum et al., 2018), and processes (e.g., eating rate (Ohkuma et al., 2015); eating mimicry
(Bell et al., 2019; Sharps et al., 2015); etc.), which is important because recent literature suggests
that they may play an important role on food selection, dietary intake, and ultimately, obesity and
disease risk.
However, many of these tools are nascent, and further research is needed to develop and
improve these tools.

110
TABLES

Table 8. Individual-level characteristics of M2FED analytic sample (n=46), by family
member role (Dissertation Study 2)

Note: The percentages

presented are column percentages.
1
n=40

Characteristic Child (n=25) Parent (n=21)
M (SD) / n (%)
Age (years) 15.48 (3.63) 44.57 (6.53)
Gender, Female (%) 9 (36.0%) 15 (71.4%)
Race/Ethnicity (%)

Asian or Pacific Islander 1 (4.0%) 1 (4.8%)
Black or African-American 0 (0.0%) 1 (4.8%)
Hispanic or Latino 15 (60.0%) 15 (71.4%)
White 3 (12.0%) 3 (14.3%)
Mixed 5 (20.0%) 0 (0.0%)
Other 0 (0.0%) 1 (4.8%)
BMI Percentile
1
22.51 (4.51) 33.27 (7.83)

111
Table 9. Summary statistics of eating event indicator variables and covariates (N=290)

Characteristic N=290
N (%)
Covariates
Morning 57 (19.66%)
Afternoon 98 (33.79%)
Evening 135 (46.55%)
Weekend, yes 95 (32.76%)
Meal, yes 204 (70.34%)
Indicator Variables
Eat with others, yes 186 (64.14%)
Negative affect 90 (31.03%)
Hunger before eating 141 (48.62%)
Fullness after eating 143 (49.31%)
EAH (start eating) 241 (83.10%)
EAH (kept eating) 234 (80.69%)
Mindfulness 127 (43.79%)

112
Table 10. Fit indices for multilevel latent classes of eating events, without covariates

Number of Level-1 Latent Classes
Indices 1 Class 2 Classes 3 Classes 4 Classes 5 Classes 6 Classes
Free
parameters
7 16 25 34 43 52
Log-
likelihood
-1220.593 -1117.098 -1068.994 -1031.984 -1014.239 -999.698
LRT -- > 0.05 > 0.05 > 0.05 > 0.05 > 0.05
AIC 2455.186 2266.195 2187.987 2131.967 2114.478 2103.396
BIC 2480.875 2324.913 2279.734 2256.743 2272.283 2294.230
SSABIC 2458.677 2274.174 2200.455 2148.923 2135.923 2129.328
Entropy -- 0.828 0.846 0.943 0.891 0.912
Note: LRT = Lo–Mendell–Rubin likelihood-ratio test; AIC = Akaike information criterion; BIC
= Bayesian information criterion; SSABIC = sample size adjusted BIC.

113
Table 11. Class counts and proportions for 4-class solution (N=290)

Latent Classes Count Proportion
1 54 0.186
2 53 0.183
3 92 0.317
4 91 0.314

114
Table 12. Fit indices for conditional 4-class model of eating events, including covariates
(N=290)

Indices 4 Classes
Free
parameters
49
AIC 2097.547
BIC 2277.372
SSABIC 2121.984
Entropy 0.871
Note: AIC = Akaike information criterion; BIC = Bayesian information criterion; SSABIC =
sample size adjusted BIC.

115
Table 13. Multi-level latent class analysis results for 4-class solution, including covariates

Class 1 (18.6%) Class 2 (18.2%) Class 3 (31.7%) Class 4 (31.4%)
Eating Events
Indicator “High NA,
Hunger &
EAH”
“Mindful”
“Low EAH &
High Satiety”
“High NA; Low
Satiety & EAH;
Eat Alone”
β (SE)
Hunger Level,
high
0.939** (0.309) -0.684 (1.130) 3.502 (2.584) -3.065 (2.348)
Satiety Level,
high
0.887 (0.803) -1.260 (1.807)
1.467***
(0.374)
-1.115* (0.502)
Negative Affect,
high
2.965** (0.923) -1.305 (0.682) 0.636 (0.887) 1.613* (0.498)
Mindful Eating,
high
0.240 (0.391)
3.418***
(0.969)
-0.066 (0.643) -0.442 (0.513)
EAH - Start,
high
0.959* (0.370) -15.000 (0.000) -3.410** (1.264) -3.654 (1.910)
EAH - Keep,
high
1.306 (0.987) -15.000 (0.000)
-2.527***
(0.457)
-3.510* (1.095)
Eating with
others, yes
-0.334 (0.318) -0.903 (0.581) 0.142 (0.744)
-1.501***
(0.316)
*p<0.05, **p<0.01, ***p<0.001

116
FIGURES
Figure 10. Overview of sampling protocol for (a) time-triggered and (b) eating event-
triggered EMAs

(a) Time-triggered EMAs

Note: “Top of hour” refers to the beginning of the hour (e.g., 10:00am, 11:00am, 12:00pm, etc.).

Top of hour:
Time-triggered EMA sent
to phone
Participant completes
survey within 3 minutes
Participant cannot receive
any TIME-TRIGGERED
surveys for the next 60
minutes
Participant does NOT
complete survey within 3
minutes
Participant receives a
reminder survey (up to 3
times)
The survey is not
completed after 3
reminders are sent
Participant cannot receive
any TIME-TRIGGERED
surveys for the next 60
minutes
The survey is completed
Participant cannot receive
any TIME-TRIGGERED
surveys for the next 60
minutes

117
(b) Eating event-triggered EMAs

Eating event detected:
Eating event-triggered
EMA sent to phone
Participant completes
survey within 5 minutes
Participant cannot receive
ANY surveys for the next
60 minutes
Participant does NOT
complete survey within 5
minutes
Participant receives a
reminder survey (up to 3
times)
The survey is not
completed after 3
reminders are sent
Participant can receive
ANY surveys for the next
60 minutes
The survey is completed
Participant cannot receive
ANY surveys for the next
60 minutes

118
CHAPTER 4: INVESTIGATING THE DYNAMICS OF IN-HOME FAMILY EATING
BEHAVIOR: A RELATIONAL EVENT MODELING APPROACH
ABSTRACT
Introduction: Features and contexts within the family system can have a large impact on eating
behaviors and health. Although the field has seen a marked increase in the study of social and
familial influences on eating behavior, few studies have taken a family systems approach:
examining how features of the family system interact over time to impact its members’ eating
behavior. In addition, a systems approach acknowledges the various individual-level factors that
have also been shown to influence eating behavior. Many intra-personal states have been
established as predictors of eating behavior in cross-sectional studies, but the temporal
relationship between these individual states and eating behavior in real-time are underexplored.
Therefore, the purpose of this paper is to use the relational event modeling framework to
investigate the role of four ego-centric (i.e., ‘person-centric’) mechanisms in predicting the
sequence of in-home intra-personal events (affective states) and behavioral events (eating events)
within a family system over a two-week period of time.
Methods: The Monitoring and Modeling Family Eating Dynamics (M2FED) study recruited 20
families to partake in an in-the-field observational study. Across a two-week period, each family
member wore a smartwatch that automatically detected their in-home eating events and reported
on their affective states via hourly Ecological Momentary Assessments (EMAs). Ego-centric
relational event modeling was used to determine how sequences of relational events (eating
events and affective states) unfolded within the family system.
Results: Various types of inertia effects, including affective state followed by affective state,
eating event followed by eating event, and affective state followed by eating event followed by

119
affective state were observed in the dataset. Some key differences emerged once the model was
stratified by family role. Children were significantly less likely to experience the sequence “high
negative affect (NA) → eating event with others/high NA”, as compared to the reference event
(low NA), while parents were significantly more likely to experience this sequence.
Conclusion: The results from our study provide insight into the interplay of affective states and
eating events, the potential differences between different family roles, and the differences
between social contexts, with implications for future just-in-time behavioral interventions to
promote healthy eating patterns.
INTRODUCTION
Obesity as a Complex System
The etiology of obesity has many complex factors and interdependences (Spruijt-Metz,
2011). Traditional health behavior models, such as the Health Belief Model (Champion &
Skinner, 2008) and the Theory of Planned Behavior (Ajzen, 1991), suggest that an individual’s
behaviors are a result of one’s thoughts, beliefs, and intentions. However, more recent models
suggest that a much wider variety of factors influence obesogenic behaviors, including an
individual’s dietary intake and eating behavior. Many conceptual models have been proposed to
expand the traditional “energy balance framework” (Spiegelman & Flier, 2001), such as the
ecological model (Egger & Swinburn, 1997); a systems-oriented, multi-level model (Huang et
al., 2009); and a complex systems model (Hammond, 2009; Lee et al., 2017). These models posit
that the determinants of obesity are both complex and potentially non-linear, and their
frameworks provide the ability to investigate factors that are at multiple levels (e.g., intra-
personal, inter-personal, community, etc.) and their interdependences.

120
For instance, an increasingly popular conceptualization of the influences on eating
behavior and drivers of obesity is the “Obesity System Map” developed in a report
commissioned by the United Kingdom’s Government Office for Science (Vandenbroeck et al.,
2007). This map depicts relevant factors and their interdependences that determine the condition
of obesity for an individual or a group of people. It consists of 108 variables and 304 causal
linkages, which are represented in seven clusters: individual physiology, individual physical
activity, physical activity environment, individual psychology, social psychology, food
consumption, and food production. Similarly, another framework often used to portray the many
influences on eating behavior is an ecological framework. Story et al. present this framework by
illustrating the individual factors (e.g., cognitions, demographics), social environments (e.g.,
family, friends, social support), physical environments (e.g., home, school, neighborhoods, food
outlets), and macro-level environments (e.g., economic systems, food marketing and media) that
influence what people eat (Story et al., 2008).
Recognizing the obesity epidemic as a complex systems problem will necessarily shift the
paradigm of how public health researchers and professionals approach obesity prevention and
intervention strategies (Luke & Stamatakis, 2012). These social-ecological and systems science
models of obesity increasingly recognize that a broad range of interrelated factors influence
individuals to adopt particular obesogenic behaviors (e.g., diet, physical activity). Individual
processing of these stimuli and subsequent engagement in behaviors is therefore central to this
system. Identifying and understanding the stimuli that influence eating behaviors within systems
is critical to the development of effective systems-focused obesity prevention and intervention
efforts.

121
A substantial amount of work has been done to identify and characterize the ways in
which the physical food environment influences eating behavior (e.g., food accessibility, food
advertisements, toxic food environment). The characterization of how one’s social environment
influences eating behavior, and moreover, how it interacts with intra-personal factors and how
the effects change across time (i.e., a systems perspective), has been less of a focus.
A Family Systems Approach
Evidence suggests that various features and contexts within the family system can have a
large impact on eating behaviors and health outcomes (i.e., obesity and chronic disease risk),
especially for children. This is because the family system and the home are common contexts in
which eating takes place (Dallacker et al., 2018; Neumark-Sztainer et al., 2003). Additionally,
parental behaviors, attitudes, and feeding styles have a strong influence on child dietary intake
(Anzman et al., 2010; Savage et al., 2007; Vollmer & Mobley, 2013; Yee et al., 2017); few
studies have examined the influence in the other direction (i.e., child’s influence on parent)
and/or the bidirectional influences between parents’ diet and children’s diet intake.
What other family members are eating and how they’re eating can be influential – family
members tend to engage in similar eating behaviors, such as in the areas of food choice (Ayala et
al., 2007; Cameron et al., 2011; Pachucki et al., 2011) and eating style (e.g., restrained eating)
(Munsch et al., 2007). More subtle features and processes can also have an impact. For example,
the mere presence of other family members can influence eating behaviors – one early study
found evidence that meals eaten with family were larger and faster compared to meals with other
types of companions (i.e., social facilitation) (de Castro, 1994). Relatedly, social cues within a
family system can have a strong impact on food intake through several mechanisms, including
mindless imitation of others’ eating behaviors (Bell et al., 2019; Hermans et al., 2012; Sharps et

122
al., 2015), food matching and modeling (Salvy et al., 2012), and situational and cultural food
norms (Herman & Polivy, 2005).
Although the field has seen a marked increase in the study of social and familial
influences on eating behavior (Herman et al., 2019), few studies have taken a family systems
approach: examining how features of the family system interact over time, to impact its
members’ eating. Another notable limitation is that much of this research has taken place in
laboratory settings (vs. in the field settings), and has relied on cross-sectional survey data and/or
in-lab measurement periods only spanning a few hours. It is uncommon to observe families in
rich settings and during everyday routines that shape their interactions and eating behavior, and
therefore, we do not have the opportunity to observe family dynamics play out over meaningful
periods of time. Overall, these limitations prohibit the exploration of the rich dynamics that may
play out as multiple eating events unfold over time. To date, no studies have attempted to
monitor or model the real-time dynamics around eating within a family system in the wild.
Affective States and Eating Behavior
A systems perspective also acknowledges that a number of individual-level factors (e.g.,
attitudes, behavior, internal states) have been shown to influence eating behavior, including:
eating in the absence of hunger (Fisher & Birch, 2002), mood, anger, and stress during eating
(Lemmens et al., 2011). In particular, the relationship between affect, stress, and eating has been
extensively investigated (Ganley, 1989; Greeno & Wing, 1994; Sominsky & Spencer, 2014;
Stone & Brownell, 1994), however the underlying mechanisms are still unclear to date. Mixed
results in the literature (Dunton et al., 2016; Groesz et al., 2012; Hou et al., 2013) indicate that
stress can either lead to undereating or overeating, possibly varied by stressor severity (e.g., mild
vs. severe, etc.) or stressor duration (e.g., acute vs. chronic) (Torres & Nowson, 2007).

123
Many of these intra-personal states, such as hunger levels, eating in the absence of
hunger, and mindful eating, have been established as associates of eating behavior in cross-
sectional studies (Beshara et al., 2013; Fogel et al., 2018; Gilbert & Waltz, 2010; Lansigan et al.,
2015; Lavender et al., 2011), but the temporal relationship between these individual states and
eating behavior in the moment are underexplored.
A handful of studies have used Ecological Momentary Assessment (EMA), a data
collection technique in which one’s behavior is repeatedly sampled in the natural environment, to
assess the relationship between an individual’s momentary negative affect/stress and their
subsequent eating behavior (Engel et al., 2016). Studies have found that momentary anger,
negative affect, and stress are significantly associated with binge eating episodes in women with
eating disorders (Anestis et al., 2010; Engel et al., 2007; Mason et al., 2018).
Overall, the reviewed evidence indicates that (i) the social environment, and family
contexts in particular, influence eating behavior and (ii) individual-level factors such as affective
states influence eating behavior. However, how these individual- and social-level features within
a family system interact and change over time has yet to be explored, and it is plausible that there
are complex interactions and sequences of events that are central to family eating. The
identification of temporally-specific processes and events at multiple levels of influence
within the family system is likely to meaningfully improve our understanding of how
children and families influence one another’s eating behaviors, and can guide future
interventions and thus have a positive impact on diet and ultimately obesity.
M2FED Study: A Systems Approach to Modeling Family Eating Dynamics
Emerging technologies are poised to collect real-time, in context data that may lend
insights into these intersecting influences on eating behavior. However, most health behavioral

124
theories and models are not ‘up to the task’ (Riley et al., 2011) of modeling these dense,
temporally-rich data produced by these technologies. New dynamic models of health-related
behaviors are needed to understand behavior as it unfolds in context and in real-time (Saranummi
et al., 2013; Spruijt-Metz & Nilsen, 2014).
To address the limitations of past behavior models (Spruijt-Metz et al., 2015) and
traditional dietary assessment methods (Spruijt-Metz et al., 2018), the Monitoring and Modeling
Family Eating Dynamics (M2FED) study developed and deployed new methods for in-home
sensing to monitor and model eating behavior and multi-level contexts in a family setting
(Spruijt-Metz et al., 2016). A suite of sensors, including smartwatches, Bluetooth beacons, and
smartphones, was used to monitor features related to eating, including behaviors, intra-personal
(individual) states (e.g., affective states), and inter-personal (social) contexts before, during, and
after eating events to understand the potential interrelationships among all behaviors. The data
for this paper come from the Monitoring and Modeling Family Eating Dynamics (M2FED)
study.
The M2FED study broadly aimed to identify temporally-specific processes and events
within the family system that can be targeted for personalized, context-specific, real-time
feedback (“Just-In-Time Adaptive Interventions” or JITAIs) (Nahum-Shani et al., 2015; Nahum-
Shani et al., 2017; Spruijt-Metz & Nilsen, 2014). Identification of these sequences can inform
future interventions to encourage healthier family eating dynamics that are likely to have a
downstream positive impact on diet and ultimately obesity.
The M2FED study is the first study that utilizes novel technology to capture and model
family dynamics in real-time to identify potential opportunities for innovative interventions in
the future.

125
Relational Event Modeling (REM) Framework
The Relational Event Modeling (REM) framework (Butts, 2008; Butts & Marcum, 2017)
is especially suited to model the sequences of eating events in these family systems, using rich
continuous time-stamped data from wearable sensors and ecological momentary assessment
(EMA). It provides a more ‘bottom-up’ perspective of the role that individual events and their
embedded contexts play in the emergent evolution of the system. Specifically, the relational
event model examines the mechanisms that drive the occurrence of future events in a continuous
sequence. Overall, relational event models enable us to test specific patterns of interdependences
between family members’ eating events and affective states.
Study Aims
This study investigates eating behavior within families as a “complex system” – namely,
the different components and factors that are interconnected and how these components interact
with and affect one another. Using a relational event modeling framework allows us to look at
complex systems and to understand how they work and what features drive particular sequences
of events, which will help to identify important leverage points to promote healthier behaviors
within the system.
Past research has looked at psychological, social, and environmental influences on eating
behavior, but often using cross-sectional, self-reported, and/or retrospective data. The analyses
for this study are exploratory in nature because little is known on how family members’
sequences of affective states and eating events unfold in real-time.
Aim 1: Investigate the role of four proposed ego-centric mechanisms in predicting the
sequence of in-home intra-personal events (affective states) and behavioral events (eating events)

126
within a family system over a two-week period of time. The four mechanisms that represent
dynamic phenomena of interest, based on the literature, are:
(i) Affective state inertia: tendency of one’s past actions (i.e., the experience of an affective
state) to predict future actions (i.e., experiencing an affective state with the same
features);
(ii) Eating event inertia: tendency of one’s past actions (i.e., the experience of an eating
event) to predict future actions (i.e., experiencing an eating event with the same features);
(iii)The sequence of an affective state followed by an eating event featuring that same
affective state (i.e., tendency of a high negative affective state → an eating event
featuring high negative affect; and a low negative affective → an eating event featuring
low negative affect); and
(iv) The sequence of an affective state followed by an eating event featuring the same
affective state followed by the same type of affective state (i.e., tendency of a high
negative affective state → an eating event featuring high negative affect → high negative
affective state; and a low negative affective state → an eating event featuring low
negative affect → low negative affective state).
Aim 2: Examine whether the presence of these specific event sequences varies by family
role (e.g., parent, child) to establish whether there are consistent temporal patterns across all
types of family members or whether novel patterns emerge for certain types of family members.

127
METHODS
Participants and Recruitment
Eligibility. The inclusion criteria for families that expressed interest in participating in the
Monitoring and Modeling Family Eating Dynamics study were:
• Families must consist of at least two related family members.
• Families must include at least one adult parent and one child between the ages of
11 and 18 years old.
• Families must live within Los Angeles County.
Families that had one or more children under the age of 11 (these children were ineligible to
participate) were still eligible to enroll other family members in the study as long as there was at
least one parent and at least one other child between the ages of 11 and 18. There were no
demographic- or disease-related exclusion criteria. However, families with one or more members
living in the home who did not primarily speak English were ineligible to enroll.
Family members that were eligible to participate in the study will be referred to as
‘participants’ hereafter.
Method of recruitment. Recruitment efforts for the M2FED study were primarily conducted
at public spaces, such as train and bus stops, and public events, such as football games and book
fairs, within Los Angeles County from May 2017 to August 2019. Our research team handed out
flyers containing the study’s website information and their contact information at these public
locations. As a secondary recruitment strategy, we also employed snowball sampling, such that
families that successfully completed the study were offered an additional $20 if they referred
other eligible families that successfully enrolled in the study.

128
Recruitment staff followed up via a telephone call on any person (typically a parent) that
expressed interest in the study and provided their contact information. Once eligibility was
confirmed, and at least one parent verbally agreed that the family would participate, the
recruitment staff administered an intake screening tool over the phone before officially enrolling
the family in the study.
This study was approved by the Institutional Review Board of the University of Southern
California. All participating parents provided informed written consent, and all children provided
assent.
M2FED System Overview
The M2FED cyberphysical system (Figure 1) was developed and deployed into families’
homes as part of the M2FED study (Spruijt-Metz et al., 2016). Cyberphysical systems can be
defined as “physical and engineered systems whose operations are monitored, coordinated,
controlled and integrated by a computing and communication core” (Rajkumar et al., 2010).
The system consisted on four primary components that were used to monitor in-home
family eating behavior for all participating family members. These components included: (1) a
base station, (2) wearable devices and in-situ sensors, (3) an ecological momentary assessment
(EMA) subsystem, and (4) a remote monitoring system. All of these system components were
connected through a Wi-Fi router (see Figure 1).
Due to the scope of this study and the capabilities of the technologies that were used, all data
collected by the system were measured in the home. Data outside of the home were not
collected.

129
Base station. The system’s base station (Lenovo ThinkPad laptop) collected and processed
the data that it received from the devices in the system through the Wi-Fi router. It also managed
the EMA subsystem described below.
Devices and sensors. Two wearable devices (smartwatches and smartphones) were given to
participants for the duration of the study. Participants were instructed to keep their smartwatch
and smartphone in the home (i.e., not take them outside of the home). If the devices were outside
the range of the Wi-Fi router, the smartwatches did not send data to the base station, and the
smartphones did not receive EMAs sent by EMA subsystem.
Smartwatches and automatic eating detection. Firstly, all participants were given
smartwatches (Sony Smartwatch 3, Android Wear operating system), and were instructed to
wear them on their dominant hand during all waking hours they were at home.
Software developed by the research team was downloaded onto the smartwatches, and was
used to automatically detect eating-related hand-to-mouth (H-t-M) gestures for each participant
in the home and in real-time. Motion data from the accelerometer and gyroscope (inertial sensors
inside the smartwatch) were used to detect clusters of at least two H-t-M gestures within a one-
minute timeframe. A group of these clusters were then ultimately classified as an “eating event”.
An eating event can be defined as a set of H-t-M gestures, representing phenomenon such as
consuming a meal, a snack, a drink or a combination of these consumption behaviors in which
H-t-M gestures are clustered temporally. The technical details of the eating event detection
algorithm are provided in detail elsewhere (Mondol et al., 2020).
Smartphones and EMA. Participants were also given smartphones (Samsung Galaxy S7,
Android operating system), and were instructed to respond to mobile questionnaires sent by the
EMA subsystem. The EMA subsystem repeatedly sent two types of mobile questionnaires to

130
participants in order to measure a number of behaviors and states in the natural environment (at
home) (Shiffman et al., 2008). The first type of questionnaire that participants received were
time-triggered mobile questionnaires. These questionnaires were sent to participants’
smartphones every hour at the top of the hour (e.g., 10am, 11am, 12pm, etc.) (Figure 2a).
Participants were asked four short questions about their negative affect (mood).
The second type of questionnaire that participants received were eating event-triggered
mobile questionnaires. When an eating event was detected by the smartwatch of a given
participant, a questionnaire was immediately sent to the corresponding participant’s smartphone
asking them to confirm whether or not they had just eaten (Figure 2b). If they indicated that they
had not just eaten, then they were asked what activity they had just been doing and no additional
questions were asked. If they confirmed that they were eating, and that they had finished eating,
then they were asked to report on various intra- and inter-personal contexts, including their
current level of negative affect and with whom they had just eaten, if anyone.
Figure 3 illustrates the full eating event-triggered EMA question logic. The full list the
questions for the time-triggered and the eating event-triggered mobile questionnaires can be
found in Table 1.
To prevent the participants from being inundated with questionnaires and experiencing
unreasonable burden, participants could only receive mobile questionnaires no more than once
per hour. For similar reasons, questionnaires timed out and could not be answered if they weren’t
answered within 15 minutes (Figure 10).
Lastly, EMAs were only sent to participants during their “personalized participation
window,” the period in which they indicated at intake that they normally woke up and went to
bed.

131
Beacons. Approximately one to two in-situ proximity beacon sensors (Estimote Bluetooth
Low Energy (BLE) beacons) were installed on the walls of each “living space” (excluding
bathrooms and bedrooms) in order to determine the approximate location of participants’
smartwatches during the length of the study. The beacons were primary used to determine
whether the watches were still in the home, and did not require any interaction with the
participants following installation.
Remote monitoring subsystem. The monitoring subsystem was used to monitor the status
and operations of the M2FED system in real-time (Ma et al., 2017). Research team members
were alerted of issues detected by the system via email alert.
Procedures
The research team (typically two team members) visited the homes of all recruited
families following their enrollment. There were two visits included in the study procedures.
Visit 1. At the first home visit, all family members who were participating in the study
provided written consent. Then, participants had their height and weight measured by a research
team member (using a Tanita scale and a stadiometer), and they completed baseline surveys on a
laptop or tablet.
While participants’ completed the surveys, the research team installed the components of
the cyberphysical system around the home. The system’s base station, Wi-Fi router, and
Bluetooth beacons were secured in discrete locations within the family’s home so they could run
without interference for the duration of the study. All participating family members received
designated Samsung Galaxy S7 smartphones (Android operating system) and Sony
Smartwatches 3 (Android Wear operating system) for the duration of the study (all smartphone
features except for answering questionnaires were turned off). To ensure that participants knew

132
which devices were theirs, each smartphone and smartwatch was labeled on the outside with the
participants’ first name. Once all family members had completed the baseline measures, the
research team provided instructions and gave a demonstration on how to properly wear, charge,
and care for the smartwatches, and how to answer an EMA on the smartphones. Participants
were instructed to both (i) wear the watch and (ii) answer all EMA questionnaires they received,
when they were at home for the duration of the 2 week study. Lastly, they were instructed to
leave their study-assigned phone and watch at home when they left the home, to prevent the
devices from getting damaged or lost.
Participating family members then underwent approximately fourteen consecutive days
of (1) use of a smartphone to complete hourly time-triggered and eating event-triggered mobile
questionnaires, up to once every hour during waking hours; and (2) eating event monitoring, in
the form of a wrist-worn smartwatch during waking hours.
Visit 2. After the study had ended, approximately two weeks following the first home
visit, the research team revisited the home to terminate data collection and uninstall and remove
all equipment from the home. Each participant received $100 in a Visa gift card format as
compensation for the two-week study.
Measures
Data pre-processing. An algorithm was generated by our team to determine the time
intervals in which we are confident participants were both (i) at home and (ii) actively
participating in the study (i.e., answering EMAs and/or wearing the smartwatch) (Figure 4).
Essentially, if participants had answered an EMA in the past 15 minutes, then they were
classified as “participating” for the past 15 and the next 15 minutes (30 minute interval).
Otherwise, if data from the smartwatch and beacons were available and indicated movement,

133
then they were classified as “participating” for the past 1 minute. Contiguous 1-minute
“participation” intervals were merged into larger “participation” time intervals (Figure 5). The
union of all “participation” intervals were used as the valid time intervals in the analyses
described below. Data for times outside of these intervals (i.e., times we are not confident
participants were at home and participating) were not included in the current analyses.
Affective states. A participant’s affective state was measured throughout the day via a time-
triggered EMA approximately every hour during waking hours during the two-week assessment
period. The timestamps of these affective states (format: mm/dd/yyyy, hh:mm:ss) were recorded
by the system.
Participants were asked to rate how they were feeling right before they received the EMA
mobile questionnaire on the following four affect items: (1) Upset, (2) Nervous, (3) Stressed, and
(4) Couldn’t cope (Cohen et al., 1983; Laurent et al., 1999; Terry et al., 1999). Participants could
choose from the following ratings for each item: 1 = “Not at all,” 2 = “A little,” 3 = “Some,” and
4 = “Very.”
A participant’s negative affect (NA) score was then calculated as the average of the four NA
items (Upset, Nervous, Stressed, Couldn’t cope). Each average score can range from 1.00 to 4.00
(lower average score indicates low NA, higher average score indicates high NA).
Affective state valence: Affective states were categorized as either low or high NA. If
any average NA score was greater than the minimum (1.00), then it was coded as high NA.
Otherwise, any average NA score that was the minimum (i.e., a participant answered “1” for all
four NA items) was coded as low NA.
Eating events. Data collected from the smartwatches, and processed using the automatic
eating event detection software, were used to generate a dataset of timestamped eating events

134
(approximate start times, format: mm/dd/yyyy, hh:mm:ss) for each participant. Only eating
events that were confirmed by the participant (via eating event-triggered EMA) were included in
these analyses.
Eating event features: Two indicators were used to characterize eating events. Firstly,
participants reported on their level of negative affect (NA) via EMA immediately following a
detected eating event (i.e., individual-level context). This indicator was measured and processed
using the same approach described above. Participants also reported on who was eating with
them (i.e., social context). The response options included: “Nobody,” “Spouse/partner,”
“Child(ren),” “Mother,” “Father,” “Sister(s),” “Brother(s),” “Grandparent(s),” “Other family,”
“Friend(s),” and “Other people.” These measures were used to create an indicator variable
representing social companionship: 1=eating with others, 0=eating alone.
From these two event-level features, four types of eating events were generated:
1. Eating with others/low NA
2. Eating with others/high NA
3. Eating alone/low NA
4. Eating alone/high NA
Other covariates.
Age, gender, family role: The survey administered during Home Visit 1 captured basic
demographic information for each participant, including their current age (in years), gender
(Female, Male), and family role (Mother, Father, Child, Grandparent, Aunt, Uncle, Other).
Analytic Approach
This paper utilized the relational event modeling (REM) framework, which allows for
the dynamic modeling of event sequences, while incorporating cognitive, behavioral, and

135
social/contextual processes (Butts, 2008). This approach draws upon event-history analysis and
“helps researchers answer the question ‘what drives what happens next’ in a complex sequence
of interdependent events” (Butts & Marcum, 2017).
The data collected in the M2FED study are inherently relational and temporally dense,
and the study aims focus on temporal processes. However, many traditional statistical
approaches in the behavioral health field (e.g., multi-level linear regression, repeated measures
ANOVA) and the social networks field (e.g., exponential random graph models, stochastic actor-
oriented models) are limited in their ability to model dynamics of behavior or other measures
with temporally-dense and real-time data. Therefore, the REM framework is appropriate because
it can examine dynamics over time without the loss or aggregation of temporally dense data.
REM is used to determine the hazard (i.e., the propensity or likelihood) of a relational
event to occur at any moment, given the event history. A relational event is traditionally defined
as a discrete event generated by a social actor (the “sender”) and directed toward one or more
targets (the “receivers”) (Butts, 2008), which could be another social actor or feature of the
system. These events are instantaneous, well-ordered in time, and are most commonly used to
represent “instances” of social behavior between individuals (Butts & Marcum, 2017).
The REM framework can be implemented to estimate dyadic relational event models, but
can also more generally accommodate ego-centric (i.e., ‘person-centric’) event histories
(Marcum & Butts, 2015). The dyadic approach can be used to analyze the unfolding of events
within a social network (e.g., a family), and can elucidate interpersonal effects (i.e., the effect of
one individual’s affective state on another family member, etc.). But, a current limitation of the
dyadic approach is that it’s methodologically challenging to examine ‘actions/events’ directed at
multiple receivers (e.g., multiple family members in the system). Therefore, the ego-centric

136
REM approach was chosen over the dyadic approach, and family dynamics were incorporated
into the analysis through the EMA self-reported measure of eating companionship and the
mechanisms tested in the relational event model. In this study, behavioral sequences of affective
states and eating events were modeled as “ego-centric” relational events (Marcum & Butts,
2015).
This statistical approach allows us to test a set of hypothesized “processes and
mechanisms” that explain the observed sequence of events. These mechanisms may increase or
decrease the propensity of a given event. These hypothesized mechanisms have been termed as
“sequential structural signatures” (also known as sufficient statistics), and can be defined as
sequences of relational events that unfold in a particular pattern (Leenders, Contractor, &
DeChurch, 2016; Schecter & Contractor, 2017). For our analyses, the four mechanisms we
were interested in are (1) affective state inertia, (2) eating event inertia, (3) the sequence of
an affective state → an eating event that shared the same affective features; and (4) the
sequence of an affective state → an eating event that shared the same affective features →
an affective state.
Our data meet the three primary assumptions for this statistical approach, namely: all
events are observed, and the period of observation is “exogenously” determined (e.g., chosen by
the researcher); events cannot occur at the same time and are temporally ordered; and event
hazards are piecewise constant (“piecewise constancy”), with changes occurring either when an
event is realized or at exogenous “clock” events. (Butts & Marcum, 2017)
REM specification
To investigate the four proposed mechanisms (Study Aim 1), the sequence of ego-centric
relational events (i.e., mechanisms) were operationalized as follows:

137
(1) Affective state inertia:
• Low NA → Low NA
• High NA → High NA
(2) Eating event inertia:
• Eating with others/low NA → Eating with others/low NA
• Eating with others/high NA → Eating with others/high NA
• Eating alone/low NA → Eating alone/low NA
• Eating alone/high NA → Eating alone/high NA
(3) Sequence of an affective state followed by an eating event that shared the same
affective features:
• Low NA → Eating with others/low NA
• Low NA → Eating alone/low NA
• High NA → Eating with others/high NA
• High NA → Eating alone/high NA
(4) Sequence of an affective state followed an eating event that shared the same affective
features followed by an affective state:
• Low NA → Eating with others/low NA → Low NA
• Low NA → Eating alone/low NA → Low NA
• High NA → Eating with others/high NA → High NA
• High NA → Eating alone/high NA → High NA
Model selection
Given the complexity of the models and number of variables, a stepwise model selection
process was used. First, we fit a model that included only the “base effects”: (1) low NA (used as

138
reference), (2) high NA, (3) eating with others/low NA, (4) eating with others/high NA, (5)
eating alone/low NA, and (6) eating alone/high NA. Then, each parameter representing a unique
mechanism was added one at a time, resulting in multiple models. Goodness of fit (GOF) indices
(log likelihood, Akaike information criterion (AIC), and Bayesian information criterion (BIC))
were used to evaluate the performance of the multiple models and ultimately select the final
model with the best fit.
Interaction effects
To examine how the mechanisms varied by family role (Study Aim 2), we ran a model
containing interaction terms that represented the various effects × family role. If any of the
interaction terms were significant, then the final REM was stratified by family role.
Missing data
If an eating event-triggered EMA observation was missing the negative affect measure or
the social companionship measure (i.e., the participant did not answer at least one of these
questions), then it was removed from the eating event dataset.
Statistical software
R (version 4.0.2) was used to perform these analyses. Specifically, we utilized the
‘relevant’ R package that was developed to facilitate relational event model analysis (Butts,
2015) and the ‘informR’ package to construct the sequence statistics terms included the
relational event models (Marcum & Butts, 2015).

139
RESULTS
Sample Characteristics
A total of 74 participants from 20 families enrolled in the M2FED study. Thirteen
participants dropped out of the study or were removed from the dataset if their participation (as
determined by the participation algorithm) was 0% (i.e., they did not answer any EMAs nor ever
wore the smartwatch) (Figure 6). Additionally, the data from three non-parent adult participants
made up approximately 1.5% of the EMAs received, so these participants were removed from
the analytic sample as well. The remaining 58 participants included in the analytic sample did
not significantly differ from the enrolled sample (N=74) by age, gender, or parent role (Table 2).
Of the 58 participants, 41.4% were parents (n=24) and 58.6% were children (n=34). On
average, children were 15.74 years old (SD=5.30 years) and parents were 44.04 years old
(SD=6.65 years). There were 13 female children (38.2% of children) and 17 female parents
(70.8% of parents). 61.8% of children and 66.7% of parents identified as Hispanic or Latino
(Table 3).
Event Descriptives
On average, the length of a family deployment lasted 14.90 days (SD=3.13) (Table 4). A
total of 287 eating events were confirmed as true eating events by participants via eating event-
triggered EMA and were not missing the negative affect measure or the social companionship
measure (Table 14). Parents accounted for the majority of these eating events (52.96%),
compared to children. Additionally, females accounted for the majority (65.16%) compared to
males. Ninety-four (32.75%) of the eating events took place on the weekend (Saturday/Sunday)
and 57 (19.86%) took place between 12:00AM-11:59AM. Two-thirds of eating events (n=185,

140
64.46%) took place with at least one other person, and the other one-third took place alone.
Ninety (31.36%) of the eating events were characterized as high negative affect.
A total of 3,451 affective states were reported by participants via time-triggered EMA
(Table 14).
11
Children reported 1,594 (46.19%), parents reported 1,362 (39.47%), and other
adult family members reported 495 (14.34%). About one-third (n=1,211, 35.09%) of the
affective states took place on the weekend (Saturday/Sunday) and 654 (18.95%) took place
between 12:00AM-11:59AM. Approximately one-third of reported affective states (n=1,202,
34.83%) were characterized as high negative affect.
Relational Event Models (Aim 1)
Table 15 summarizes the results of the REM, including Model 1 that includes the base
effects only, and three subsequent models (2 through 4) that incrementally include mechanisms
of interest. Model 4, which contains all of the mechanism terms, was selected as the final model
because it produced the lowest goodness of fit values (AIC=6265.81, BIC=6384.11).
Base effects. Compared to the reference event (low NA), high NA (MLE=0.28, SE=0.12),
eating with others/low NA (MLE=-2.29, SE=0.11), eating alone/low NA (MLE=-3.52, SE=0.20),
eating with others/high NA (MLE=-3.61, SE=0.21), and eating alone/high NA (MLE=-3.84,
SE=0.23) were all significantly less likely to occur (Table 15; Model 4).
Dynamic effects. Both of the affective state inertia terms were positive and significant:
the event sequences “low NA followed by ( →) low NA” (MLE=1.32, SE=0.12) and “high NA →
high NA” (MLE=1.04, SE=0.12) were significantly more likely to occur compared to the

11
The number of answered time-triggered EMAs in this study (i.e., affective states) (n=3,451) is
higher than the number of answered time-triggered EMAs reported in Dissertation Study 1
(n=3,328 (Table 6) because the eating event-triggered EMAs in which the participant reported
they were not eating (false positive) were converted to affective states for these analyses.

141
reference event (low NA), holding all other effects constant. Additionally, a majority of the
eating event inertia terms were positive and significant. The following event sequences:
• Eating alone/low NA → Eating alone/low NA (MLE=1.58, SE=0.57);
• Eating with others/high NA → Eating with others/high NA (MLE=1.98,
SE=0.48); and
• Eating alone/high NA → Eating alone/high NA (MLE=1.58, SE=0.76)
were more also likely to occur compared to the reference event (low NA states), holding all other
effects constant.
The effect representing the event sequence “low NA → eating with others/low NA” was
significantly less likely to occur (MLE=-0.78, SE=0.25), compared to the reference event (low
NA) (Table 15; Model 4). No other effects representing an affective state followed by an eating
event were significant (all p>0.05).
All effects that represented sequences of three events, i.e., an affective state followed by
an eating event followed by an affective state, were significant. The following sequences were all
more likely to occur, compared to the occurrence of low NA, holding all other effects constant:
• Low NA → Eating with others/low NA → Low NA (MLE=1.35, SE=0.23);
• Low NA → Eating alone/low NA → Low NA (MLE=1.89, SE=0.31);
• High NA → Eating with others/high NA → High NA (MLE=1.68, SE=0.30); and
• High NA → Eating alone/high NA → High NA (MLE=2.02, SE=0.41) (Table
15).

142
Differences by Family Role (Aim 2)
Multiple interaction terms containing the ‘parent’ dummy-coded variable  the final
model terms were significant (p<0.05), so the final model was stratified by family role (Table
16).
Some differences emerged once the model was stratified. Children were more than 5
times as likely (compared to the reference event) to experience the sequence “eating with
others/high NA → eating with others/high NA” (eating event inertia) (MLE=1.72, SE=0.50); and
more than 4 times as likely to experience “eating alone/high NA → eating alone/high NA”
(eating event inertia)” (MLE=1.59, SE=0.80). While these effects significantly explained the
observed sequence of events in children, they did not do so in parents. Children were also
significantly less likely to experience the sequence “high NA → eating with others/high NA”
(MLE=-1.18, SE=0.48), as compared to low NA and holding all other effects constant, while
parents were significantly more likely to experience this sequence (MLE=1.37, SE=0.66). All
stratified model results are reported in Table 16.
DISCUSSION
General Discussion
This study applied the Relational Event Modeling (REM) framework to ordinal event data
measured by wearable sensors and ecological momentary assessment (EMA). Multiple
mechanisms that predicted the sequence of in-home affective states and eating events were
identified. Various types of inertia effects, including affective state → affective state, eating
event → eating event, and affective state → eating event → affective state were observed in the
dataset. These findings indicate that the events and states being modeled were not randomly

143
sequenced, and that there was a tendency for similar types of states and events to persist over
time (rather than being followed by a different random event or state). Thus, the sequencing of
events is meaningful, not random, as one might expect with family systems.
Specifically, in all models, “affective state inertia” was prevalent and significantly explained
the observed sequence of events. Affective state inertia can be described as the tendency for an
affective state to be followed by the same type of affective state. This phenomenon has been
identified by others in the literature. For instance, an early study had male participants record
their mood via diary multiple times per day, across eight days; the study reported evidence of
affective inertia (i.e., bad mood carrying over to the next assessment) in this sample (Suls, Green,
& Hillis, 1998). More recently, EMA has been used to investigate affective dynamics in real-
time (Kamarck, Shiffman, & Wethington, 2011; Keng & Tong, 2016; Lamers et al., 2018;
Vansteelandt & Verbeke, 2016). Because affective inertia has been proposed as a depressive
symptom (Brose, Schmiedek, Koval, & Kuppens, 2015; Kuppens et al., 2012), the field of
research has focused a lot of attention on examining this effect in clinical populations (e.g.,
participants with a mood or anxiety disorder) or how it related to clinical outcomes (e.g., clinical
depression).
Fewer studies have examined the relationship between this inertia effect and health
behaviors, especially in non-clinical populations. One study also using EMA reported that
negative affective inertia was negatively associated with physical activity in a sample of college
students (Wen, An, Li, Du, & Xu, 2020); and another EMA study found that certain eating-
disorder diagnostic groups (e.g., those with anorexia nervosa, bulimia nervosa, or binge-eating
disorder) displayed negative affective inertia (Williams-Kerver et al., 2020). One likely reason
that few studies have examined the relationship between momentary affective inertia and eating

144
behavior could be that health behaviors are challenging to accurately measure in real-time
(Klurfeld, Hekler, Nebeker, Patrick, & Khoo, 2018; McClung et al., 2018; Spruijt-Metz et al.,
2018; Takemoto et al., 2018), alongside affective states. The M2FED study built a novel system
that can measure both affective states and automatically detect eating events in real time. In
addition to observing affective state inertia in our sample of family members, we found evidence
that affective states were followed by eating events that shared the same affective features.
Our study was also innovative in studying these phenomena in family systems, and we
detected differences in these sequences of events between parents and children, as hypothesized
in Study Aim 2. Parents were more likely to experience high NA followed by an eating event
with others/high NA, whereas that sequence was significantly less likely to occur in children.
Overall, this is unsurprising given that parents reported 45 (42.1%) eating events alone, and were
more likely to eat with others at home (n=107, 57.9%). It is also possible that parents feel an
obligation to eat with others while at home, despite their current affective state, but they have
different eating patterns and social contexts outside of the home.
Children, on the other hand, were also more likely to experience high NA → eating
alone/high NA → high NA; this sequence was not observed in the parent subsample, and was
therefore not included as a term in the parent model. These findings suggest that both parents and
children are likely to experience this affect-eating inertia effect, but in different social contexts
and in different sequences.
One EMA study using a sample of adults with binge eating disorder found that participants
who experienced “stress pileup” (continuous reports of stress) reported greater binge eating
symptoms (Smith et al., 2020). Although this was in a clinical sample, the results suggest that
this type of sequence of events can produce unhealthy or undesirable eating patterns. The results

145
from our study provide insight into the interplay of affective states and eating events; the
potential differences between different family roles; and the differences between social contexts,
with implications for future just-in-time behavioral interventions to promote healthy eating
patterns.
Intervention Implications
Our study was unable to examine the influence of other family member’s actions on an
individual’s affective states and/or eating events; however, a few recent studies have utilized
EMA to measure the relationship between momentary affect and eating behavior within family
systems. One study using EMA found that mothers and children who reported higher-than-
average negative affect consumed more pastries/sweets, children with higher-than-average
negative affect consumed more fast food, and mothers’ momentary positive affect predicted their
own fruit/vegetable consumption (Mason et al., 2019). Another EMA study found that higher
parental stress/depressed mood earlier in the day predicted fewer homemade foods served at
dinner that same night (Berge et al., 2017). Thus, the affective states of individuals within a
family and interpersonal influences are not only likely to influence family eating behavior, but
the types and quality of food consumed as well.
The potential interdependent relationship of affect/mood and obesogenic behaviors between
two family members (typically mother and child) has begun to be explored for eating behavior
(Mason et al., 2019) and physical activity behavior (Yang et al., 2020). However, future research
is needed to explore these independences beyond a dyad (i.e., two family members), and the
possible inter-personal effects of affect/stress on subsequent eating behavior in families.
The findings from this study provide insights into the dynamic nature of in-home affective
states and eating behavior within families, and taken together with the findings from the studies

146
reviewed above, can be used to uniquely inform just-in-time adaptive interventions (JITAIs) that
target eating behaviors in families. A JITAI is a type of intervention design that aims to provide
the right amount of support at the right time and in the right contexts to promote behavior change
(Nahum-Shani et al., 2015; Nahum-Shani et al., 2017; Spruijt-Metz & Nilsen, 2014). The results
from our relational event model offer key insights into the temporal progression (Nahum-Shani
et al., 2015) of key factors and contexts that precede “unhealthy” or “healthy” eating behaviors
and that can be potential targets for future intervention. For example, based on the M2FED
monitoring system, our study results, and evidence in the broader literature, if persistent
‘negative affect’ is detected among children/parents, then an intervention that interrupts negative
affect and/or stress accumulation could ultimately lead to healthier eating events.
Methodological Implications
The application of REM to study health behaviors in general is nascent. REM has often been
utilized to study to animal behavior (Patison, Quintane, Swain, Robins, & Pattison, 2015;
Tranmer, Marcum, Morton, Croft, & de Kort, 2015) and online/digital communications data
(Butts, 2008; Chen & Poquet, 2020; Foucault Welles, Vashevko, Bennett, & Contractor, 2014).
One other paper in the literature applying REM framework to investigate eating behavior has
been identified. Marcum and colleagues (2018) used the framework to investigate the micro-
processes that underlie food choice within a virtual buffet setting (Marcum et al., 2018). They
found that the order in which food items were placed in the virtual buffet (an aspect of “built
environment”) had an influential effect on the food that participants selected from the buffet;
also, they identified a significant “choice inertia” effect, such that participants were more likely
to select a particular food item if they had just served themselves that same food item (Marcum
et al., 2018). Our findings are the first to extend this work on eating behaviors by collecting

147
observations on micro-processes and sequences of states and events in-the-wild (i.e., outside of a
lab or controlled virtual setting).
Other applications in the literature have proposed ad hoc solutions to meet the REM
assumptions, and methodological extensions have been proposed to increase REM’s usability
and flexibility, including at least two extensions to account for multiple receivers (although both
of these are still in preprint) (Lerner, Tranmer, Mowbray, & Hancean, 2019; Mulder & Hoff,
2021). While early REM work focused on expanding the methods and applying REM to mock
datasets for illustration purposes (Butts, 2008), more recent work has applied REM to a variety
of relational event datasets. Although still a novel application, a few papers in addition to this
current study have examined the mechanisms that drive the occurrence of certain health
behaviors, such as play behavior (i.e., physical activity) in adolescents (Ellis-Davies, Lew-Levy,
Fleming, Boyette, & Baguley, 2021) and food choice behavior (Marcum et al., 2018). The REM
framework provides ample opportunity for methodological expansion and for the application to
the study of health behaviors. It’s application can lend new and unique insights into behavioral
health from a relational perspective.
Overall, the findings from this study suggest that the application of REM is a feasible
approach to examining intensive longitudinal data (ILD) collected by mHealth technologies,
including wearable sensors and mobile devices. Additionally, we contribute to the literature by
providing evidence for affective state inertia and eating event inertia operating within family
systems and discussing the implications for behavioral health interventions.
Limitations and Advantages
We observed low variability in the negative affect measure, which was measured in both
time-triggered and eating event-triggered EMA surveys and was ultimately used to characterize

148
the affective states and eating events in this study. We made the decision to dummy code this
variable rather than keep it as a continuous variable; however, in doing so we were unable to
observe a potentially more nuanced relationship between negative affect and eating events.
Secondly, because the scope of this study only covered in-home eating behavior, we observed
few eating events per person per day (reported in Dissertation Chapter 2). This informed our
decision to use the ordinal-approach to REM (vs. using interval time), but we do acknowledge
this approach limited our ability to examine certain temporal effects (e.g., examining the duration
between affective states and subsequent eating events). Relatedly, the statistical approach that we
took did not account for the breaks in the day (i.e., start of a new day) across the two-week
observation period, which should be addressed in future work. Despite its limitations, this study
took a unique methodological approach to modeling intensive longitudinal data on health
behaviors; and allowed us to better understand how the sequences of eating events and affective
states unfold within a family system.
Future Directions
Emerging technologies present the opportunity for measuring and examining sequences
of events (Pentland, Pentland, & Calantone, 2017) and/or contextual factors in real-time, and
applying new analytic methods such as REM for modeling the multiple and interdependent
signals from these complex systems. Specifically, technology-assisted eating assessment tools
offer the ability to explore eating behaviors/activities at various levels of granularity, such as
meal microstructure (Doulah et al., 2017), food choices (Marcum et al., 2018), and processes
(Bell et al., 2019; Ohkuma et al., 2015; Sharps et al., 2015), and their social influences. Recent
literature, along with the findings from this study, suggest that these complex eating behaviors

149
and their interaction with inter-personal factors may ultimately play an important role on food
selection, dietary intake, and thus, obesity and disease risk.
Another advantage of employing mobile health (mHealth) technologies for data
collection includes the ability to measure and model these behaviors and social mechanisms “in
the wild”. The majority of the research on the social influences on eating has taken place in
laboratory settings (vs. in the wild), and has relied on cross-sectional survey data and/or in-lab
measurement periods only spanning a few hours (Herman et al., 2019). For instance, a scoping
review conducted by Bell and colleagues (2020) report on two studies (Gemming et al., 2015a;
Vaizman, Ellis, & Lanckriet, 2017) that use wearable sensors and mobile devices to
automatically capture both eating behavior and social contexts in free-living environments (Bell
et al., 2020). Similar to our approach in the M2FED study, in Vaizman (2017), wrist-worn
smartwatches were used to automatically detect eating events, while mobile phones were used to
collect information on one’s social context while eating (e.g., with friends) (Vaizman et al.,
2017). In Gemming (2015), wearable cameras were used to automatically detect both eating
episodes and social interaction (defined as active conversations or body language/movements
and eye contact toward others) during eating episodes (Gemming et al., 2015a). Overall, these
technologies offer exciting new avenues of research that can take the broader system into
account when measuring and intervening on health behaviors.
CONCLUSION
In conclusion, wearable sensors and other mHealth technologies that produce “big data”
will be increasingly integrated into cyber-physical systems like M2FED to monitor complex
individual and group (family) phenomena that are relevant to health. The application of REM
and other social network analysis methods are likely to be very valuable for understanding these

150
new streams of data and elucidating unique insights into the influences (and potential leverage
points) on health behaviors.

151
TABLES
Table 14. Counts and row proportions of observed affective states and eating events

High Negative Affect Low Negative Affect Total
Affective State 1,202 (34.8%) 2,249 (65.2%) 3,451
Eat With Others 54 (29.2%) 131 (70.8%) 185
Eat Alone 36 (35.3%) 66 (64.7%) 102
Total 1,292 (34.6%) 2,446 (65.4%) 3,738

Table 15. Results from the egocentric relational event models (ordinal likelihood)

Model 1 Model 2 Model 3 Model 4
MLE / SE
Base Effects
HighNA -0.63*** -0.48*** -0.47*** -0.28*
0.025 0.066 0.081 0.12
EatWithOthers/LowNA -2.84*** -2.13*** -2.41*** -2.29***
0.064 0.078 0.11 0.11
EatAlone/LowNA -3.53*** -2.82*** -3.65*** -3.52***
0.088 0.099 0.20 0.20
EatWithOthers/HighNA -3.73*** -3.06*** -3.87*** -3.61***
0.097 0.11 0.20 0.21
EatAlone/HighNA -4.14*** -3.44*** -4.10*** -3.84***
0.12 0.13 0.22 0.23
Dynamic Effects
LowNA →LowNA 1.09*** 1.13*** 1.32***
0.069 0.090 0.12
HighNA →HighNA 1.21*** 1.22*** 1.04***
0.073 0.092 0.12
EatWithOthers/LowNA →EatWithOthers/LowNA -0.18 0.065 0.52
0.28 0.30 0.31
EatAlone/LowNA →EatAlone/LowNA -0.047 0.75 1.58**
0.52 0.55 0.57
EatWithOthers/HighNA →EatWithOthers/HighNA 0.83 1.60*** 1.98***
0.43 0.46 0.48
EatAlone/HighNA →EatAlone/HighNA 0.53 1.14 1.58*
0.73 0.75 0.76
Low NA →EatWithOthers/LowNA 0.50** -0.78**
0.16 0.25
Low NA →EatAlone/LowNA 1.27*** -0.56
152
Note: MLE=Maximum likelihood estimate; SE=standard error
***
p<0.001;
**
p<0.01;
*
p<0.05

0.24 0.37
HighNA →EatWithOthers/HighNA 1.60*** -0.33
0.23 0.36
HighNA →EatAlone/HighNA 1.39*** -0.88
0.27 0.47
Low NA →EatWithOthers/LowNA →Low NA 1.35***
0.23
Low NA →EatAlone/LowNA →Low NA 1.89***
0.31
HighNA →EatWithOthers/HighNA →HighNA 1.68***
0.30
HighNA →EatAlone/HighNA →HighNA 2.02***
0.41
Num. events 3738 3738 3738 3738
AIC 7225.51 6417.33 6351.87 6265.81
AICC 7225.52 6417.41 6352.00 6266.01
BIC 7256.64 6485.82 6445.27 6384.11
153
Table 16. Results from final relational event model, stratified by family role

Parents Children
MLE (SE) OR MLE (SE) OR
Base Effects
HighNA
-0.74***
(0.18)
0.48
0.21
(0.16)
1.23
EatWithOthers/LowNA
-1.84***
(0.15)
0.16
-2.70***
(0.18)
0.067
EatAlone/LowNA
-3.45***
(0.29)
0.032
-3.60***
(0.27)
0.027
EatWithOthers/HighNA
-5.15***
(0.53)
0.0058
-2.77***
(0.25)
0.062
EatAlone/HighNA
-4.23***
(0.36)
0.015
-3.40***
(0.31)
0.033
Dynamic Effects
Low NA →Low NA
1.02***
(0.19)
2.78
1.69***
(0.16)
5.41
HighNA →HighNA
1.02***
(0.19)
2.77
0.75***
(0.17)
2.12
EatWithOthers/LowNA →EatWithOthers/LowNA
0.14
(0.35)
1.16
-18.56
(9744.10)
0.00
EatAlone/LowNA →EatAlone/LowNA
1.40
(0.81)
4.06
1.87*
(0.79)
6.47
EatWithOthers/HighNA →EatWithOthers/HighNA
-16.43
(14815.89)
0.00
1.72***
(0.50)
5.58
EatAlone/HighNA →EatAlone/HighNA
-17.78
(19520.26 )
0.00
1.59*
(0.80)
4.88
Low NA →EatWithOthers/LowNA
-0.53
(0.30)
0.59
-1.53**
(0.49)
0.22
Low NA →EatAlone/LowNA -0.74 0.48 -0.41 0.67
154
(0.56) (0.51)
HighNA →EatWithOthers/HighNA
1.37*
(0.66)
3.94
-1.18*
(0.48)
0.31
HighNA →EatAlone/HighNA
0.49
(0.91)
1.64
-0.37
(0.52)
0.69
Low NA →EatWithOthers/LowNA →Low NA
0.78**
(0.28)
2.19
2.30***
(0.46)
9.93
Low NA →EatAlone/LowNA →Low NA
1.96***
(0.47)
7.07
1.90***
(0.43)
6.68
HighNA →EatWithOthers/HighNA →HighNA
1.71***
(0.42)
5.54
1.57***
(0.43)
4.80
HighNA →EatAlone/HighNA →HighNA
⎯
⎯
1.39**
(0.43)
4.02
AIC 2618.738 3584.061
AICC 2619.196 3584.405
BIC 2714.543 3692.495

Note: MLE=Maximum likelihood estimate; OR=odds ratio; SE=standard error
***
p<0.001;
**
p<0.01;
*
p<0.05
155

156
CHAPTER 5: DISCUSSION
SUMMARY OF FINDINGS
The Monitoring and Modeling Family Eating Dynamics (M2FED) study takes a
dramatically different mobile health (mHealth) approach to childhood obesity by not focusing
directly on diet and activity, but rather on family eating dynamics (FED). Many mHealth efforts
addressing childhood obesity have focused on measuring and intervening on diet intake and
activity (Turner, Spruijt-Metz, Wen, & Hingle, 2015), two of the most important behaviors that
directly affect body mass index (BMI) and central adiposity (accumulation of abdominal fat)
(Spruijt-Metz, 2011). However, this focus presents a number of challenges, including the
accurate assessment of dietary intake (i.e., what and how much is being eaten). Furthermore, this
approach ignores the decades of research indicating the importance of family systems and family
social networks on obesity-related behaviors (Bates et al., 2018; Birch & Davison, 2001; Patrick
& Nicklas, 2005; Savage et al., 2007). Therefore, this dissertation utilized data from the
M2FED study to address three main goals: (i) to gain deeper insight into how children and
families influence one another’s eating behaviors in real-time using novel technologies and
(ii) to develop new dynamic models of family eating behavior that depict how behavior
unfolds in context and in real-time, and thus (iii) to better inform future obesity prevention
and intervention strategies.
Dissertation Study 1 sought to address methodological gaps in the dietary assessment
literature described above. The M2FED study utilized a passive measurement tool, a smart-watch
wearable sensor, in order to objectively measure eating events rather than dietary intake (i.e.,
what and how much is being eaten). The findings from Study 1 indicate that this approach can be
used to automatically detect in-home eating activity with a high level of precision, compared to

157
other automatic dietary assessment technologies deployed in the field (Bell et al., 2020). We also
employed Ecological Momentary Assessment (EMA) to measure self-reported intra- and inter-
personal contexts during eating events. Inter-personal (social) context was a significant predictor
of EMA compliance, suggesting that interpersonal relations can be leveraged to increase
compliance to mHealth research studies.
The primary research question for Dissertation Study 2 was to investigate how social
context interacts with intra-personal context to influence eating behavior together in real-time.
Our analyses identified four distinct “contextualized” eating event classes in the data: two
“healthy” characterizations and two “unhealthy” characterizations, the latter of which was
significantly less likely to occur in social contexts, when others were present. The findings from
this study indicate that further research into real-time intra- and inter-personal features (e.g.,
negative affect, eating companionship) may yield Just-In-Time (JIT) points of intervention that
will help to promote healthier eating behaviors and avoid unhealthy eating behaviors within
families.
Dissertation Study 3 implemented a Relational Event Model (REM) approach to examine
how in-home affective states and eating behaviors unfold over time and in various contexts.
Multiple inertia effects, such as an affective state being followed by the same type of affective
state, and an eating event with certain features being followed by the same type of eating event,
were observed in the dataset. Some of the event sequences observed in the dataset differed by
family role. For instance, children were more likely than parents to experience high NA followed
by a high NA eating event alone, followed by high NA. We applied the REM framework as a
novel approach to analyzing and interpreting intensive longitudinal data (ILD) produced by
mHealth technologies.

158
IMPLICATIONS
The findings from this dissertation have important methodological, theoretical, and
intervention implications.
Methodological Implications
Firstly, Dissertation Study 1 extends the literature by showcasing the technical capability
of a real-time monitoring system of family eating behavior via smartwatch and mobile phone,
and improves upon limitations of past research. For instance, one persisting critical barrier to
dietary assessment in research settings is the ability to accurately measure dietary intake (many
measures have low validity due to recall bias) (Burrows et al., 2012; Livingstone et al., 2004).
The M2FED study utilizes a passive measurement tool, a smart-watch wearable sensor, in order
to objectively measure eating events rather than dietary intake in the field. Past studies have also
utilized wearable devices/sensors to automatically detect eating-related activity (Bell et al., 2020;
Vu et al., 2017). Limitations of this past work include in-lab testing, small sample sizes, short
length of deployment, and reliance on retrospective self-report methods for ground-truth
validation (Bell et al., 2020).
Our study improves upon past work by deploying our system in-the-wild and using
event-triggered EMA for ground-truth validation. Additionally, we deployed the system to a
relatively sizeable sample of diverse study participants for a two-week period of time, as
compared to other in-field studies. Overall, the findings from this dissertation indicate the
feasibility to monitor and model in-home family eating behavior and context with wearable
sensors and mobile phones.

159
Theoretical and Intervention Implications
Secondly, Dissertation Studies 2 and 3 advance scientific knowledge by identifying the
ways in which intra-personal and inter-personal factors, and the synergy of these factors,
influence individual and family eating behavior in the wild. The results from the latent class
analysis (Study 2) and the relational event model (Study 3) refine and develop our theoretical
understanding of eating behavior, and enable the identification of temporally specific processes
(i.e., mood inertia) and events within the family system that can be targeted for personalized,
context-specific, real-time feedback.
This approach aids us in understanding how family systems work and where to best
leverage them in order to promote healthier eating behaviors within the system. Specifically,
these results can inform family-based just-in-time adaptive interventions (JITAIs). Although the
influence of family context and characteristics on childhood obesity, obesity-related behaviors,
and treatment has been recognized for decades (Epstein et al., 2007; Epstein et al., 1994), and
there is evidence that interventions that target the family system may be most effective at
treating obesity (vs. those that target individuals), the recent Cochrane Review on interventions
to treat obesity in youth show that fewer than 25% of interventions are family-based or include
the family (Oude Luttikhuis et al., 2009) and fewer than 10% incorporate the home as an
intervention setting (T. Brown et al., 2019). The findings from this dissertation can be translated
into intervention strategies that intervene on cues or states that trigger eating behaviors or
synchronous patterns of eating among family members.
For example, the eating events within Latent Class 1 (from Study 2) were partially
characterized by high negative affect, eating in the absence of hunger (EAH), and were more
likely to be a shorter eating activity (snack/drink episode). We see the co-occurrence of two

160
intra-personal contexts that have previously been associated with unhealthy eating behavior
(negative affect and EAH) (Anestis et al., 2010; Engel et al., 2007; Fogel et al., 2018; Lansigan
et al., 2015; Mason et al., 2018). This finding could, for instance, be leveraged to develop a
family-based intervention that targets affect and stress reduction in family members across the
day to potentially decrease the likelihood of affect-induced snacking.
Furthermore, the results from Study 3 indicate that a “high negative affect” eating event
is likely to be both preceded by and followed by a negative affective state. Again, consider the
following intervention scenario: the M2FED system detects, via time-triggered EMA (or through
passive assessment), that a participant just reported a negative affective state, and it is
programmed to know that there is a tendency for this participant to report a negative affective
state preceding a “negative” eating event. If a wrist-worn smartwatch soon detects an eating
event following this negative affective state, then the system could provide a brief message via
mobile phone to discourage eating in the absence of hunger or ameliorate a high negative affect
state.
OVERALL DISSERTATION LIMITATIONS
1. In-home family meals are a common context for food consumption (Dallacker et al.,
2018; Neumark-Sztainer et al., 2003; Saad, 2013), which is why the M2FED study
focused on measuring and monitoring in-home family eating behavior. Because the scope
of the M2FED project was to examine in-home eating behavior, data collected by the
system were only measured in the home (i.e., no data on family eating events and
affective states were collected outside of the home). Therefore, the findings from this
dissertation may not be generalizable to all types of eating behavior, especially eating that
takes place outside of the home.

161
2. Although typically offered as a solution to traditional measurement tools, the mHealth
technologies utilized in the M2FED study are not without limitations. To some extent,
they are still susceptible to the same limitations as traditional measurement tools,
including self-report biases and participant burden (Spruijt-Metz et al., 2018).
Additionally, mHealth technologies have their own unique challenges, such as
technology usability and possible malfunctions, cost, suitability for low-literacy
populations, and decreased compliance and attrition rates due to respondent burden (as
seen in Dissertation Study 1) (McClung et al., 2018; Spruijt-Metz et al., 2018).
3. Recruitment of entire families was a challenging task, and this resulted in a smaller
sample size of families than initially planned. This may have limited our ability to detect
signals from the data.
4. Although the M2FED study embraced the systems approach in examining the varying
levels of influence on eating behaviors (individual level and social environment level),
there are still many other levels of influence (e.g., physical environment, macro-level
environments) that were beyond the scope of the project and were not explored. The
results from this dissertation should be interpreted with this in mind.
FUTURE RESEARCH DIRECTIONS
mHealth and Behavioral Health Fields
Broadly speaking, mHealth technologies are increasingly able to passively detect health
behaviors including and beyond eating behavior (i.e., physical activity, sleep, smoking), either
standalone or in conjunction with other behaviors (de Zambotti, Cellini, Goldstone, Colrain, &
Baker, 2019; Imtiaz, Ramos-Garcia, Wattal, Tiffany, & Sazonov, 2019; O’Reilly & Spruijt-

162
Metz, 2013). These technologies also are increasingly able to collect vast amounts of information
on the multiple contexts in which an individual is embedded, and, data from smartphones and
wearable technologies can be integrated with other “big data” to more broadly assess an
important level of influence – the food environment. Big data previously used in this manner
include retail sales data, geospatial data, social media data, and transport data (Timmins, Green,
Radley, Morris, & Pearce, 2018).
These technologies and streams of data can enable public health researchers to recognize and
subsequently treat the obesity epidemic as a complex systems problem, and can shift our
approach toward obesity prevention and intervention strategies. This is an important and
necessary shift because, to date, the impacts of behavioral interventions to prevent and treat
obesity have been modest at best.
Despite decades of research on childhood obesity prevention and intervention strategies, the
overall prevalence of obesity has continued to rise in both U.S. children and adults (Han et al.,
2010; Ng et al., 2014; Skinner et al., 2018; The GBD 2015 Obesity Collaborators, 2017) –
although, the prevalence has decreased or leveled off for certain age groups in recent years (e.g.,
children aged 2 to 5 years old and children aged 6 to 11 years old, respectively) (Ogden et al.,
2016). Furthermore, strategies that target individual-level behavior have for the most part failed
to change one’s long-term level of adiposity (T. Brown et al., 2019; Oude Luttikhuis et al., 2009;
Waters et al., 2011). Embracing conceptual models that recognize obesity as a complex, multi-
level, and multi-factorial problem is seemingly the only path forward.
With this influx of data from technologies, however, comes many challenges. There may be
difficulties working successfully in interdisciplinary research teams as well as deciding how to
operationalize and model these incredibly rich data so that they can tell us something useful and

163
help us improve people’s lives. It will be necessary to continuously refine and update our
approach to this interdisciplinary work, but the insights we can glean and the impact we can
potentially have will be worth the effort.
In conclusion, using a variety of big data streams from emerging technologies can help us
gain deeper understanding of complex eating behaviors and the environments in which they
occur. Thus, this approach also can help better inform eating behavior interventions and health
policies, and ultimately, improving people’s health and quality of life.
Broader Public Health Field
There are two emerging areas within public health research broadly related to the topics
and themes of this dissertation, namely, (i) using a systems approach to better address health
disparities and (ii) the impact of climate change on eating behaviors and vice versa.
Systems approach to address health disparities
The application of systems science theory and methods to understand eating behavior,
and specifically family eating behavior, necessarily shifts the paradigm of how public health
researchers and professionals approach obesity prevention and intervention strategies, yet this
approach remains underutilized in the public health field (Luke & Stamatakis, 2012). Traditional
health behavior models, such as the Health Belief Model (Champion & Skinner, 2008) and the
Theory of Planned Behavior (Ajzen, 1991), suggest that an individual’s behaviors are a result of
one’s thoughts, beliefs, and intentions. However, this oversimplification of the problem does not
reflect the current state of scientific evidence (Hammond, 2009; Lee et al., 2017; Story et al.,
2008; Vandenbroeck et al., 2007).

164
Basic social conditions and racism have been established as fundamental causes of health
disparities based on socioeconomic status (SES) and race, respectively (Ford, Griffith, Bruce, &
Gilbert, 2019; Link & Phelan, 1995; Phelan & Link, 2015). There is also a growing literature that
aims to look at how the intersection of these two social identities impact health disparities
(Evans, Williams, Onnela, & Subramanian, 2018; Green, Evans, & Subramanian, 2017;
Lapalme, Haines-Saah, & Frohlich, 2019). Even though past research on SES-based and race-
based health disparities has helped document the “persistent gaps in morbidity and mortality”
among these groups (Bassett, 2015), research routinely fails to address the role of racism, and
intersecting forms of oppression, on health disparities (Boyd, Lindo, Weeks, & McLemore,
2020).
Using a systems science approach provides the opportunity to study and better understand
the complex factors and interdependences that play a role in health disparities. In fact, systems
science was identified as one of the three major opportunities in the 21
st
Century to understand
the etiology of health disparities and guide intervention development (Zhang et al., 2017). Fink,
Keyes, and Cerdá (2016) posit that a systems science framework can contribute to the
understanding of health disparities in three ways:
1. Modeling how health inequalities shape population health;
2. Understanding feedback between micro- and macro-level processes; and
3. Identifying targets for intervention to reduce inequalities in health (Fink,
Keyes, & Cerdá, 2016).
Health disparity interventions to date have primarily focused on modifying individual
factors, thereby limiting the ability for sustained change (A. F. Brown et al., 2019). Systems
science approaches can be used to inform evidence-based, multi-level, structural interventions

165
that address the structural determinants of health and health disparities (A. F. Brown et al., 2019;
Carey & Crammond, 2015; Trickett & Beehler, 2013). This approach has the opportunity to
change the narrative surrounding the fundamental causes of health disparities, and to reveal
powerful “system intervention points” (Carey & Crammond, 2015) that could potentially
improve population health and advance health equity.
Use of this approach does come with challenges. ‘Upstream’ governmental, policy, and
structural changes can take many years, or even decades, to accomplish. Also, the results from
systems science work needs to be made understandable and accessible to policy makers and
change agents (Dubowitz et al., 2016).
By using a systems science approach in this dissertation and investigating the family
system-level factors (beyond individual-level factors) that play a role in family eating behavior,
we contribute to changing the narrative surrounding obesity and health disparities, and highlight
the role that systems science can play in understanding the complex drivers of obesity.
The intersection of climate change, eating behaviors, and health
Relatedly, one of these system-level influences on health outcomes that can no longer be
ignored is climate change. The American Public Health Association (APHA) has stated that
climate change is one of the greatest threats to health America has ever faced—and that it is a
true public health emergency (Benjamin, 2019). There will undoubtedly be short- and long-term
impacts that climate change will have on health behaviors, disease risk, and human health (Watts
et al., 2021). The IPCC Special Report on Climate Change and Land makes clear that the effects
of climate change have already begun to and will continue to have increasingly dire
consequences on food security and human health, among many other adverse consequences

166
(IPCC, 2019). This intersection has recently been coined “The Global Syndemic” of obesity,
undernutrition, and climate change by The Lancet Commission (Swinburn et al., 2019).
Food and eating behaviors will play a key role in addressing climate change and ensuring
long-term public health. The EAT Lancet Commission has recently proposed that food – i.e.,
individual-level eating behavior and system-level food production – has the potential to be the
single strongest lever to optimize human health and environmental sustainability in the midst of
climate change (Willett et al., 2019). A growing body of literature suggests that a ‘planetary
health diet’ – rich in plant-based foods and with fewer animal source foods – confer both
improved health benefits, such as lower obesity risk, and environmental benefits, such as lower
greenhouse gas emissions (Semba et al., 2020; Willett et al., 2019). While the type of eating
behavior changes (e.g., increased fruit and vegetable intake) and food system changes (e.g.,
diversified farming methods) that are needed have been proposed by the Lancet Commission and
others, how these behavior changes will be enacted on both a local and global scale is still under
investigation.
A substantial amount of work has been done to identify and characterize the ways in
which the physical environment (e.g., food accessibility, built food environment) and the social
environment (e.g., social networks, social norms, social cues) can ultimately influence one’s
eating behavior. Understanding where to best leverage these complex environments can
ultimately promote healthier and sustainable behaviors and systems.
Building on past work, future work should investigate how social environments (i.e.,
social networks) and physical environments (i.e., food environments) can be leveraged to
promote heathy and sustainable eating behaviors and thus improve both population and
planetary health. The wide-scale consumption of healthy foods and sustainable foods (e.g., plant-

167
based, low red meat intake) produces co-benefits: these foods (i) improve human health by
decreasing cardiovascular disease and other chronic disease risk and (ii) improve planetary
health by decreasing greenhouse gas emissions and mitigating climate change.
Climate change and its imminent effects on population health is an urgent, complex issue
that requires rapid and sustainable solutions. Public health researchers need to quickly and
efficiently identify ways to encourage a nationwide transition towards healthy and sustainable eating
behaviors and food environments. However, a recent meta-analysis found that behavioral
interventions deployed to promote more sustainable (“pro-environmental”) behaviors (e.g.,
recycling, decreasing meat intake) had little to no effect on adoption of these behaviors during
the intervention, and led to no sustained behavior changes after the intervention ended (Nisa,
Bélanger, Schumpe, & Faller, 2019). Novel research technologies (e.g., wearable and deployable
sensors, mobile phones), data collection methodologies (e.g., ecological momentary assessment),
and intervention designs (e.g., multiphase optimization strategies, JITAIs) offer promising
methodological solutions to urgently address this public health emergency. These technologies
and designs can be used to create pro-environmental behavioral interventions that are efficient,
scalable, and highly effective.
In conclusion, a major global transition towards healthy and sustainable eating behaviors
and food environments presents a major opportunity for improving both human and planetary
health. This transition has the potential to improve human health by improving food security and
dietary intake and decreasing chronic disease risk; and can improve planetary health by
decreasing greenhouse gas emissions from food systems and mitigating climate change. The
mHealth technologies and adaptive intervention designs described throughout this dissertation
can be applied to promote behaviors that improve both human health and planetary health.

168
REFERENCES
Afshin, A., Sur, P. J., Fay, K. A., Cornaby, L., Ferrara, G., Salama, J. S., . . . Murray, C. J. L.
(2019). Health effects of dietary risks in 195 countries, 1990-2017: a systematic analysis
for the Global Burden of Disease Study 2017. The Lancet, 393(10184), 1958-1972.
doi:10.1016/S0140-6736(19)30041-8
Ajzen, I. (1991). The theory of planned behavior. Organ Behav Hum Decis Process, 50(2), 179-
211.
Al‐Khudairy, L., Loveman, E., Colquitt, J. L., Mead, E., Johnson, R. E., Fraser, H., . . . et al.
(2017). Diet, physical activity and behavioural interventions for the treatment of
overweight or obese adolescents aged 12 to 17 years. Cochrane Database of Systematic
Reviews(6). doi:10.1002/14651858.CD012691
Althubaiti, A. (2016). Information bias in health research: definition, pitfalls, and adjustment
methods. Journal of multidisciplinary healthcare, 9, 211-217.
doi:10.2147/JMDH.S104807
Anestis, M. D., Selby, E. A., Crosby, R. D., Wonderlich, S. A., Engel, S. G., & Joiner, T. E.
(2010). A comparison of retrospective self-report versus ecological momentary
assessment measures of affective lability in the examination of its relationship with
bulimic symptomatology. Behaviour Research and Therapy, 48(7), 607-613.
doi:https://doi.org/10.1016/j.brat.2010.03.012
Anzman, S. L., Rollins, B. Y., & Birch, L. L. (2010). Parental influence on children's early
eating environments and obesity risk: implications for prevention. Int J Obes (Lond),
34(7), 1116-1124. doi:10.1038/ijo.2010.43

169
Archer, E., & Blair, S. N. (2015). Implausible Data, False Memories, and the Status Quo in
Dietary Assessment. Advances in Nutrition, 6(2), 229-230. doi:10.3945/an.114.007799
Archer, E., Hand, G. A., & Blair, S. N. (2013). Validity of U.S. nutritional surveillance:National
Health and Nutrition Examination Survey caloric energy intake data, 1971-2010. PLoS
One, 8(10), e76632. doi:10.1371/journal.pone.0076632
Askari, M., Heshmati, J., Shahinfar, H., Tripathi, N., & Daneshzad, E. (2020). Ultra-processed
food and the risk of overweight and obesity: a systematic review and meta-analysis of
observational studies. International Journal of Obesity, 44(10), 2080-2091.
doi:10.1038/s41366-020-00650-z
Asparouhov, T., Hamaker, E. L., & Muthén, B. (2017). Dynamic Latent Class Analysis.
Structural Equation Modeling: A Multidisciplinary Journal, 24(2), 257-269.
doi:10.1080/10705511.2016.1253479
Ayala, G. X., Baquero, B., Arredondo, E. M., Campbell, N., Larios, S., & Elder, J. P. (2007).
Association between family variables and Mexican American children’s dietary
behaviors. Journal of Nutrition Education and Behavior, 39(2), 62-69.
Bassett, M. T. (2015). #BlackLivesMatter — A Challenge to the Medical and Public Health
Communities. New England Journal of Medicine, 372(12), 1085-1087.
doi:10.1056/NEJMp1500529
Bates, C. R., Buscemi, J., Nicholson, L. M., Cory, M., Jagpal, A., & Bohnert, A. M. (2018).
Links between the organization of the family home environment and child obesity: a
systematic review. Obesity Reviews, 19(5), 716-727.
doi:https://doi.org/10.1111/obr.12662

170
Bauer, K. W., Hearst, M. O., Escoto, K., Berge, J. M., & Neumark-Sztainer, D. (2012). Parental
employment and work-family stress: Associations with family food environments. Social
Science & Medicine, 75(3), 496-504. doi:https://doi.org/10.1016/j.socscimed.2012.03.026
Beaton, G. H., Burema, J., & Ritenbaugh, C. (1997). Errors in the interpretation of dietary
assessments. The American Journal of Clinical Nutrition, 65(4), 1100S-1107S.
doi:10.1093/ajcn/65.4.1100S
Bell, B. M., Alam, R., Alshurafa, N., Thomaz, E., Mondol, A. S., de la Haye, K., . . . Spruijt-
Metz, D. (2020). Automatic, wearable-based, in-field eating detection approaches for
public health research: a scoping review. npj Digital Medicine, 3(1), 38.
doi:10.1038/s41746-020-0246-2
Bell, B. M., Spruijt-Metz, D., Vega Yon, G. G., Mondol, A. S., Alam, R., Ma, M., . . . De la
Haye, K. (2019). Sensing eating mimicry among family members. Translational
Behavioral Medicine, 9(3), 422-430. doi:10.1093/tbm/ibz051
Benjamin, G. C. (2019). APHA supports work to drive action on climate change and health
[Press release]. Retrieved from https://www.apha.org/News-and-Media/News-
Releases/APHA-News-Releases/2019/APHA-supports-action-on-climate-change
Berge, J. M., & Everts, J. C. (2011). Family-Based Interventions Targeting Childhood Obesity:
A Meta-Analysis. Childhood Obesity, 7(2), 110-121.
doi:10.1089/chi.2011.07.02.1004.berge
Berge, J. M., Tate, A., Trofholz, A., Fertig, A. R., Miner, M., Crow, S., & Neumark-Sztainer, D.
(2017). Momentary Parental Stress and Food-Related Parenting Practices. Pediatrics,
140(6), e20172295. doi:10.1542/peds.2017-2295

171
Beshara, M., Hutchinson, A. D., & Wilson, C. (2013). Does mindfulness matter? Everyday
mindfulness, mindful eating and self-reported serving size of energy dense foods among
a sample of South Australian adults. Appetite, 67, 25-29.
doi:https://doi.org/10.1016/j.appet.2013.03.012
Best, J. R., Goldschmidt, A. B., Mockus-Valenzuela, D. S., Stein, R. I., Epstein, L. H., &
Wilfley, D. E. (2016). Shared weight and dietary changes in parent–child dyads following
family-based obesity treatment. Health Psychology, 35(1), 92-95.
doi:10.1037/hea0000247
Bin Morshed, M., Kulkarni, S. S., Li, R., Saha, K., Roper, L. G., Nachman, L., . . . Abowd, G. D.
(2020). A Real-Time Eating Detection System for Capturing Eating Moments and
Triggering Ecological Momentary Assessments to Obtain Further Context: System
Development and Validation Study. JMIR Mhealth Uhealth, 8(12), e20625.
doi:10.2196/20625
Birch, L. L., & Davison, K. K. (2001). FAMILY ENVIRONMENTAL FACTORS
INFLUENCING THE DEVELOPING BEHAVIORAL CONTROLS OF FOOD
INTAKE AND CHILDHOOD OVERWEIGHT. Pediatric Clinics, 48(4), 893-907.
doi:10.1016/S0031-3955(05)70347-3
Bleich, S. N., Thorpe, R. J., Sharif-Harris, H., Fesahazion, R., & LaVeist, T. A. (2010). Social
context explains race disparities in obesity among women. Journal of Epidemiology and
Community Health, 64(5), 465. doi:10.1136/jech.2009.096297
Bolck, A., Croon, M., & Hagenaars, J. (2004). Estimating Latent Structure Models with
Categorical Variables: One-Step Versus Three-Step Estimators. Political Analysis, 12(1),
3-27.

172
Booth, B. M., Mundnich, K., Feng, T., Nadarajan, A., Falk, T. H., Villatte, J. L., . . . Narayanan,
S. (2019). Multimodal Human and Environmental Sensing for Longitudinal Behavioral
Studies in Naturalistic Settings: Framework for Sensor Selection, Deployment, and
Management. J Med Internet Res, 21(8), e12832. doi:10.2196/12832
Boushey, C. J., Spoden, M., Zhu, F. M., Delp, E. J., & Kerr, D. A. (2017). New mobile methods
for dietary assessment: review of image-assisted and image-based dietary assessment
methods. Proceedings of the Nutrition Society, 76(3), 283-294.
doi:10.1017/S0029665116002913
Boyd, R. W., Lindo, E. G., Weeks, L. D., & McLemore, M. R. (2020). On Racism: A New
Standard For Publishing On Racial Health Inequities. Health Affairs Blog.
doi:10.1377/hblog20200630.939347
Brose, A., Schmiedek, F., Koval, P., & Kuppens, P. (2015). Emotional inertia contributes to
depressive symptoms beyond perseverative thinking. Cognition and Emotion, 29(3), 527-
538. doi:10.1080/02699931.2014.916252
Brown, A. F., Ma, G. X., Miranda, J., Eng, E., Castille, D., Brockie, T., . . . Trinh-Shevrin, C.
(2019). Structural Interventions to Reduce and Eliminate Health Disparities. American
Journal of Public Health, 109(S1), S72-S78. doi:10.2105/AJPH.2018.304844
Brown, T., Moore, T. H. M., Hooper, L., Gao, Y., Zayegh, A., Ijaz, S., . . . et al. (2019).
Interventions for preventing obesity in children. Cochrane Database of Systematic
Reviews(7). doi:10.1002/14651858.CD001871.pub4
Buchowski, M. S. (2014). Doubly Labeled Water Is a Validated and Verified Reference Standard
in Nutrition Research. The Journal of Nutrition, 144(5), 573-574.
doi:10.3945/jn.114.191361

173
Burrows, T., Golley, R. K., Khambalia, A., McNaughton, S. A., Magarey, A., Rosenkranz, R. R.,
. . . Collins, C. (2012). The quality of dietary intake methodology and reporting in child
and adolescent obesity intervention trials: a systematic review. Obesity Reviews, 13(12),
1125-1138. doi:https://doi.org/10.1111/j.1467-789X.2012.01022.x
Butts, C. T. (2008). A relational event framework for social action. Sociological Methodology,
38(1), 155-200.
Butts, C. T. (2015). relevant: Relational Event Models (Version R package version 1.0-4).
Butts, C. T., & Marcum, C. S. (2017). A Relational Event Approach to Modeling Behavioral
Dynamics. In A. Pilny & M. S. Poole (Eds.), Group Processes: Data-Driven
Computational Approaches (pp. 51-92). Cham: Springer International Publishing.
Cameron, A. J., Crawford, D. A., Salmon, J., Campbell, K., McNaughton, S. A., Mishra, G. D.,
& Ball, K. (2011). Clustering of Obesity-Related Risk Behaviors in Children and Their
Mothers. Annals of Epidemiology, 21(2), 95-102.
doi:https://doi.org/10.1016/j.annepidem.2010.11.001
Cardello, A. V., Schutz, H. G., Lesher, L. L., & Merrill, E. (2005). Development and testing of a
labeled magnitude scale of perceived satiety. Appetite, 44(1), 1-13.
Carey, G., & Crammond, B. (2015). Systems change for the social determinants of health. BMC
Public Health, 15(1), 662. doi:10.1186/s12889-015-1979-8
Casey, B. J., Jones, R. M., Levita, L., Libby, V., Pattwell, S. S., Ruberry, E. J., . . . Somerville, L.
H. (2010). The storm and stress of adolescence: Insights from human imaging and mouse
genetics. Developmental Psychobiology, 52(3), 225-235. doi:10.1002/dev.20447

174
CDC Division of Nutrition, Physical Activity, and Obesity. (2018). Children's BMI Tool for
Schools. Retrieved from
https://www.cdc.gov/healthyweight/assessing/bmi/childrens_bmi/tool_for_schools.html
Centers for Disease Control and Prevention. (2018). About Child & Teen BMI. Retrieved from
https://www.cdc.gov/healthyweight/assessing/bmi/childrens_bmi/about_childrens_bmi.ht
ml
Centers for Disease Control and Prevention. (2020). About Adult BMI. Retrieved from
https://www.cdc.gov/healthyweight/assessing/bmi/adult_bmi/index.html
Cetateanu, A., & Jones, A. (2016). How can GPS technology help us better understand exposure
to the food environment? A systematic review. SSM - Population Health, 2, 196-205.
doi:https://doi.org/10.1016/j.ssmph.2016.04.001
Chae, W., Ju, Y. J., Shin, J., Jang, S.-I., & Park, E.-C. (2018). Association between eating
behaviour and diet quality: eating alone vs. eating with others. Nutrition Journal, 17(1),
117. doi:10.1186/s12937-018-0424-0
Champion, V. L., & Skinner, C. S. (2008). Chapter 3: The Health Belief Model. In Health
Behavior and Health Education (4th ed.). San Francisco, CA: Jossey-Bass.
Chen, B., & Poquet, O. (2020). Socio-temporal dynamics in peer interaction events. Paper
presented at the Proceedings of the Tenth International Conference on Learning Analytics
& Knowledge, Frankfurt, Germany. https://doi.org/10.1145/3375462.3375535
Cialdini, R. B., & Goldstein, N. J. (2004). Social Influence: Compliance and Conformity. Annual
Review of Psychology, 55(1), 591-621. doi:10.1146/annurev.psych.55.090902.142015

175
Cialdini, R. B., & Trost, M. R. (1998). Social influence: Social norms, conformity and
compliance. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social
psychology (4th ed., pp. 151-192). New York, NY, US: McGraw-Hill.
Cohen, S., Kamarck, T., & Mermelstein, R. (1983). A global measure of perceived stress.
Journal of Health & Social Behavior, 24(4), 385-396.
Collins, L. M., & Lanza, S. T. (2009). Latent class and latent transition analysis: With
applications in the social, behavioral, and health sciences. Hoboken, New Jersey: John
Wiley & Sons.
Colquitt, J. L., Loveman, E., O'Malley, C., Azevedo, L. B., Mead, E., Al‐Khudairy, L., . . . Rees,
K. (2016). Diet, physical activity, and behavioural interventions for the treatment of
overweight or obesity in preschool children up to the age of 6 years. Cochrane Database
of Systematic Reviews(3). doi:10.1002/14651858.CD012105
Contento, I. R., Williams, S. S., Michela, J. L., & Franklin, A. B. (2006). Understanding the food
choice process of adolescents in the context of family and friends. Journal of Adolescent
Health, 38(5), 575-582. doi:https://doi.org/10.1016/j.jadohealth.2005.05.025
Cruwys, T., Bevelander, K. E., & Hermans, R. C. J. (2015). Social modeling of eating: A review
of when and why social influence affects food intake and choice. Appetite, 86, 3-18.
doi:https://doi.org/10.1016/j.appet.2014.08.035
Dallacker, M., Hertwig, R., & Mata, J. (2018). The frequency of family meals and nutritional
health in children: a meta-analysis. Obesity Reviews, 19(5), 638-653.
doi:10.1111/obr.12659

176
Dallacker, M., Hertwig, R., & Mata, J. (2019). Quality matters: A meta-analysis on components
of healthy family meals. Health Psychology, 38(12), 1137-1149.
doi:http://dx.doi.org.libproxy2.usc.edu/10.1037/hea0000801
Dattilo, A. M., Birch, L., Krebs, N. F., Lake, A., Taveras, E. M., & Saavedra, J. M. (2012). Need
for Early Interventions in the Prevention of Pediatric Overweight: A Review and
Upcoming Directions. Journal of Obesity, 2012, 123023. doi:10.1155/2012/123023
de Castro, J. M. (1994). Family and friends produce greater social facilitation of food intake than
other companions. Physiology & Behavior, 56(3), 445-455.
de Castro, J. M. (1995). Social facilitation of food intake in humans. Appetite, 24(3), 260.
doi:https://doi.org/10.1016/S0195-6663(95)99835-7
de Castro, J. M., & de Castro, E. S. (1989). Spontaneous meal patterns of humans: influence of
the presence of other people. The American Journal of Clinical Nutrition, 50(2), 237-247.
doi:10.1093/ajcn/50.2.237
de Zambotti, M., Cellini, N., Goldstone, A., Colrain, I. M., & Baker, F. C. (2019). Wearable
Sleep Technology in Clinical and Research Settings. Medicine and science in sports and
exercise, 51(7), 1538-1557. doi:10.1249/MSS.0000000000001947
Delormier, T., Frohlich, K. L., & Potvin, L. (2009). Food and eating as social practice –
understanding eating patterns as social phenomena and implications for public health.
Sociology of Health & Illness, 31(2), 215-228. doi:10.1111/j.1467-9566.2008.01128.x
Diep, C. S., Hingle, M., Chen, T. A., Dadabhoy, H. R., Beltran, A., Baranowski, J., . . .
Baranowski, T. (2015). The Automated Self-Administered 24-Hour Dietary Recall for
Children, 2012 Version, for youth aged 9 to 11 years: A validation study. J Acad Nutr
Diet, 115(10), 1591-1598. doi:10.1016/j.jand.2015.02.021

177
Dombrowski, S. U., Knittle, K., Avenell, A., Araújo-Soares, V., & Sniehotta, F. F. (2014). Long
term maintenance of weight loss with non-surgical interventions in obese adults:
systematic review and meta-analyses of randomised controlled trials. BMJ : British
Medical Journal, 348, g2646. doi:10.1136/bmj.g2646
Doulah, A., Farooq, M., Yang, X., Parton, J., McCrory, M. A., Higgins, J. A., & Sazonov, E.
(2017). Meal Microstructure Characterization from Sensor-Based Food Intake Detection.
Frontiers in Nutrition, 4, 31. doi:10.3389/fnut.2017.00031
Dubowitz, T., Orleans, T., Nelson, C., May, L. W., Sloan, J. C., & Chandra, A. (2016). Creating
Healthier, More Equitable Communities By Improving Governance And Policy. Health
Affairs, 35(11), 1970-1975. doi:10.1377/hlthaff.2016.0608
Dunton, G. F., Dzubur, E., Huh, J., Belcher, B. R., Maher, J. P., O’Connor, S., & Margolin, G.
(2016). Daily Associations of Stress and Eating in Mother–Child Dyads. Health
Education & Behavior, 44(3), 365-369. doi:10.1177/1090198116663132
Dzubur, E., Huh, J., Maher, J. P., Intille, S. S., & Dunton, G. F. (2018). Response patterns and
intra-dyadic factors related to compliance with ecological momentary assessment among
mothers and children. Translational Behavioral Medicine, 8(2), 233-242.
doi:10.1093/tbm/ibx002
Egger, G., & Swinburn, B. (1997). An “ecological” approach to the obesity pandemic. BMJ,
315(7106), 477-480. doi:10.1136/bmj.315.7106.477
Ellis-Davies, K., Lew-Levy, S., Fleming, E., Boyette, A. H., & Baguley, T. (2021).
Demonstrating the Utility of Egocentric Relational Event Modeling Using Focal Follow
Data from Congolese BaYaka Children and Adolescents Engaging in Work and Play.
Field Methods, 1525822X20987073. doi:10.1177/1525822X20987073

178
Elliston, K. G., Ferguson, S. G., Schüz, N., & Schüz, B. (2017). Situational cues and momentary
food environment predict everyday eating behavior in adults with overweight and
obesity. Health Psychology, 36(4), 337-345. doi:10.1037/hea0000439
Encyclopedia of Survey Research Methods. (2008). In P. J. Lavrakas (Ed.), (Vol. 1). Thousand
Oaks, California: Sage Publications, Inc.
Engel, S. G., Boseck, J. J., Crosby, R. D., Wonderlich, S. A., Mitchell, J. E., Smyth, J., . . .
Steiger, H. (2007). The relationship of momentary anger and impulsivity to bulimic
behavior. Behaviour Research and Therapy, 45(3), 437-447.
doi:https://doi.org/10.1016/j.brat.2006.03.014
Engel, S. G., Crosby, R. D., Thomas, G., Bond, D., Lavender, J. M., Mason, T., . . . Wonderlich,
S. A. (2016). Ecological Momentary Assessment in Eating Disorder and Obesity
Research: a Review of the Recent Literature. Current Psychiatry Reports, 18(4), 37.
doi:10.1007/s11920-016-0672-7
Epel, E., Lapidus, R., McEwen, B., & Brownell, K. (2001). Stress may add bite to appetite in
women: a laboratory study of stress-induced cortisol and eating behavior.
Psychoneuroendocrinology, 26(1), 37-49. doi:https://doi.org/10.1016/S0306-
4530(00)00035-4
Epstein, L. H., Paluch, R. A., Roemmich, J. N., & Beecher, M. D. (2007). Family-based obesity
treatment, then and now: Twenty-five years of pediatric obesity treatment. Health
Psychology, 26(4), 381-391. doi:10.1037/0278-6133.26.4.381
Epstein, L. H., Valoski, A., Wing, R. R., & McCurley, J. (1994). Ten-year outcomes of
behavioral family-based treatment for childhood obesity. Health Psychol, 13(5), 373-383.
doi:10.1037//0278-6133.13.5.373

179
Evans, C. R., Williams, D. R., Onnela, J.-P., & Subramanian, S. V. (2018). A multilevel
approach to modeling health inequalities at the intersection of multiple social identities.
Social Science & Medicine, 203, 64-73.
doi:https://doi.org/10.1016/j.socscimed.2017.11.011
Fink, D. S., Keyes, K. M., & Cerdá, M. (2016). Social Determinants of Population Health: A
Systems Sciences Approach. Current epidemiology reports, 3(1), 98-105.
doi:10.1007/s40471-016-0066-8
Fisher, J. O., & Birch, L. L. (2002). Eating in the absence of hunger and overweight in girls from
5 to 7 y of age. American Journal of Clinical Nutrition., 76(1), 226-231.
Flegal, K. M., Carroll, M. D., Kuczmarski, R. J., & Johnson, C. L. (1998). Overweight and
obesity in the United States: prevalence and trends, 1960–1994. International Journal of
Obesity, 22(1), 39-47. doi:10.1038/sj.ijo.0800541
Flegal, K. M., Carroll, M. D., Ogden, C. L., & Johnson, C. L. (2002). Prevalence and Trends in
Obesity Among US Adults, 1999-2000. JAMA, 288(14), 1723-1727.
doi:10.1001/jama.288.14.1723
Fogel, A., McCrickerd, K., Fries, L. R., Goh, A. T., Quah, P. L., Chan, M. J., . . . Forde, C. G.
(2018). Eating in the absence of hunger: Stability over time and associations with eating
behaviours and body composition in children. Physiology & Behavior, 192, 82-89.
doi:https://doi.org/10.1016/j.physbeh.2018.03.033
Fontana, J. M., Farooq, M., & Sazonov, E. (2021). Chapter 20 - Detection and characterization
of food intake by wearable sensors. In E. Sazonov (Ed.), Wearable Sensors (Second
Edition) (pp. 541-574). Oxford: Academic Press.

180
Ford, C. L., Griffith, D. M., Bruce, M. A., & Gilbert, K. L. (2019). Racism: Science & Tools for
the Public Health Professional: APHA Press.
Forrest, C. B., Ravens-Sieberer, U., Devine, J., Becker, B. D., Teneralli, R. E., Moon, J., . . .
Bevans, K. B. (2018). Development and Evaluation of the PROMIS® Pediatric Positive
Affect Item Bank, Child-Report and Parent-Proxy Editions. Journal of Happiness
Studies, 19(3), 699-718. doi:10.1007/s10902-016-9843-9
Forrestal, S. G. (2011). Energy intake misreporting among children and adolescents: a literature
review. Maternal & Child Nutrition, 7(2), 112-127. doi:10.1111/j.1740-
8709.2010.00270.x
Foucault Welles, B., Vashevko, A., Bennett, N., & Contractor, N. (2014). Dynamic Models of
Communication in an Online Friendship Network. Communication Methods and
Measures, 8(4), 223-243. doi:10.1080/19312458.2014.967843
Fowler, L. A., Grammer, A. C., Staiano, A. E., Fitzsimmons-Craft, E. E., Chen, L., Yaeger, L.
H., & Wilfley, D. E. (2021). Harnessing technological solutions for childhood obesity
prevention and treatment: a systematic review and meta-analysis of current applications.
International Journal of Obesity. doi:10.1038/s41366-021-00765-x
Framson, C., Kristal, A. R., Schenk, J. M., Littman, A. J., Zeliadt, S., & Benitez, D. (2009).
Development and Validation of the Mindful Eating Questionnaire. Journal of the
American Dietetic Association, 109(8), 1439-1444.
doi:https://doi.org/10.1016/j.jada.2009.05.006
Frederick, C. B., Snellman, K., & Putnam, R. D. (2014). Increasing socioeconomic disparities in
adolescent obesity. Proceedings of the National Academy of Sciences, 111(4), 1338.
doi:10.1073/pnas.1321355110

181
French, S. A., Epstein, L. H., Jeffery, R. W., Blundell, J. E., & Wardle, J. (2012). Eating
behavior dimensions. Associations with energy intake and body weight. A review.
Appetite, 59(2), 541-549. doi:https://doi.org/10.1016/j.appet.2012.07.001
Gallagher, R., Appenzeller, T., Normile, D., Service, R. F., & et al. (1999). Beyond
reductionism. Science, 284(5411), 79-109.
Ganley, R. M. (1989). Emotion and eating in obesity: A review of the literature. International
Journal of Eating Disorders, 8(3), 343-361.
Gemming, L., Doherty, A., Utter, J., Shields, E., & Ni Mhurchu, C. (2015a). The use of a
wearable camera to capture and categorise the environmental and social context of self-
identified eating episodes. Appetite, 92, 118-125. doi:10.1016/j.appet.2015.05.019
Gemming, L., & Ni Mhurchu, C. (2016). Dietary under-reporting: what foods and which meals
are typically under-reported? European Journal of Clinical Nutrition, 70(5), 640-641.
doi:10.1038/ejcn.2015.204
Gemming, L., Utter, J., & Ni Mhurchu, C. (2015b). Image-assisted dietary assessment: A
systematic review of the evidence. J Acad Nutr Diet, 115(1), 64-77.
doi:10.1016/j.jand.2014.09.015
Gilbert, D., & Waltz, J. (2010). Mindfulness and Health Behaviors. Mindfulness, 1(4), 227-234.
doi:10.1007/s12671-010-0032-3
Goldschmidt, A. B., Crosby, R. D., Cao, L., Pearson, C. M., Utzinger, L. M., Pacanowski, C. R.,
. . . Peterson, C. B. (2017). Contextual factors associated with eating in the absence of
hunger among adults with obesity. Eating Behaviors, 26, 33-39.
doi:https://doi.org/10.1016/j.eatbeh.2017.01.005

182
Goldschmidt, A. B., Engel, S. G., Wonderlich, S. A., Crosby, R. D., Peterson, C. B., Le Grange,
D., . . . Mitchell, J. E. (2012). Momentary Affect Surrounding Loss of Control and
Overeating in Obese Adults With and Without Binge Eating Disorder. Obesity, 20(6),
1206-1211. doi:10.1038/oby.2011.286
Goldschmidt, A. B., Smith, K. E., Crosby, R. D., Boyd, H. K., Dougherty, E., Engel, S. G., &
Haedt-Matt, A. (2018). Ecological momentary assessment of maladaptive eating in
children and adolescents with overweight or obesity. International Journal of Eating
Disorders, 51(6), 549-557. doi:10.1002/eat.22864
Goldschmidt, A. B., Wonderlich, S. A., Crosby, R. D., Cao, L., Engel, S. G., Lavender, J. M., . . .
Le Grange, D. (2014). Latent profile analysis of eating episodes in anorexia nervosa.
Journal of Psychiatric Research, 53, 193-199.
doi:https://doi.org/10.1016/j.jpsychires.2014.02.019
Goldstein, S. P., Hoover, A., Evans, E. W., & Thomas, J. G. (2021). Combining ecological
momentary assessment, wrist-based eating detection, and dietary assessment to
characterize dietary lapse: A multi-method study protocol. DIGITAL HEALTH, 7,
2055207620988212. doi:10.1177/2055207620988212
Gomes, D., & Sousa, I. (2019). Real-Time drink trigger detection in free-living conditions using
inertial sensors. Sensors, 19(9), 2145.
Goodman, L. A. (1974). Exploratory latent structure analysis using both identifiable and
unidentifiable models. Biometrika, 61(2), 215-231. doi:10.1093/biomet/61.2.215
Green, M. A., Evans, C. R., & Subramanian, S. V. (2017). Can intersectionality theory enrich
population health research? Social Science & Medicine, 178, 214-216.
doi:10.1016/j.socscimed.2017.02.029

183
Greeno, C. G., & Wing, R. R. (1994). Stress-induced eating. Psychological Bulletin, 115(3), 444-
464. doi:http://dx.doi.org/10.1037/0033-2909.115.3.444
Groesz, L. M., McCoy, S., Carl, J., Saslow, L., Stewart, J., Adler, N., . . . Epel, E. (2012). What
is eating you? Stress and the drive to eat. Appetite, 58(2), 717-721.
doi:https://doi.org/10.1016/j.appet.2011.11.028
Grolemund, G., & Wickham, H. (2011). Dates and Times Made Easy with lubridate. Journal of
Statistical Software; Vol 1, Issue 3 (2011).
Hales, C. M., Carroll, M. D., Fryar, C. D., & Ogden, C. L. (2017). Prevalence of Obesity Among
Adults and Youth: United States, 2015-2016. NCHS Data Brief(288), 1-8.
Hales, C. M., Fryar, C. D., Carroll, M. D., Freedman, D. S., & Ogden, C. L. (2018). Trends in
obesity and severe obesity prevalence in us youth and adults by sex and age, 2007-2008
to 2015-2016. JAMA, 319(16), 1723-1725. doi:10.1001/jama.2018.3060
Hammond, R. A. (2009). Complex systems modeling for obesity research. Preventing chronic
disease, 6(3), A97.
Han, J. C., Lawlor, D. A., & Kimm, S. Y. S. (2010). Childhood obesity. Lancet, 375(9727),
1737-1748. doi:https://doi.org/10.1016/S0140-6736(10)60171-7
Hassannejad, H., Matrella, G., Ciampolini, P., De Munari, I., Mordonini, M., & Cagnoni, S.
(2017). Automatic diet monitoring: a review of computer vision and wearable sensor-
based methods. International Journal of Food Sciences and Nutrition, 68(6), 656-670.
doi:10.1080/09637486.2017.1283683
He, J., Chen, G., Wu, S., Niu, R., & Fan, X. (2020). Patterns of negative emotional eating among
Chinese young adults: A latent class analysis. Appetite, 155, 104808.
doi:https://doi.org/10.1016/j.appet.2020.104808

184
Hebert, J. R., Clemow, L., Pbert, L., Ockene, I. S., & Ockene, J. K. (1995). Social Desirability
Bias in Dietary Self-Report May Compromise the Validity of Dietary Intake Measures.
International Journal of Epidemiology, 24(2), 389-398. doi:10.1093/ije/24.2.389
Herman, C. P. (2015). The social facilitation of eating. A review. Appetite, 86, 61-73.
doi:https://doi.org/10.1016/j.appet.2014.09.016
Herman, C. P., & Polivy, J. (2005). Normative influences on food intake. Physiol Behav, 86(5),
762-772. doi:10.1016/j.physbeh.2005.08.064
Herman, C. P., Polivy, J., Pliner, P., & Vartanian, L. R. (2019). Social Influences on Eating:
Springer.
Hermans, R. C. J., Lichtwarck-Aschoff, A., Bevelander, K. E., Herman, C. P., Larsen, J. K., &
Engels, R. C. M. E. (2012). Mimicry of Food Intake: The Dynamic Interplay between
Eating Companions. PLoS One, 7(2), e31027. doi:10.1371/journal.pone.0031027
Heron, K. E., Everhart, R. S., McHale, S. M., & Smyth, J. M. (2017). Using Mobile-Technology-
Based Ecological Momentary Assessment (EMA) Methods With Youth: A Systematic
Review and Recommendations. Journal of Pediatric Psychology, 42(10), 1087-1107.
doi:10.1093/jpepsy/jsx078
Heron, K. E., Miadich, S. A., Everhart, R. S., & Smyth, J. M. (2019). Ecological momentary
assessment and related intensive longitudinal designs in family and couples research. In
APA handbook of contemporary family psychology: Foundations, methods, and
contemporary issues across the lifespan, Vol. 1 (pp. 369-385). Washington, DC, US:
American Psychological Association.

185
Heron, K. E., & Smyth, J. M. (2010). Ecological momentary interventions: Incorporating mobile
technology into psychosocial and health behaviour treatments. British Journal of Health
Psychology, 15(1), 1-39. doi:https://doi.org/10.1348/135910709X466063
Heydarian, H., Adam, M., Burrows, T., Collins, C., & Rollo, E. M. (2019). Assessing Eating
Behaviour Using Upper Limb Mounted Motion Sensors: A Systematic Review.
Nutrients, 11(5), 1168. doi:10.3390/nu11051168
Higgs, S. (2015). Social norms and their influence on eating behaviours. Appetite, 86, 38-44.
doi:https://doi.org/10.1016/j.appet.2014.10.021
Higgs, S., & Thomas, J. (2016). Social influences on eating. Curr Opin Behav Sci, 9, 1-6.
doi:http://dx.doi.org/10.1016/j.cobeha.2015.10.005
Hou, F., Xu, S., Zhao, Y., Lu, Q., Zhang, S., Zu, P., . . . Tao, F. (2013). Effects of emotional
symptoms and life stress on eating behaviors among adolescents. Appetite, 68, 63-68.
doi:https://doi.org/10.1016/j.appet.2013.04.010
Huang, T. T., Drewnosksi, A., Kumanyika, S., & Glass, T. A. (2009). A systems-oriented
multilevel framework for addressing obesity in the 21st century. Preventing chronic
disease, 6(3), A82-A82.
Imtiaz, M. H., Ramos-Garcia, R. I., Wattal, S., Tiffany, S., & Sazonov, E. (2019). Wearable
Sensors for Monitoring of Cigarette Smoking in Free-Living: A Systematic Review.
Sensors, 19(21). doi:10.3390/s19214678
IPCC. (2019). Climate Change and Land: an IPCC special report on climate change,
desertification, land degradation, sustainable land management, food security, and
greenhouse gas fluxes in terrestrial ecosystems. Retrieved from www.ipcc.ch/srccl

186
Isasi, C. R., Parrinello, C. M., Jung, M. M., Carnethon, M. R., Birnbaum-Weitzman, O.,
Espinoza, R. A., . . . Gallo, L. C. (2015). Psychosocial stress is associated with obesity
and diet quality in Hispanic/Latino adults. Annals of Epidemiology, 25(2), 84-89.
doi:https://doi.org/10.1016/j.annepidem.2014.11.002
Jannasch, F., Kröger, J., & Schulze, M. B. (2017). Dietary Patterns and Type 2 Diabetes: A
Systematic Literature Review and Meta-Analysis of Prospective Studies. The Journal of
Nutrition, 147(6), 1174-1182. doi:10.3945/jn.116.242552
Johnson, J. A., & Johnson, A. M. (2015). Urban-Rural Differences in Childhood and Adolescent
Obesity in the United States: A Systematic Review and Meta-Analysis. Childhood
Obesity, 11(3), 233-241. doi:10.1089/chi.2014.0085
Kamarck, T. W., Shiffman, S., & Wethington, E. (2011). Measuring psychosocial stress using
ecological momentary assessment methods. In The handbook of stress science: Biology,
psychology, and health. (pp. 597-617). New York, NY, US: Springer Publishing
Company.
Keng, S.-L., & Tong, E. M. W. (2016). Riding the tide of emotions with mindfulness:
Mindfulness, affect dynamics, and the mediating role of coping. Emotion, 16(5), 706-
718. doi:10.1037/emo0000165
Khare, A., & Inman, J. J. (2006). Habitual Behavior in American Eating Patterns: The Role of
Meal Occasions. J Consum Res, 32(4), 567-575. doi:10.1086/500487
Klurfeld, D. M., Hekler, E. B., Nebeker, C., Patrick, K., & Khoo, C. S. H. (2018). Technology
Innovations in Dietary Intake and Physical Activity Assessment: Challenges and
Recommendations for Future Directions. American Journal of Preventive Medicine,
55(4), e117-e122. doi:https://doi.org/10.1016/j.amepre.2018.06.013

187
Kral, T. V. E., & Rauh, E. M. (2010). Eating behaviors of children in the context of their family
environment. Physiology & Behavior, 100(5), 567-573.
doi:https://doi.org/10.1016/j.physbeh.2010.04.031
Krueger, P. M., & Reither, E. N. (2015). Mind the Gap: Race/Ethnic and Socioeconomic
Disparities in Obesity. Current Diabetes Reports, 15(11), 95. doi:10.1007/s11892-015-
0666-6
Kuppens, P., Sheeber, L. B., Yap, M. B. H., Whittle, S., Simmons, J. G., & Allen, N. B. (2012).
Emotional inertia prospectively predicts the onset of depressive disorder in adolescence.
Emotion, 12(2), 283-289. doi:10.1037/a0025046
L'Hommedieu, M., L'Hommedieu, J., Begay, C., Schenone, A., Dimitropoulou, L., Margolin, G.,
. . . Narayanan, S. (2019). Lessons Learned: Recommendations For Implementing a
Longitudinal Study Using Wearable and Environmental Sensors in a Health Care
Organization. JMIR mHealth and uHealth, 7(12), e13305-e13305. doi:10.2196/13305
Laitinen, J., Ek, E., & Sovio, U. (2002). Stress-Related Eating and Drinking Behavior and Body
Mass Index and Predictors of This Behavior. Preventive Medicine, 34(1), 29-39.
doi:https://doi.org/10.1006/pmed.2001.0948
Lamers, F., Swendsen, J., Cui, L., Husky, M., Johns, J., Zipunnikov, V., & Merikangas, K. R.
(2018). Mood reactivity and affective dynamics in mood and anxiety disorders. Journal
of Abnormal Psychology, 127(7), 659-669. doi:10.1037/abn0000378
Lansigan, R. K., Emond, J. A., & Gilbert-Diamond, D. (2015). Understanding eating in the
absence of hunger among young children: A systematic review of existing studies.
Appetite, 85, 36-47. doi:https://doi.org/10.1016/j.appet.2014.10.032

188
Lapalme, J., Haines-Saah, R., & Frohlich, K. L. (2019). More than a buzzword: how
intersectionality can advance social inequalities in health research. Critical Public
Health, 1-7. doi:10.1080/09581596.2019.1584271
Lauby-Secretan, B., Scoccianti, C., Loomis, D., Grosse, Y., Bianchini, F., & Straif, K. (2016).
Body Fatness and Cancer — Viewpoint of the IARC Working Group. New England
Journal of Medicine, 375(8), 794-798. doi:10.1056/NEJMsr1606602
Laurent, J., Catanzaro, S. J., Joiner, T. E., Rudolph, K. D., Potter, K. I., Lambert, S., . . .
Gathright, T. (1999). A measure of positive and negative affect for children: Scale
development and preliminary validation. Psychological Assessment, 11(3), 326-338.
doi:Doi 10.1037/1040-3590.11.3.326
LaVeist, T., Pollack, K., Thorpe, R., Fesahazion, R., & Gaskin, D. (2011). Place, Not Race:
Disparities Dissipate In Southwest Baltimore When Blacks And Whites Live Under
Similar Conditions. Health Affairs, 30(10), 1880-1887. doi:10.1377/hlthaff.2011.0640
Lavender, J. M., Gratz, K. L., & Tull, M. T. (2011). Exploring the Relationship between Facets
of Mindfulness and Eating Pathology in Women. Cognitive Behaviour Therapy, 40(3),
174-182. doi:10.1080/16506073.2011.555485
Lazarsfeld, P. F., & Henry, N. W. (1968). Latent structure analysis. Boston, MA: Houghton
Mifflin Co.
Lee, B. Y., Bartsch, S. M., Mui, Y., Haidari, L. A., Spiker, M. L., & Gittelsohn, J. (2017). A
systems approach to obesity. Nutr Rev, 75(suppl 1), 94-106. doi:10.1093/nutrit/nuw049
Leech, R. M., Worsley, A., Timperio, A., & McNaughton, S. A. (2017). Temporal eating
patterns: a latent class analysis approach. International Journal of Behavioral Nutrition
and Physical Activity, 14(1), 3. doi:10.1186/s12966-016-0459-6

189
Leenders, R. T. A. J., Contractor, N. S., & DeChurch, L. A. (2016). Once upon a time:
Understanding team processes as relational event networks. Organizational Psychology
Review, 6(1), 92-115. doi:10.1177/2041386615578312
Lemmens, S. G., Rutters, F., Born, J. M., & Westerterp-Plantenga, M. S. (2011). Stress augments
food "wanting" and energy intake in visceral overweight subjects in the absence of
hunger. Physiology & Behavior, 103(2), 157-163.
doi:10.1016/j.physbeh.2011.01.009
Lenardson, J. D., Hansen, A. Y., & Hartley, D. (2015). Rural and Remote Food Environments
and Obesity. Current Obesity Reports, 4(1), 46-53. doi:10.1007/s13679-014-0136-5
Lerner, J., Tranmer, M., Mowbray, J., & Hancean, M.-G. (2019). REM beyond dyads: relational
hyperevent models for multi-actor interaction networks. arXiv preprint
arXiv:1912.07403.
Liao, Y., Schembre, S. M., O'Connor, S. G., Belcher, B. R., Maher, J. P., Dzubur, E., & Dunton,
G. F. (2018). An Electronic Ecological Momentary Assessment Study to Examine the
Consumption of High-Fat/High-Sugar Foods, Fruits/Vegetables, and Affective States
Among Women. Journal of Nutrition Education and Behavior, 50(6), 626-631.
doi:https://doi.org/10.1016/j.jneb.2018.02.003
Liao, Y., Skelton, K., Dunton, G., & Bruening, M. (2016). A Systematic Review of Methods and
Procedures Used in Ecological Momentary Assessments of Diet and Physical Activity
Research in Youth: An Adapted STROBE Checklist for Reporting EMA Studies
(CREMAS). J Med Internet Res, 18(6), e151. doi:10.2196/jmir.4954
Link, B. G., & Phelan, J. (1995). Social Conditions As Fundamental Causes of Disease. Journal
of Health and Social Behavior, 80-94. doi:10.2307/2626958

190
Liu, J. L., Han, B., & Cohen, D. A. (2015). Associations between eating occasions and places of
consumption among adults. Appetite, 87, 199-204.
doi:https://doi.org/10.1016/j.appet.2014.12.217
Livingstone, M. B., Robson, P. J., & Wallace, J. M. (2004). Issues in dietary intake assessment
of children and adolescents. Br J Nutr, 92 Suppl 2, S213-222.
Luke, D. A., & Stamatakis, K. A. (2012). Systems science methods in public health: dynamics,
networks, and agents. Annu Rev Public Health, 33, 357.
Lytle, L. A., Hearst, M. O., Fulkerson, J., Murray, D. M., Martinson, B., Klein, E., . . .
Samuelson, A. (2011). Examining the relationships between family meal practices,
family stressors, and the weight of youth in the family. Annals of Behavioral Medicine,
41(3), 353-362. doi:10.1007/s12160-010-9243-z
Ma, M., Alam, R., Bell, B., de la Haye, K., Spruijt-Metz, D., Lach, J., & Stankovic, J. (2017).
M^2G: A Monitor of Monitoring Systems with Ground Truth Validation Features for
Research-Oriented Residential Applications. IEEE 14th International Conference on
Mobile Ad Hoc and Sensor Systems (MASS), 10-18. doi:10.1109/MASS.2017.107
Magarey, A., Watson, J., Golley, R. K., Burrows, T., Sutherland, R., McNaughton, S. A., . . .
Collins, C. (2011). Assessing dietary intake in children and adolescents: Considerations
and recommendations for obesity research. Int J Pediatr Obes, 6(1), 2-11.
doi:10.3109/17477161003728469
Maher, J. P., Rebar, A. L., & Dunton, G. F. (2018). Ecological Momentary Assessment Is a
Feasible and Valid Methodological Tool to Measure Older Adults’ Physical Activity and
Sedentary Behavior. Frontiers in Psychology, 9, 1485.

191
Malik, V. S., Schulze, M. B., & Hu, F. B. (2006). Intake of sugar-sweetened beverages and
weight gain: a systematic review. The American Journal of Clinical Nutrition, 84(2),
274-288. doi:10.1093/ajcn/84.2.274
Marcum, C. S., & Butts, C. T. (2015). Constructing and Modifying Sequence Statistics for
relevent Using informR in R. Journal of statistical software, 64(5), 1-36.
Marcum, C. S., Goldring, M. R., McBride, C. M., & Persky, S. (2018). Modeling Dynamic Food
Choice Processes to Understand Dietary Intervention Effects. Annals of Behavioral
Medicine, 52(3), 252-261.
Mason, T. B., Do, B., Wang, S., & Dunton, G. F. (2020). Ecological momentary assessment of
eating and dietary intake behaviors in children and adolescents: A systematic review of
the literature. Appetite, 144, 104465. doi:https://doi.org/10.1016/j.appet.2019.104465
Mason, T. B., Lavender, J. M., Wonderlich, S. A., Crosby, R. D., Engel, S. G., Mitchell, J. E., . .
. Peterson, C. B. (2018). Examining a momentary mediation model of appearance-related
stress, anxiety, and eating disorder behaviors in adult anorexia nervosa. Eating and
Weight Disorders - Studies on Anorexia, Bulimia and Obesity, 23(5), 637-644.
doi:10.1007/s40519-017-0404-y
Mason, T. B., O'Connor, S. G., Schembre, S. M., Huh, J., Chu, D., & Dunton, G. F. (2019).
Momentary affect, stress coping, and food intake in mother–child dyads. Health
Psychology, 38(3), 238-247. doi:10.1037/hea0000714
Maugeri, A., & Barchitta, M. (2019). A Systematic Review of Ecological Momentary
Assessment of Diet: Implications and Perspectives for Nutritional Epidemiology.
Nutrients, 11(11). doi:10.3390/nu11112696

192
McClung, H. L., Ptomey, L. T., Shook, R. P., Aggarwal, A., Gorczyca, A. M., Sazonov, E. S., . .
. Das, S. K. (2018). Dietary Intake and Physical Activity Assessment: Current Tools,
Techniques, and Technologies for Use in Adult Populations. American Journal of
Preventive Medicine, 55(4), e93-e104. doi:https://doi.org/10.1016/j.amepre.2018.06.011
Mead, E., Brown, T., Rees, K., Azevedo, L. B., Whittaker, V., Jones, D., . . . et al. (2017). Diet,
physical activity and behavioural interventions for the treatment of overweight or obese
children from the age of 6 to 11 years. Cochrane Database of Systematic Reviews(6).
doi:10.1002/14651858.CD012651
Mondol, A., Bell, B. M., Ma, M., Alam, R., Emi, I. A., Preum, S. M., . . . Stankovic, J. A.
(2020). MFED: A System for Monitoring Family Eating Dynamics. arXiv preprint
arXiv:2007.05831.
Mulder, J., & Hoff, P. D. (2021). A Latent Variable Model for Relational Events with Multiple
Receivers. arXiv preprint arXiv:2101.05135.
Mundnich, K., Booth, B. M., L’Hommedieu, M., Feng, T., Girault, B., L’Hommedieu, J., . . .
Narayanan, S. (2020). TILES-2018, a longitudinal physiologic and behavioral data set of
hospital workers. Scientific Data, 7(1), 354. doi:10.1038/s41597-020-00655-3
Munsch, S., Hasenboehler, K., Michael, T., Meyer, A. H., Roth, B., Biedert, E., & Margraf, J.
(2007). Restrained eating in overweight children: Does eating style run in families?
International journal of pediatric obesity, 2(2), 97-103.
Muthén, L. K., & Muthén, B. O. (1998-2017). Mplus User’s Guide. Los Angeles, CA: Muthén &
Muthén.

193
Nahum-Shani, I., Hekler, E. B., & Spruijt-Metz, D. (2015). Building health behavior models to
guide the development of just-in-time adaptive interventions: A pragmatic framework.
Health Psychology, 34(Suppl), 1209-1219. doi:10.1037/hea0000306
Nahum-Shani, I., Smith, S. N., Spring, B. J., Collins, L. M., Witkiewitz, K., Tewari, A., &
Murphy, S. A. (2017). Just-in-Time Adaptive Interventions (JITAIs) in Mobile Health:
Key Components and Design Principles for Ongoing Health Behavior Support. Annals of
Behavioral Medicine, 52(6), 446-462. doi:10.1007/s12160-016-9830-8
Nam, S., Whittemore, R., Vlahov, D., & Dunton, G. (2019). Ecological Momentary Assessment
of Diabetes Self-Management: A Systematic Review of Methods and Procedures. Paper
presented at the Diabetes.
Navarathna, P., Bequette, B. W., & Cameron, F. (2018, 27-29 June 2018). Wearable Device
Based Activity Recognition and Prediction for Improved Feedforward Control. Paper
presented at the 2018 Annual American Control Conference (ACC).
Neuhouser, M. L. (2018). The importance of healthy dietary patterns in chronic disease
prevention. Nutrition Research. doi:https://doi.org/10.1016/j.nutres.2018.06.002
Neumark-Sztainer, D., Hannan, P. J., Story, M., Croll, J., & Perry, C. (2003). Family meal
patterns: associations with sociodemographic characteristics and improved dietary intake
among adolescents. Journal of the American Dietetic Association, 103(3), 317-322.
Neumark-Sztainer, D., Wall, M., Fulkerson, J. A., & Larson, N. (2013). Changes in the
Frequency of Family Meals From 1999 to 2010 in the Homes of Adolescents: Trends by
Sociodemographic Characteristics. Journal of Adolescent Health, 52(2), 201-206.
doi:https://doi.org/10.1016/j.jadohealth.2012.06.004

194
Neumark-Sztainer, D., Wall, M., Story, M., & Fulkerson, J. A. (2004). Are family meal patterns
associated with disordered eating behaviors among adolescents? Journal of Adolescent
Health, 35(5), 350-359. doi:https://doi.org/10.1016/j.jadohealth.2004.01.004
Ng, M., Fleming, T., Robinson, M., Thomson, B., Graetz, N., Margono, C., . . . Gakidou, E.
(2014). Global, regional, and national prevalence of overweight and obesity in children
and adults during 1980–2013: a systematic analysis for the Global Burden of Disease
Study 2013. The Lancet, 384(9945), 766-781. doi:https://doi.org/10.1016/S0140-
6736(14)60460-8
Nisa, C. F., Bélanger, J. J., Schumpe, B. M., & Faller, D. G. (2019). Meta-analysis of
randomised controlled trials testing behavioural interventions to promote household
action on climate change. Nature Communications, 10(1), 4545. doi:10.1038/s41467-
019-12457-2
Nishida, C., Uauy, R., Kumanyika, S., & Shetty, P. (2004). The Joint WHO/FAO Expert
Consultation on diet, nutrition and the prevention of chronic diseases: process, product
and policy implications. Public Health Nutrition, 7(1a), 245-250.
doi:10.1079/PHN2003592
Novotny, J. A., Rumpler, W. V., Riddick, H., Hebert, J. R., Rhodes, D., Judd, J. T., . . . Briefel,
R. (2003). Personality characteristics as predictors of underreporting of energy intake on
24-hour dietary recall interviews. Journal of the American Dietetic Association, 103(9),
1146-1151. doi:https://doi.org/10.1016/S0002-8223(03)00975-1
Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the Number of Classes in
Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study.

195
Structural Equation Modeling: A Multidisciplinary Journal, 14(4), 535-569.
doi:10.1080/10705510701575396
O'Connor, S. G., Maher, J. P., Belcher, B. R., Leventhal, A. M., Margolin, G., Shonkoff, E. T., &
Dunton, G. F. (2017). Associations of maternal stress with children's weight-related
behaviours: a systematic literature review. Obesity Reviews, 18(5), 514-525.
doi:10.1111/obr.12522
O’Connor, S. G., Habre, R., Bastain, T. M., Toledo-Corral, C. M., Gilliland, F. D., Eckel, S. P., .
. . Dunton, G. F. (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(1), 253.
doi:10.1186/s12889-019-6583-x
O’Reilly, G. A., & Spruijt-Metz, D. (2013). Current mHealth Technologies for Physical Activity
Assessment and Promotion. American Journal of Preventive Medicine, 45(4), 501-507.
doi:https://doi.org/10.1016/j.amepre.2013.05.012
Ogden, C. L., Carroll, M. D., Curtin, L. R., McDowell, M. A., Tabak, C. J., & Flegal, K. M.
(2006). Prevalence of Overweight and Obesity in the United States, 1999-2004. JAMA,
295(13), 1549-1555. doi:10.1001/jama.295.13.1549
Ogden, C. L., Carroll, M. D., Lawman, H. G., & et al. (2016). Trends in obesity prevalence
among children and adolescents in the united states, 1988-1994 through 2013-2014.
JAMA, 315(21), 2292-2299. doi:10.1001/jama.2016.6361
Ogden, C. L., Fakhouri, T. H., Carroll, M. D., Hales, C. M., Fryar, C. D., Li, X., & Freedman, D.
S. (2017). Prevalence of Obesity Among Adults, by Household Income and Education -

196
United States, 2011-2014. MMWR. Morbidity and mortality weekly report, 66(50), 1369-
1373. doi:10.15585/mmwr.mm6650a1
Oh, A., Erinosho, T., Dunton, G., M Perna, F., & Berrigan, D. (2014). Cross-sectional
examination of physical and social contexts of episodes of eating and drinking in a
national sample of US adults. Public Health Nutrition, 17(12), 2721-2729.
doi:10.1017/S1368980013003315
Ohkuma, T., Hirakawa, Y., Nakamura, U., Kiyohara, Y., Kitazono, T., & Ninomiya, T. (2015).
Association between eating rate and obesity: a systematic review and meta-analysis. Int J
Obes, 39, 1589–1596. doi:https://doi.org/10.1038/ijo.2015.96
Oude Luttikhuis, H., Baur, L., Jansen, H., Shrewsbury, V. A., O'Malley, C., Stolk, R. P., &
Summerbell, C. D. (2009). Interventions for treating obesity in children. Cochrane
Database Syst Rev(1), CD001872. doi:10.1002/14651858.CD001872.pub2 [doi]
Pachucki, M. A., Jacques, P. F., & Christakis, N. A. (2011). Social Network Concordance in
Food Choice Among Spouses, Friends, and Siblings. American Journal of Public Health,
101(11), 2170-2177. doi:10.2105/AJPH.2011.300282
Park, J., & Yu, H.-T. (2015). The Impact of Ignoring the Level of Nesting Structure in
Nonparametric Multilevel Latent Class Models. Educational and Psychological
Measurement, 76(5), 824-847. doi:10.1177/0013164415618240
Parks, E. P., Kumanyika, S., Moore, R. H., Stettler, N., Wrotniak, B. H., & Kazak, A. (2012).
Influence of Stress in Parents on Child Obesity and Related Behaviors. Pediatrics,
130(5), e1096. doi:10.1542/peds.2012-0895
Pate, R. R., O'Neill, J. R., Liese, A. D., Janz, K. F., Granberg, E. M., Colabianchi, N., . . .
Taverno Ross, S. E. (2013). Factors associated with development of excessive fatness in

197
children and adolescents: a review of prospective studies. Obesity Reviews, 14(8), 645-
658. doi:10.1111/obr.12035
Patison, K. P., Quintane, E., Swain, D. L., Robins, G., & Pattison, P. (2015). Time is of the
essence: an application of a relational event model for animal social networks.
Behavioral Ecology and Sociobiology, 69(5), 841-855. doi:10.1007/s00265-015-1883-3
Patrick, H., & Nicklas, T. A. (2005). A Review of Family and Social Determinants of Children’s
Eating Patterns and Diet Quality. Journal of the American College of Nutrition, 24(2),
83-92. doi:10.1080/07315724.2005.10719448
Pearson, N., Atkin, A. J., Biddle, S. J. H., Gorely, T., & Edwardson, C. (2010). Parenting styles,
family structure and adolescent dietary behaviour. Public Health Nutrition, 13(8), 1245-
1253. doi:10.1017/S1368980009992217
Pentland, B. T., Pentland, A. P., & Calantone, R. J. (2017). Bracketing off the actors: Towards an
action-centric research agenda. Information and Organization, 27(3), 137-143.
doi:https://doi.org/10.1016/j.infoandorg.2017.06.001
Peters, A., Kubera, B., Hubold, C., & Langemann, D. (2011). The Selfish Brain: Stress and
Eating Behavior. Front Neurosci, 5. doi:10.3389/fnins.2011.00074
Phelan, J. C., & Link, B. G. (2015). Is Racism a Fundamental Cause of Inequalities in Health?
Annual Review of Sociology, 41(1), 311-330. doi:10.1146/annurev-soc-073014-112305
Puhl, R., & Brownell, K. D. (2001). Bias, discrimination, and obesity. Obes Res, 9(12), 788-805.
doi:10.1038/oby.2001.108
Puhl, R. M., & Heuer, C. A. (2009). The Stigma of Obesity: A Review and Update. Obesity,
17(5), 941-964. doi:10.1038/oby.2008.636

198
Puhl, R. M., & Latner, J. D. (2007). Stigma, obesity, and the health of the nation's children.
Psychological Bulletin, 133(4), 557-580. doi:10.1037/0033-2909.133.4.557
Rajkumar, R., Lee, I., Sha, L., & Stankovic, J. (2010, 13-18 June 2010). Cyber-physical systems:
The next computing revolution. Paper presented at the Design Automation Conference.
Rast, F. M., & Labruyère, R. (2020). Systematic review on the application of wearable inertial
sensors to quantify everyday life motor activity in people with mobility impairments.
Journal of NeuroEngineering and Rehabilitation, 17(1), 148. doi:10.1186/s12984-020-
00779-y
Rebro, S. M., Patterson, R. E., Kristal, A. R., & Cheney, C. L. (1998). The Effect of Keeping
Food Records on Eating Patterns. Journal of the American Dietetic Association, 98(10),
1163-1165. doi:https://doi.org/10.1016/S0002-8223(98)00269-7
Reicks, M., Banna, J., Cluskey, M., Gunther, C., Hongu, N., Richards, R., . . . Wong, S. S.
(2015). Influence of Parenting Practices on Eating Behaviors of Early Adolescents during
Independent Eating Occasions: Implications for Obesity Prevention. Nutrients, 7(10),
8783-8801. doi:10.3390/nu7105431
Reilly, J. J., & Kelly, J. (2011). Long-term impact of overweight and obesity in childhood and
adolescence on morbidity and premature mortality in adulthood: systematic review. Int J
Obes (Lond), 35(7), 891-898. doi:10.1038/ijo.2010.222
Riley, W. T., Rivera, D. E., Atienza, A. A., Nilsen, W., Allison, S. M., & Mermelstein, R.
(2011). Health behavior models in the age of mobile interventions: are our theories up to
the task? Transl Behav Med, 1(1), 53-71. doi:10.1007/s13142-011-0021-7
Robinson, E., Thomas, J., Aveyard, P., & Higgs, S. (2014). What Everyone Else Is Eating: A
Systematic Review and Meta-Analysis of the Effect of Informational Eating Norms on

199
Eating Behavior. Journal of the Academy of Nutrition and Dietetics, 114(3), 414-429.
doi:https://doi.org/10.1016/j.jand.2013.11.009
Robson, S. M., McCullough, M. B., Rex, S., Munafò, M. R., & Taylor, G. (2020). Family Meal
Frequency, Diet, and Family Functioning: A Systematic Review With Meta-analyses.
Journal of Nutrition Education and Behavior, 52(5), 553-564.
doi:https://doi.org/10.1016/j.jneb.2019.12.012
Rosenheck, R. (2008). Fast food consumption and increased caloric intake: a systematic review
of a trajectory towards weight gain and obesity risk. Obesity Reviews, 9(6), 535-547.
doi:https://doi.org/10.1111/j.1467-789X.2008.00477.x
Ruanpeng, D., Thongprayoon, C., Cheungpasitporn, W., & Harindhanavudhi, T. (2017). Sugar
and artificially sweetened beverages linked to obesity: a systematic review and meta-
analysis. QJM: An International Journal of Medicine, 110(8), 513-520.
doi:10.1093/qjmed/hcx068
Saad, L. (2013). Most U.S. Families Still Routinely Dine Together at Home. Retrieved from
https://news.gallup.com/poll/166628/families-routinely-dine-together-home.aspx
Salvy, S. J., de la Haye, K., Bowker, J. C., & Hermans, R. C. (2012). Influence of peers and
friends on children's and adolescents' eating and activity behaviors. Physiol Behav,
106(3), 369-378. doi:10.1016/j.physbeh.2012.03.022
Saranummi, N., Spruijt-Metz, D., Intille, S., Korhonen, I., Nilsen, W. J., & Pavel, M. (2013).
Moving the science of behavioral change into the 21st century. IEEE PULSE, 4(5), 23-
24.

200
Satija, A., Yu, E., Willett, W. C., & Hu, F. B. (2015). Understanding nutritional epidemiology
and its role in policy. Advances in nutrition (Bethesda, Md.), 6(1), 5-18.
doi:10.3945/an.114.007492
Savage, J. S., Fisher, J. O., & Birch, L. L. (2007). Parental influence on eating behavior:
conception to adolescence. The Journal of law, medicine & ethics : a journal of the
American Society of Law, Medicine & Ethics, 35(1), 22-34. doi:10.1111/j.1748-
720X.2007.00111.x
Schafer, T. A. W. J. L. (2006). Models for intensive longitudinal data: Oxford University Press.
Schecter, A., & Contractor, N. (2017). Understanding and Assessing Collaborative Processes
Through Relational Events. In A. A. von Davier, M. Zhu, & P. C. Kyllonen (Eds.),
Innovative Assessment of Collaboration (pp. 223-231). Cham: Springer International
Publishing.
Schembre, S. M., Liao, Y., O'Connor, S. G., Hingle, M. D., Shen, S.-E., Hamoy, K. G., . . .
Boushey, C. J. (2018). Mobile Ecological Momentary Diet Assessment Methods for
Behavioral Research: Systematic Review. JMIR Mhealth Uhealth, 6(11), e11170.
doi:10.2196/11170
Schoeller, D. A. (1995). Limitations in the assessment of dietary energy intake by self-report.
Metabolism, 44, 18-22. doi:https://doi.org/10.1016/0026-0495(95)90204-X
Schoeller, D. A., Thomas, D., Archer, E., Heymsfield, S. B., Blair, S. N., Goran, M. I., . . .
Allison, D. B. (2013). Self-report-based estimates of energy intake offer an inadequate
basis for scientific conclusions. The American Journal of Clinical Nutrition, 97(6), 1413-
1415. doi:10.3945/ajcn.113.062125

201
Semba, R. D., de Pee, S., Kim, B., McKenzie, S., Nachman, K., & Bloem, M. W. (2020).
Adoption of the ‘planetary health diet’ has different impacts on countries’ greenhouse gas
emissions. Nature Food, 1(8), 481-484. doi:10.1038/s43016-020-0128-4
Sen, S., Subbaraju, V., Misra, A., Balan, R., & Lee, Y. (2020). Annapurna: An automated
smartwatch-based eating detection and food journaling system. Pervasive and Mobile
Computing, 68, 101259. doi:https://doi.org/10.1016/j.pmcj.2020.101259
Sharifi, M., Sequist, T. D., Rifas-Shiman, S. L., Melly, S. J., Duncan, D. T., Horan, C. M., . . .
Taveras, E. M. (2016). The role of neighborhood characteristics and the built
environment in understanding racial/ethnic disparities in childhood obesity. Preventive
Medicine, 91, 103-109. doi:https://doi.org/10.1016/j.ypmed.2016.07.009
Sharps, M., Higgs, S., Blissett, J., Nouwen, A., Chechlacz, M., Allen, H. A., & Robinson, E.
(2015). Examining evidence for behavioural mimicry of parental eating by adolescent
females. An observational study. Appetite, 89, 56-61.
doi:http://dx.doi.org/10.1016/j.appet.2015.01.015
Shiffman, S., Stone, A. A., & Hufford, M. R. (2008). Ecological momentary assessment. Annu
Rev Clin Psychol, 4, 1-32.
Shim, J.-S., Oh, K., & Kim, H. C. (2014). Dietary assessment methods in epidemiologic studies.
Epidemiology and health, 36, e2014009-e2014009. doi:10.4178/epih/e2014009
Shonkoff, E. T., Dunton, G. F., Chou, C.-P., Leventhal, A. M., Bluthenthal, R., & Pentz, M. A.
(2017). Direct and indirect effects of parent stress on child obesity risk and added sugar
intake in a sample of Southern California adolescents. Public Health Nutrition, 20(18),
3285-3294. doi:10.1017/S136898001700252X

202
Singh, A. S., Mulder, C., Twisk, J. W., van Mechelen, W., & Chinapaw, M. J. (2008). Tracking
of childhood overweight into adulthood: a systematic review of the literature. Obes Rev,
9(5), 474-488. doi:10.1111/j.1467-789X.2008.00475.x
Singh, G. K., Kogan, M. D., Van Dyck, P. C., & Siahpush, M. (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(9), 682-695. doi:https://doi.org/10.1016/j.annepidem.2008.05.001
Skinner, A. C., Ravanbakht, S. N., Skelton, J. A., Perrin, E. M., & Armstrong, S. C. (2018).
Prevalence of Obesity and Severe Obesity in US Children, 1999–2016. Pediatrics,
141(3), e20173459. doi:10.1542/peds.2017-3459
Smith, K. E., Mason, T. B., Crosby, R. D., Engel, S. G., Crow, S. J., Wonderlich, S. A., &
Peterson, C. B. (2018). State and trait positive and negative affectivity in relation to
restraint intention and binge eating among adults with obesity. Appetite, 120, 327-334.
doi:https://doi.org/10.1016/j.appet.2017.09.020
Smith, K. E., Mason, T. B., Schaefer, L. M., Anderson, L. M., Critchley, K., Crosby, R. D., . . .
Peterson, C. B. (2020). Dynamic Stress Responses and Real-Time Symptoms in Binge-
Eating Disorder. Annals of Behavioral Medicine. doi:10.1093/abm/kaaa061
Smyth, J. M., Wonderlich, S. A., Heron, K. E., Sliwinski, M. J., Crosby, R. D., Mitchell, J. E., &
Engel, S. G. (2007). Daily and momentary mood and stress are associated with binge
eating and vomiting in bulimia nervosa patients in the natural environment. Journal of
consulting and clinical psychology, 75(4), 629.
Sominsky, L., & Spencer, S. J. (2014). Eating behavior and stress: a pathway to obesity.
Frontiers in Psychology, 5(434). doi:10.3389/fpsyg.2014.00434

203
Spiegelman, B. M., & Flier, J. S. (2001). Obesity and the Regulation of Energy Balance. Cell,
104(4), 531-543. doi:10.1016/S0092-8674(01)00240-9
Spruijt-Metz, D. (2011). Etiology, Treatment, and Prevention of Obesity in Childhood and
Adolescence: A Decade in Review. Journal of Research on Adolescence, 21(1), 129-152.
doi:10.1111/j.1532-7795.2010.00719.x
Spruijt-Metz, D., de la Haye, K., Lach, J., & Stankovic, J. A. (2016). M2FED: Monitoring and
Modeling Family Eating Dynamics. Proceedings of the 14th ACM Conference on
Embedded Network Sensor Systems, 352-353. doi:10.1145/2994551.2996702
Spruijt-Metz, D., Hekler, E., Saranummi, N., Intille, S., Korhonen, I., Nilsen, W., . . . Pavel, M.
(2015). Building new computational models to support health behavior change and
maintenance: new opportunities in behavioral research. Transl Behav Med, 5(3), 335-346.
doi:10.1007/s13142-015-0324-1
Spruijt-Metz, D., & Nilsen, W. (2014). Dynamic Models of Behavior for Just-in-Time Adaptive
Interventions. Pervasive Computing, IEEE, 13(3), 13-17. doi:10.1109/MPRV.2014.46
Spruijt-Metz, D., Wen, C. K. F., Bell, B. M., Intille, S., Huang, J. S., & Baranowski, T. (2018).
Advances and Controversies in Diet and Physical Activity Measurement in Youth.
American Journal of Preventive Medicine, 55(4), e81-e91.
doi:https://doi.org/10.1016/j.amepre.2018.06.012
Stone, A. A., & Brownell, K. D. (1994). The stress-eating paradox: Multiple daily measurements
in adult males and females. Psychology & Health, 9(6), 425-436.
doi:10.1080/08870449408407469

204
Story, M., Kaphingst, K. M., Robinson-O'Brien, R., & Glanz, K. (2008). Creating Healthy Food
and Eating Environments: Policy and Environmental Approaches. Annual Review of
Public Health, 29(1), 253-272. doi:10.1146/annurev.publhealth.29.020907.090926
Subar, A. F., Kirkpatrick, S. I., Mittl, B., Zimmerman, T. P., Thompson, F. E., Bingley, C., . . .
Potischman, N. (2012). The Automated Self-Administered 24-hour dietary recall
(ASA24): a resource for researchers, clinicians, and educators from the National Cancer
Institute. J Acad Nutr Diet, 112(8), 1134-1137. doi:S2212-2672(12)00589-8 [pii]
10.1016/j.jand.2012.04.016
Suggs, L. S., Della Bella, S., Rangelov, N., & Marques-Vidal, P. (2018). Is it better at home with
my family? The effects of people and place on children's eating behavior. Appetite, 121,
111-118. doi:https://doi.org/10.1016/j.appet.2017.11.002
Suls, J., Green, P., & Hillis, S. (1998). Emotional Reactivity to Everyday Problems, Affective
Inertia, and Neuroticism. Personality and Social Psychology Bulletin, 24(2), 127-136.
doi:10.1177/0146167298242002
Sung-Chan, P., Sung, Y. W., Zhao, X., & Brownson, R. C. (2013). Family-based models for
childhood-obesity intervention: a systematic review of randomized controlled trials.
Obesity Reviews, 14(4), 265-278. doi:10.1111/obr.12000
Swinburn, B. A., Kraak, V. I., Allender, S., Atkins, V. J., Baker, P. I., Bogard, J. R., . . . Dietz,
W. H. (2019). The Global Syndemic of Obesity, Undernutrition, and Climate Change:
The Lancet Commission report. The Lancet, 393(10173), 791-846. doi:10.1016/S0140-
6736(18)32822-8
Takemoto, M., Manini, T. M., Rosenberg, D. E., Lazar, A., Zlatar, Z. Z., Das, S. K., & Kerr, J.
(2018). Diet and Activity Assessments and Interventions Using Technology in Older

205
Adults. American Journal of Preventive Medicine, 55(4), e105-e115.
doi:https://doi.org/10.1016/j.amepre.2018.06.005
Tani, Y., Kondo, N., Takagi, D., Saito, M., Hikichi, H., Ojima, T., & Kondo, K. (2015).
Combined effects of eating alone and living alone on unhealthy dietary behaviors, obesity
and underweight in older Japanese adults: Results of the JAGES. Appetite, 95, 1-8.
doi:https://doi.org/10.1016/j.appet.2015.06.005
Tanofsky-Kraff, M., Ranzenhofer, L. M., Yanovski, S. Z., Schvey, N. A., Faith, M., Gustafson,
J., & Yanovski, J. A. (2008). Psychometric properties of a new questionnaire to assess
eating in the absence of hunger in children and adolescents. Appetite, 51(1), 148-155.
doi:10.1016/j.appet.2008.01.001
Terry, P. C., Lane, A. M., Lane, H. J., & Keohane, L. (1999). Development and validation of a
mood measure for adolescents. Journal of Sports Sciences, 17(11), 861-872.
The GBD 2015 Obesity Collaborators. (2017). Health Effects of Overweight and Obesity in 195
Countries over 25 Years. New England Journal of Medicine, 377(1), 13-27.
doi:10.1056/NEJMoa1614362
Theim, K. R., Sinton, M. M., Stein, R. I., Saelens, B. E., Thekkedam, S. C., Welch, R. R., . . .
Wilfley, D. E. (2012). Preadolescents’ and Parents’ Dietary Coping Efficacy During
Behavioral Family-Based Weight Control Treatment. Journal of Youth and Adolescence,
41(1), 86-97. doi:10.1007/s10964-011-9728-5
Thomaz, E., Essa, I., & Abowd, G. D. (2015). A practical approach for recognizing eating
moments with wrist-mounted inertial sensing. Paper presented at the Proceedings of the
2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing,
Osaka, Japan. https://doi.org/10.1145/2750858.2807545

206
Thompson, E. L., Vamos, C. A., & Daley, E. M. (2017). Physical activity during pregnancy and
the role of theory in promoting positive behavior change: A systematic review. Journal of
Sport and Health Science, 6(2), 198-206. doi:https://doi.org/10.1016/j.jshs.2015.08.001
Thompson, F. E., Kirkpatrick, S. I., Subar, A. F., Reedy, J., Schap, T. E., Wilson, M. M., &
Krebs-Smith, S. M. (2015). The National Cancer Institute’s Dietary Assessment Primer:
A Resource for Diet Research. Journal of the Academy of Nutrition and Dietetics,
115(12), 1986-1995. doi:https://doi.org/10.1016/j.jand.2015.08.016
Thompson, F. E., & Subar, A. F. (2017). Chapter 1 - Dietary Assessment Methodology. In A. M.
Coulston, C. J. Boushey, M. G. Ferruzzi, & L. M. Delahanty (Eds.), Nutrition in the
Prevention and Treatment of Disease (Fourth Edition) (pp. 5-48): Academic Press.
Thompson, F. E., Subar, A. F., Loria, C. M., Reedy, J. L., & Baranowski, T. (2010). Need for
Technological Innovation in Dietary Assessment. Journal of the American Dietetic
Association, 110(1), 48-51. doi:10.1016/j.jada.2009.10.008
Timmins, K. A., Green, M. A., Radley, D., Morris, M. A., & Pearce, J. (2018). How has big data
contributed to obesity research? A review of the literature. International Journal of
Obesity, 42(12), 1951-1962. doi:10.1038/s41366-018-0153-7
Torres, S. J., & Nowson, C. A. (2007). Relationship between stress, eating behavior, and obesity.
Nutrition, 23(11), 887-894. doi:https://doi.org/10.1016/j.nut.2007.08.008
Tourlouki, E., Matalas, A.-L., & Panagiotakos, D. B. (2009). Dietary habits and cardiovascular
disease risk in middle-aged and elderly populations: a review of evidence. Clinical
interventions in aging, 4, 319-330.
Tranmer, M., Marcum, C. S., Morton, F. B., Croft, D. P., & de Kort, S. R. (2015). Using the
relational event model (REM) to investigate the temporal dynamics of animal social

207
networks. Animal Behaviour, 101, 99-105.
doi:https://doi.org/10.1016/j.anbehav.2014.12.005
Trickett, E. J., & Beehler, S. (2013). The Ecology of Multilevel Interventions to Reduce Social
Inequalities in Health. American Behavioral Scientist, 57(8), 1227-1246.
doi:10.1177/0002764213487342
Turner, T., Spruijt-Metz, D., Wen, C. K. F., & Hingle, M. D. (2015). Prevention and treatment of
pediatric obesity using mobile and wireless technologies: a systematic review. Pediatric
Obesity, 10(6), 403-409. doi:https://doi.org/10.1111/ijpo.12002
Vaizman, Y., Ellis, K., & Lanckriet, G. (2017). Recognizing Detailed Human Context in the
Wild from Smartphones and Smartwatches. IEEE Pervasive Computing, 16(4), 62-74.
doi:10.1109/MPRV.2017.3971131
Vandenbroeck, P., Goossens, L., & Clemens, M. (2007). Tackling Obesities: Future Choices —
Obesity System Atlas. Retrieved from
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_
data/file/295153/07-1177-obesity-system-atlas.pdf
Vansteelandt, K., & Verbeke, G. (2016). A Mixed Model to Disentangle Variance and Serial
Autocorrelation in Affective Instability Using Ecological Momentary Assessment Data.
Multivariate Behavioral Research, 51(4), 446-465. doi:10.1080/00273171.2016.1159177
Vartanian, L. R., Spanos, S., Herman, C. P., & Polivy, J. (2015). Modeling of food intake: a
meta-analytic review. Social Influence, 10(3), 119-136.
doi:10.1080/15534510.2015.1008037
Verhoeven Aukje, A. C., Adriaanse Marieke, A., Evers, C., & de Ridder Denise, T. D. (2012).
The power of habits: Unhealthy snacking behaviour is primarily predicted by habit

208
strength. British Journal of Health Psychology, 17(4), 758-770. doi:10.1111/j.2044-
8287.2012.02070.x
Videon, T. M., & Manning, C. K. (2003). Influences on adolescent eating patterns: the
importance of family meals. Journal of Adolescent Health, 32(5), 365-373.
doi:https://doi.org/10.1016/S1054-139X(02)00711-5
Volkow, N. D., Wang, G.-J., & Baler, R. D. (2011). Reward, dopamine and the control of food
intake: implications for obesity. Trends in cognitive sciences, 15(1), 37-46.
Vollmer, R. L., & Mobley, A. R. (2013). Parenting styles, feeding styles, and their influence on
child obesogenic behaviors and body weight. A review. Appetite, 71, 232-241.
doi:https://doi.org/10.1016/j.appet.2013.08.015
Vu, T., Lin, F., Alshurafa, N., & Xu, W. (2017). Wearable Food Intake Monitoring
Technologies: A Comprehensive Review. Computers, 6(1), 4.
doi:10.3390/computers6010004
Warren, J. M., Smith, N., & Ashwell, M. (2017). A structured literature review on the role of
mindfulness, mindful eating and intuitive eating in changing eating behaviours:
effectiveness and associated potential mechanisms. Nutrition Research Reviews, 30(2),
272-283. doi:10.1017/S0954422417000154
Waters, E., de Silva-Sanigorski, A., Hall, B. J., Brown, T., Campbell, K. J., Gao, Y., . . .
Summerbell, C. D. (2011). Interventions for preventing obesity in children. Retrieved
from
http://www.mrw.interscience.wiley.com/cochrane/clsysrev/articles/CD001871/frame.htm
l

209
Watts, N., Amann, M., Arnell, N., Ayeb-Karlsson, S., Beagley, J., Belesova, K., . . . Costello, A.
(2021). The 2020 report of The Lancet Countdown on health and climate change:
responding to converging crises. The Lancet, 397(10269), 129-170.
doi:https://doi.org/10.1016/S0140-6736(20)32290-X
Wehling, H., & Lusher, J. (2017). People with a body mass index ⩾30 under-report their dietary
intake: A systematic review. Journal of Health Psychology, 24(14), 2042-2059.
doi:10.1177/1359105317714318
Weinsier, R. L., Hunter, G. R., Heini, A. F., Goran, M. I., & Sell, S. M. (1998). The etiology of
obesity: relative contribution of metabolic factors, diet, and physical activity. The
American Journal of Medicine, 105(2), 145-150. doi:https://doi.org/10.1016/S0002-
9343(98)00190-9
Wells, J. C. K., & Siervo, M. (2011). Obesity and energy balance: is the tail wagging the dog?
European Journal of Clinical Nutrition, 65(11), 1173-1189. doi:10.1038/ejcn.2011.132
Wen, C. K. F., Schneider, S., Stone, A. A., & Spruijt-Metz, D. (2017). Compliance With Mobile
Ecological Momentary Assessment Protocols in Children and Adolescents: A Systematic
Review and Meta-Analysis. J Med Internet Res, 19(4), e132. doi:10.2196/jmir.6641
Wen, X., An, Y., Li, W., Du, J., & Xu, W. (2020). How could physical activities and sleep
influence affect inertia and affect variability? Evidence based on ecological momentary
assessment. Current Psychology. doi:10.1007/s12144-020-00803-3
Westerterp, K. R., & Goris, A. H. C. (2002). Validity of the assessment of dietary intake:
problems of misreporting. Current Opinion in Clinical Nutrition & Metabolic Care, 5(5),
489-493.

210
Wilfley, D. E., Stein, R. I., Saelens, B. E., Mockus, D. S., Matt, G. E., Hayden-Wade, H. A., . . .
Epstein, L. H. (2007). Efficacy of maintenance treatment approaches for childhood
overweight: a randomized controlled trial. JAMA, 298(14), 1661-1673.
doi:10.1001/jama.298.14.1661
Wilfley, D. E., Vannucci, A., & White, E. K. (2010). Family-Based Behavioral Interventions. In
M. Freemark (Ed.), Pediatric Obesity: Etiology, Pathogenesis, and Treatment (pp. 281-
301). New York, NY: Springer New York.
Wilkinson, M. J., Manoogian, E. N. C., Zadourian, A., Lo, H., Fakhouri, S., Shoghi, A., . . .
Taub, P. R. (2020). Ten-Hour Time-Restricted Eating Reduces Weight, Blood Pressure,
and Atherogenic Lipids in Patients with Metabolic Syndrome. Cell Metabolism, 31(1),
92-104.e105. doi:https://doi.org/10.1016/j.cmet.2019.11.004
Willett, W. (1998). 24-Hour Dietary Recall and Food Record Methods. In Nutritional
Epidemiology: Oxford University Press.
Willett, W. (2013). Nutritional epidemiology. In Monographs in epidemiology and biostatistics v
40 (pp. 1 online resource). Retrieved from
http://dx.doi.org/10.1093/acprof:oso/9780199754038.001.0001
Willett, W., Rockström, J., Loken, B., Springmann, M., Lang, T., Vermeulen, S., . . . Murray, C.
J. L. (2019). Food in the Anthropocene: the EAT-Lancet Commission on healthy diets
from sustainable food systems. The Lancet, 393(10170), 447-492. doi:10.1016/S0140-
6736(18)31788-4
Williams, D. R., & Jackson, P. B. (2005). Social Sources Of Racial Disparities In Health. Health
Affairs, 24(2), 325-334. doi:10.1377/hlthaff.24.2.325

211
Williams-Kerver, G. A., Wonderlich, S. A., Crosby, R. D., Cao, L., Smith, K. E., Engel, S. G., . .
. Le Grange, D. (2020). Differences in Affective Dynamics Among Eating-Disorder
Diagnostic Groups. Clinical Psychological Science, 8(5), 857-871.
doi:10.1177/2167702620917196
Wood, W., & Neal, D. T. (2016). Healthy through habit: Interventions for initiating &
maintaining health behavior change. Behavioral Science & Policy, 2(1), 71-83.
Wood, W., & Rünger, D. (2016). Psychology of Habit. Annual Review of Psychology, 67(1),
289-314. doi:10.1146/annurev-psych-122414-033417
World Health Organization. (2011). Global status report on noncommunicable diseases.
Retrieved from https://www.who.int/nmh/publications/ncd_report_full_en.pdf
Yang, C.-H., Huh, J., Mason, T. B., Belcher, B. R., Kanning, M., & Dunton, G. F. (2020).
Mother-child dyadic influences of affect on everyday movement behaviors: evidence
from an ecological momentary assessment study. International Journal of Behavioral
Nutrition and Physical Activity, 17(1), 56. doi:10.1186/s12966-020-00951-6
Yang, J., Wang, J., Nakandala, S., Kumar, A., & Jankowska, M. M. (2019, 24-27 Sept. 2019).
Predicting Eating Events in Free Living Individuals. Paper presented at the 2019 15th
International Conference on eScience (eScience).
Yau, Y. H. C., & Potenza, M. N. (2013). Stress and eating behaviors. Minerva endocrinologica,
38(3), 255-267.
Ye, X., Chen, G., Gao, Y., Wang, H., & Cao, Y. (2016). Assisting Food Journaling with
Automatic Eating Detection. Proceedings of the 2016 CHI Conference Extended
Abstracts on Human Factors in Computing Systems, 3255-3262.
doi:10.1145/2851581.2892426

212
Yee, A. Z. H., Lwin, M. O., & Ho, S. S. (2017). The influence of parental practices on child
promotive and preventive food consumption behaviors: a systematic review and meta-
analysis. International Journal of Behavioral Nutrition and Physical Activity, 14(1), 47.
doi:10.1186/s12966-017-0501-3
Zhang, Q., & Wang, Y. (2004). Socioeconomic inequality of obesity in the United States: do
gender, age, and ethnicity matter? Social Science & Medicine, 58(6), 1171-1180.
doi:https://doi.org/10.1016/S0277-9536(03)00288-0
Zhang, X., Pérez-Stable, E. J., Bourne, P. E., Peprah, E., Duru, O. K., Breen, N., . . . Denny, J.
(2017). Big Data Science: Opportunities and Challenges to Address Minority Health and
Health Disparities in the 21st Century. Ethnicity & disease, 27(2), 95-106.
doi:10.18865/ed.27.2.95

Abstract (if available)

Abstract This dissertation utilized data from the Monitoring and Modeling Family Eating Dynamics (M2FED) study, which developed and deployed novel methods for in-home sensing that could accurately monitor and model family eating dynamics using intensive longitudinal data from Ecological Momentary Assessment (EMA) and wearable sensors. This dissertation investigated the influence of intra-personal and inter-personal factors, and the synergy of these factors, on both individual and family eating behavior in order to increase our understanding of how children and families influence one another’s eating behaviors and to ultimately inform future obesity prevention and intervention strategies. The broad aims of this dissertation were to: (1) investigate the validity and feasibility of the M2FED cyberphysical system to detect in-field family eating behavior

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Sleep health and variability in youth: a real-time data capture study to examine influences on daily dietary intake patterns and longitudinal weight trajectories

PDF

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

PDF

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

PDF

Influences of specific environmental domains on childhood obesity and related behaviors

PDF

Psychosocial and behavioral ractors associated with emotional eating in adolescents

PDF

Motivation and the meanings of health behavior as factors associated with eating behavior in Latino youth

PDF

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

PDF

Diffusion of a peer‐led suicide preventive intervention in secondary schools: strategies to increase effectiveness of peer‐led interventions

PDF

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

PDF

Investigating factors that influence peer relationships and obesity during middle childhood

PDF

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

PDF

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

PDF

Using a human factors engineering perspective to design and evaluate communication and information technology tools to support depression care and physical activity behavior change among low-inco...

PDF

Tobacco and marijuana surveillance using Twitter data

PDF

A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system

PDF

Forensic markers of physical elder abuse in medical and community contexts: implications for criminal justice interventions

PDF

Acculturation team-based clinical program: pilot program to address acculturative stress and mental health in the Latino community

PDF

Essays on improving human interactions with humans, algorithms, and technologies for better healthcare outcomes

Asset Metadata

Creator Bell, Brooke Marie (author)

Core Title System dynamics of in-home family eating behavior: insights from intensive longitudinal data using Ecological Momentary Assessment and wearable sensors

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Preventive Medicine (Health Behavior Research)

Publication Date 04/26/2021

Defense Date 03/18/2021

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

Tag affect,Children,context,diet assessment,eating behavior,Ecological Momentary Assessment (EMA),families,family system,in-field,latent class analysis,mHealth,mood,OAI-PMH Harvest,obesity,relational event model,social influence,social networks,system dynamics,wearable sensors

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor de la Haye, Kayla (committee chair), Spruijt-Metz, Donna (committee chair), Belcher, Britni (committee member), Valente, Thomas (committee member), Wu, Shinyi (committee member)

Creator Email brookebe@usc.edu,brookebellusc@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-453779

Unique identifier UC11668443

Identifier etd-BellBrooke-9548.pdf (filename),usctheses-c89-453779 (legacy record id)

Legacy Identifier etd-BellBrooke-9548.pdf

Dmrecord 453779

Document Type Dissertation

Rights Bell, Brooke Marie

Type texts

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

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

Repository Name University of Southern California Digital Library

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