Investigating the associations of affective variability and physical activity among young adults using ecological momentary assessment

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Copyright 2023 Bridgette Do
INVESTIGATING THE ASSOCIATIONS OF AFFECTIVE VARIABILITY AND PHYSICAL
ACTIVITY AMONG YOUNG ADULTS USING ECOLOGICAL MOMENTARY
ASSESSMENT
By
Bridgette Do
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
May 2023
ii

ACKNOWLEDGMENTS
I would like to first acknowledge my mentor and committee chair, Dr. Genevieve
Dunton. Thank you for your guidance throughout the years, for creating an incredibly supportive
and collaborative community, and for setting an example as an incredible scientist, mentor, and
leader. You inspired me to pursue a PhD, and this would not be possible without you! I would
also like to acknowledge and thank my wonderful committee members Drs. Tyler Mason, Britni
Belcher, Kimberly Miller, and Donald Hedeker. Thank you for your optimism, lending your
expertise, and for helping me grow as a researcher. I am immensely grateful for each of your
contributions to this work and for your mentorship!
I would like to thank my parents Mimi Dinh and Steve Do for their life-long hard work
and sacrifices, which allowed me to have the privilege of pursuing higher education. I wouldn’t
be where I am today without you both. Thank you to my older sister and role model, Erica
Chorbajian. Your ambition and work ethic constantly inspires me, and your empathy reminds me
to show myself compassion. Thank you for always supporting me and for believing in me every
single day.
I would like to thank my partner, Kurt Taillon, for his endless support over the years.
You have helped me in so many ways: helping with doctoral program applications, reading
drafts, listening to my thoughts and presentations, testing out study materials, and of course
putting a big smile on my face every day. You celebrated every win with me—no matter how
small—and cheered me up during the difficult times. You have been my biggest supporter
throughout this journey, and I feel incredibly grateful to have you by my side.
I am extremely thankful for the wonderful friends and colleagues in the REACH Lab,
both past and present. Our research would not be possible without this amazing team, and I have
iii

learned so much by working alongside you all. It has been an honor to be part of the REACH
Lab for nearly 8 years. I would also like to acknowledge my doctoral colleagues and fellow
cohort members for building a supportive community and for making this journey fun!
I would like to give a big thank you to my dear friends—both near and far—for their
constant love and kind words. I am tremendously appreciative for your compassion, patience,
and hype over the years. I am so grateful for your love and support!
Lastly, a special thank you to the two cutest supporters, Peanut and Murphy. Your
unconditional love and companionship made each day brighter. Thank you for filling my heart
with so much joy and always giving me a reason to smile!
This dissertation research was made possible through a Research Supplement to Promote
Diversity in the Health-Related Research from the National Institutes of Health/National Heart,
Lung, and Blood Institute (3U01HL146327-03S1). Data for this dissertation came from the
Temporal Influences on Movement and Exercise (TIME) Study, which was funded by the
National Institutes of Health/National Heart, Lung, and Blood Institute (U01HL146327; PIs:
Genevieve F. Dunton and Stephen Intille). I would also like to acknowledge and thank the TIME
Study participants for their contributions to science. This work was also supported by the
Achievement Rewards for College Scientists Foundation Los Angeles Founder Chapter.

iv

TABLE OF CONTENTS
Acknowledgements………………………………………………………………………... ii

List of Tables……………………………………………………………………………… vii

List of Figures……………………………………………………………………………... ix

Abstract……………………………………………………………………………………. x

Chapter 1: Introduction……………………………………………………………………. 1
Background and Significance……………………………………………………... 1
Physical Activity and Health Outcomes…………………………………... 1
Affect as a Putative Factor Contributing to Physical Activity…………..... 3
Affective Variability……………………………………………………..... 9
Real-Time Data Capture Methods: Ecological Momentary Assessment
(EMA)……………………………………………………………………... 13
Specific Aims……………………………………………………………………… 17

Chapter 2: Examining the associations between subject-level affective variability and
overall physical activity using ecological momentary assessment: Exploring differences
by trait self-control………………………………………………………………………... 19
Abstract……………………………………………………………………………. 19
Introduction………………………………………………………………………... 21
Methods…………………………………………………………………………………… 26
Study Design….….….….…………………………………………………. 26
Recruitment and Participants……………………………………………… 26
Study Procedures………………………………………………………….. 27
Measures………………………………………………………………....... 29
Statistical Analysis……………………………………………………........ 31
Results……………………………………………………………………………... 35
Data Availability………………………………………………………….. 35
Descriptive Statistics……………………………………………………… 37
Variability in positive-activated affect predicting physical activity………. 38
Variability in positive-deactivated affect predicting physical activity……. 40
Variability in negative-activated affect predicting physical activity……… 41
Variability in negative-deactivated affect predicting physical activity…… 43
Discussion………………………………………………………………………... 49
Conclusions………………………………………………………………………. 56

Chapter 3: Investigating the day-level associations between affective variability and
physical activity using ecological momentary assessment………………………………... 57
Abstract…………………………………………………………………………..... 57
Introduction………………………………………………………………………... 59
Methods…………………………………………………………………………… 62
Study Design….….….….…………………………………………………. 62
v

Recruitment and Participants……………………………………………… 63
Study Procedures………………………………………………………….. 64
Measures………………………………………………………………....... 65
Statistical Analysis……………………………………………………….... 67
Results…………………………………………………………………………....... 69
Data Availability…………………………………………………………... 69
Participant Characteristics………………………………………………… 71
Day-level associations of positive-activated affective variability and
physical activity…………………………………………………………… 72
Day-level associations of positive-deactivated affective variability and
physical activity…………………………………………………………… 73
Day-level associations of negative-activated affective variability and
physical activity…………………………………………………………… 74
Day-level associations of negative-deactivated affective variability and
physical activity…………………………………………………………… 75
Discussion…………………………………………………………………………. 80
Conclusions………………………………………………………………………... 85

Chapter 4: Assessing whether subject-level associations of momentary affect and
subsequent physical activity predict future physical activity levels…….………………… 86
Abstract……………………………………………………………………………. 86
Introduction……………………………………………………………………....... 88
Methods…………………………………………………………………………… 91
Study Design….….….….…………………………………………………. 91
Recruitment and Participants……………………………………………… 92
Study Procedures………………………………………………………….. 93
Measures………………………………………………………………....... 94
Statistical Analysis………………………………………………………… 97
Results……………………………………………………………………………... 99
Data Availability………………………………………………………....... 99
Descriptive Statistics……………………………………………………… 100
Associations between momentary positive-activated affect and
subsequent physical activity predicting future physical activity…….……. 101
Associations between momentary positive-deactivated affect and
subsequent physical activity predicting future physical activity….………. 102
Associations between momentary negative-activated affect and
subsequent physical activity predicting future physical activity………….. 104
Associations between momentary negative-deactivated affect and
subsequent physical activity predicting future physical activity………….. 105
Discussion…………………………………………………………………………. 111
Conclusions………………………………………………………………………... 117

Chapter 5: Discussion…………………………………………………………………....... 118
Summary of Findings……………………………………………………………... 118
Overall Strengths………………………………………………………………….. 121
Overall Limitations………………………………………………………...……… 122
vi

Implications………………………………………………………………………..
. 124
Future Research Directions………………………………………………………... 127
Conclusions………………………………………………………………………... 130

References…………………………………………………………………………………. 131

vii

LIST OF TABLES

Table 1. Participant Characteristics (N=236)…….……………………………………….. 37

Table 2. Results from first-stage mixed effects location scale modeling for positive-
activated affect…………………………………………………………………………….. 44

Table 3. Results from first-stage mixed effects location scale modeling for positive-
deactivated ………………………………………………………………………………... 45

Table 4. Results from first-stage mixed effects location scale modeling for negative-
activated affect…………………………………………………………………………….. 46

Table 5. Results from first-stage mixed effects location scale modeling for negative-
deactivated affect………………………………………………………………………….. 47

Table 6. Results from second-stage mixed-effects location scale modeling for positive-
activated and positive-deactivated affect …………………………………………………. 48

Table 7. Results from second-stage mixed-effects location scale modeling for negative-
activated and negative-deactivated affect…………………………………………………. 49

Table 8. Participant Characteristics (N=231)…………………………………………… 71

Table 9. Mixed-effects location scale models examining day-level associations of
within-subject variability in positive-activated affect with physical activity……………... 76

Table 10. Mixed-effects location scale models examining day-level associations of
within-subject variability in positive-deactivated affect with physical activity…………... 77

Table 11. Mixed-effects location scale models examining day-level associations of
within-subject variability in negative-activated affect with physical activity…………….. 78

Table 12. Mixed-effects location scale models examining day-level associations of
within-subject variability in negative-deactivated affect with physical activity………….. 79

Table 13. Participant Characteristics (N=174)……………………………………………. 100

Table 14. Results of Stage 1 mixed-effects model (positive-activated affect) and Stage 2
linear regression model……………………………………………………………………. 107

Table 15. Results of Stage 1 mixed-effects model (positive-deactivated affect) and Stage
2 linear regression model………………………………………………………………… 108

viii

Table 16. Results of Stage 1 mixed-effects model (negative-activated affect) and Stage 2
linear regression model……………………………………………………………………. 109

Table 17. Results of Stage 1 mixed-effects model (negative-deactivated affect) and Stage
2 linear regression model………………………………………………………………….. 110

ix

LIST OF FIGURES

Figure 1. Representation of a circumplex model in a two-dimensional space defined by
valence (pleasure-displeasure) and activation (arousal).…………………………………... 6

Figure 2. Basic conceptual model of the three dissertation studies……………………….. 16

x

ABSTRACT

This dissertation is comprised of three unique studies that leveraged real-time data
capture methods (i.e., ecological momentary assessment [EMA]) and novel statistical modeling
strategies to examine the influence of affective variability (i.e., fluctuations in affect intensity) on
physical activity among young adults. The overarching objective of this dissertation was to
increase our scientific understanding on the potential associations between affective variability
(i.e., positive-activated, positive-deactivated, negative-activated, negative-deactivated affect) and
physical activity among young adults by coupling smartphone-based EMA with smartwatch
technology. The specific aims of this dissertation were to (1) determine whether subject-level
affective variability is associated with overall levels of physical activity, (2) determine whether
trait self-control moderates the associations between subject-level variability and overall levels
of physical activity, (3) investigate the day-level associations between affective variability and
physical activity, and (4) examine whether the subject-level association between momentary
affect and subsequent physical activity (e.g., next 30 minutes) predicts future physical activity
levels. Findings suggest that (1) subject-level variability in negative-deactivated affect was
associated with greater overall levels of physical activity, while variability in positive-activated,
positive-deactivated, and negative-activated affect were not associated with overall levels of
physical activity, (2) trait self-control did not moderate the associations between subject-level
affective variability and overall physical activity, (3) greater day-level variability in positive-
activated affect was associated with greater physical activity on that same day compared to other
days, whereas greater day-level variability in negative-deactivated affect was associated with less
physical activity on that same day compared to other days, and (4) the strength of association
between momentary affect and subsequent physical activity (i.e., 30 minutes later) did not
xi

predict future daily physical activity levels one month later. Taken together, the findings
highlight how variability in different affective states may be differently associated with physical
activity depending on the affective valence, arousal/activation state, and the temporality of the
associations examined. In addition, the dissertation underscores the importance of assessing
affective variability above and beyond mean or momentary levels. Future research can expand
upon this work by elucidating mechanisms underlying the associations of affective variability
and physical activity, considering important mental health characteristics, and developing EMA
studies to further explore the temporal relationships at the day-level and longitudinally. Overall,
this dissertation provides a foundation for future studies to unique role that affective variability
may play in physical activity engagement both within- and between-subjects. Ultimately,
knowledge on affect dynamics and momentary affective processes in free-living situations can
help optimize future intervention strategies by targeting fluctuations in affect to help increase
physical activity among young adults.

1

CHAPTER 1: INTRODUCTION
BACKGROUND AND SIGNIFICANCE
Physical Activity and Health Outcomes
Prevalence and Trends of Physical Activity
Physical activity is any bodily movement produced by skeletal muscles that results in
energy expenditure (Caspersen et al., 1985). Physical activity refers to all types of movement
including those during leisure time, as part of work, or for transport to get to and from places.
Energy expenditure is expressed as metabolic equivalent of task (MET) and rates of energy
expenditure during physical activity are often described by intensities: light, moderate, or
vigorous. Light-intensity activity requires less than 3.0 METs, moderate-intensity activity
requires 3.0-6.0 METs, and vigorous-intensity activity requires 6.0 or more METs. The Physical
Activity Guidelines for Americans (2018 Physical Activity Guidelines Advisory Committee
Scientific Report, 2018) recommends that adults engage in two types of physical activity each
week for health benefits: aerobic activity and muscle strengthening. For adults, it is
recommended to do at least 150 minutes of moderate-intensity aerobic activity or 75 minutes of
vigorous-intensity aerobic activity a week. Bout lengths of moderate-to-vigorous physical
activity (MVPA) of any duration can be included in the total volume of physical activity and
bouts of any length contribute to health benefits associated with physical activity (Piercy et al.,
2018). It is also recommended for adults to do muscle-strengthening activities of moderate or
greater intensity that engage all major muscle groups on two or more days a week.
Only one in four adults in the U.S. met the physical activity guidelines for aerobic and
muscle-strengthening activities in 2016 (2018 Physical Activity Guidelines Advisory Committee
Scientific Report, 2018). About 54.2% of U.S. adults engaged in recommended levels of aerobic

2

activity and 27.6% of U.S. performed muscle-strengthening activities on two or more days a
week. Even fewer young adults—typically defined as ages 18-29 years (Arnett, 2015; Arnett et
al., 2014)—are obtaining sufficient levels of physical activity for health benefits: only 10% of
young adults met the national recommendations (Tucker et al., 2011). Given the substantial
health benefits of physical activity, high inactivity rates among young adults are a critical public
health concern.
Associations between Physical Activity and Health Outcomes
There is well-established evidence on the health benefits of engaging in regular physical
activity. Regular physical activity is associated with a marked reduction in the risk for premature
mortality (Warburton & Bredin, 2017). The available evidence reveals a dose-response
relationship between physical activity and health—a 20-30% risk reduction for premature
mortality and chronic disease among those that meet or exceed physical activity guidelines
(Leitzmann et al., 2007; Nocon et al., 2008; Warburton & Bredin, 2017). According to the World
Health Organization, physical inactivity is the fourth leading risk factor for globally mortality,
resulting in an estimated 4-5 million deaths annually (WHO Guidelines on Physical Activity and
Sedentary Behaviour, 2020). Regular physical activity can also reduce the risk for more than 25
chronic conditions, including metabolic conditions, cardiovascular disease, hypertension, type 2
diabetes, osteoporosis, breast cancer, and colon cancer (Booth et al., 2012; Pedersen & Saltin,
2015; Piercy et al., 2018; Warburton & Bredin, 2017). Increasing regular physical activity
among young adults can help reduce the risk for a variety of chronic conditions and enhance the
short- and long-term health of adults.
Given the overwhelming evidence on the extensive health benefits of regular physical
activity, prioritizing regular physical activity is imperative for the long-term health of adults.

3

Furthermore, health behaviors shaped during these formative years persist throughout adulthood
(Daw et al., 2017; Telama et al., 2014); therefore, promoting physical is a critical public health
issue. The transition to adulthood is considered a crucial period of development when health
practices can be adopted or stopped, subsequently influencing behavioral and health trajectories.
Promoting regular physical activity among young adults, however interventions targeting
physical activity thus far typically focus on social, cognitive, or environmental factors.
Affect as a Putative Factor Contributing to Physical Activity

Previous literature has examined a wide range of potential social, cognitive, and
environmental determinants of physical activity. In example, studies have reported that family
involvement, social support, and marital status (social), motivation, self-efficacy, and attitudes
(cognitive), and proximity to activity facilities, greenness, walkability (environmental) are
associated with activity levels (Bauman et al., 2012; De Bourdeaudhuij et al., 2003; Kaczynski &
Henderson, 2007; Rhodes et al., 2012; Saelens et al., 2012; Yi et al., 2019). However, these
factors only explain only a small amount of the variance in behavior. Research needs to look
beyond these commonly studied constructs and into other potential determinants in order to
improve theoretical frameworks, research, and interventions. Recently, research has turned its
attention to examining affective determinants in hopes of elucidating the “intention-behavior”
gap which is evident in general knowledge of the benefits of physical activity yet simultaneous
low rates of physical activity (Rhodes & Gray, 2018; Sheeran et al., 2013). Research on affective
influences as a putative factor for physical activity has gained increased attention over the past
few decades (Ekkekakis & Brand, 2019; Stevens et al., 2020; Williams et al., 2018).

4

Conceptualizing Affect: Theoretical Models and Measurement
The terms “affect”, “emotion”, and “mood” have often been used synonymously in the
psychological and behavioral science literature and their definitions are not universally accepted
(Ekkekakis & Petruzzello, 2000). Among these affective constructs, affect—specifically core
affect—is the most general and is the broadest, most comprehensive form of affect (Ekkekakis,
2013). Core affect has been defined as the most elementary consciously accessible affective
feelings that are not directed at anything or a neurophysiological state accessible as a simple
primitive and non-reflective feeling (Russell, 2009; Russell & Barrett, 1999). Core affect is
always consciously available and is argued to be both non-cognitive and non-reflective, whereas
emotions (and sometimes moods) are the result of a specific cognitive appraisal (Russell, 2009).
Emotions and moods are reactions to the way one appraises relationships with the environment;
emotions are immediate responses to specific stimuli and moods can refer to the larger, pervasive
issues of one’s life (Lazarus, 1991). Affect is considered to be rapid and automatically occurring
feeling states, whereas emotions are short (i.e., seconds to minutes) and moods are longer lasting
(i.e., hours, days, or longer). Examples of core affect include: pleasure, displeasure, tension,
calmness, energy, and tiredness (Russell, 2009).
The conceptualization of affect has its basis in two common theoretical perspectives,
namely whether affective constructs represent distinct states or represent underlying dimensions
(Ekkekakis, 2013). The first theoretical model posits that affective states should be
conceptualized and modeled as distinct entities. This “distinct-states” approach considers each
state as unique and distinct from all others and should be considered independently from each
other. In example, affective constructs can be assessed by discrete states such as anxiety, anger,
or calmness. While assessing discrete states lends to more specificity, it is difficult to capture the

5

global domain of affect through this method. Alternatively, the second theoretical model argues
that affective constructs exhibit systematic interrelationships that can be modeled by a set of
underlying dimensions (Ekkekakis, 2013; Ekkekakis & Petruzzello, 2002). The “dimensional”
approach identifies elemental dimensions that account for the similarities and differences among
affective states, and affective states are thought to be positioned along dimensions. The main
advantage attributed to the dimensional approach is the breadth of scope and parsimony,
resulting in adequate representations of the entire affective space. Therefore, dimensional models
are considered well-suited for studying affect from a global perspective (Ekkekakis &
Petruzzello, 2002).
Researchers utilizing the dimensional approach for studying affective constructs
identified a two-dimensional structure consisting of two dimensions: valence (pleasure-
displeasure) and activation (arousal) (Russell, 1978). The widely accepted circumplex model
proposed by Russell is a variation of the two-dimensional structure (Russell, 1980). Based on
this model, there two orthogonal (i.e., unrelated) and bipolar dimensions: affective valence and
perceived activation. Different affective states are considered combinations of varying degrees of
these two dimensions, such that affective states are located around the perimeter of the circle
(Ekkekakis & Petruzzello, 2000; Russell, 1980; Russell et al., 1989) (Figure 1). Affective states
positioned close together on the circle represent similar combinations of valence and activation
(e.g., happy and glad). On the other hand, states diametrically across from one another (e.g.,
happy and sad) differ maximally in terms of one or the other dimension (i.e., valence or
activation). Russell’s circumplex model has become the standard for two-dimensional
approaches for measuring affect.

6

Figure 1. Representation of a circumplex model in a two-dimensional space defined by
valence (pleasure-displeasure; x-axis) and activation (arousal; y-axis).
In addition to acknowledging the varying conceptualizations of affect, it is important to
distinguish between affect occurring outside versus within the context of a target behavior.
Affect can be further categorized into incidental affect or integral affect. Incidental affect
describes how one feels throughout the day and is unrelated to any specific target behavior, but
nonetheless may influence the behavior or be influenced by the behavior. Integral affect refers to
one’s affective response to a target behavior (e.g., how they feel while performing the behavior)
or the immediate consequence of the behavior (Williams et al., 2019; Williams & Evans, 2014).

7

This proposal will focus on the former—incidental affect—and its associations with physical
activity in naturalistic settings because affect measurements are not tied directly to physical
activity. Although incidental affect is not a direct result of a target behavior, incidental affect
may influence the behavior and/or be influenced by a target behavior (Williams & Evans, 2014).
Associations between Affect and Subsequent Physical Activity
Several theoretical frameworks or conceptual models have emphasized the effects of
incidental affect on behaviors. The main underlying premise of these models suggest that
behaviors are a function of how an individual feels leading up to the behavior, but is independent
of any anticipation of the behavior (Lazarus, 1993; Williams et al., 2019). According to affect
regulation theories, the effect of incidental affect is moderated by how one expects to feel as a
result of the behavior, such that individuals will engage in behaviors that are expected to regulate
their current affective state (Hall et al., 2018; Sheeran et al., 2018; Tice et al., 2001).
For example, negative incidental affect may lead to engagement in maladaptive behaviors
if individuals anticipate that the behavior will alleviate their negative affective state (Baker et al.,
2004; Larsen, 1987; Stevens et al., 2020). Similar to negative reinforcement, behaviors are
reinforced when associated with the removal or avoidance of an aversive stimuli (e.g., negative
affect) (Bandura, 1989; McKee et al., 2003). In example, an individual may engage in physical
activity if they anticipate that physical activity will produce a positive affective response (e.g.,
“exercise makes you feel better”). On the other hand, an individual may engage in sedentary
behaviors (e.g., watching TV), if they expect it to alleviate feelings of negative affect, even when
it is not in line with their intentions or long-term goals (Tice et al., 2001).
Empirical evidence suggests that instantaneous levels of affect at any given moment (i.e.,
incidental affect) predict subsequent levels of physical activity within short time scales (e.g., next

8

15, 30, or 60 minutes). A systematic review (Liao et al., 2015) on the acute relationships between
affective states and physical activity in naturalistic settings concluded that positive affective
states were positively associated with physical activity over the next few hours (Carels et al.,
2007; Dunton et al., 2009; Schwerdtfeger et al., 2010). On the other hand, negative affective
states were considered not an antecedent for physical activity given the null or inconsistent
results (Dunton et al., 2014; Mata et al., 2012; Schwerdtfeger et al., 2010; Wichers et al., 2012).
In addition, feeling more energetic led to more physical activity, whereas feeling tired was
associated with less subsequent physical activity among children (Dunton et al., 2014). However,
there no significant associations between feeling states (i.e., energy and fatigue) were found in a
study among middle-aged to older adults (Dunton et al., 2009).
Several studies examining the acute relationships between affect and physical activity in
naturalistic settings have been published after the review, yielding similar results for positive
affect and providing additional support for negative affect and feeling states. Positive affect
predicted more objectively-measured and self-reported MVPA after work, whereas negative
predicted less MVPA (Niermann et al., 2016). A separate study revealed independent positive
within-subjects effects of valence and feeling energetic, and negative within-subject effects of
calmness on non-exercise activity (e.g., biking for transport, climbing stairs) (Reichert et al.,
2016). Among young adult sub-populations (i.e., university students), more positive affective
valence, more energetic arousal, and less calmness was associated with greater momentary
physical activity (Kanning & Schoebi, 2016). Given that the young adult population experiences
an array of daily hassles, life changes, and demands due to responsibilities and life transitions
related to education, interpersonal relationships, and occupations, additional research in free-
living environments is needed.

9

Affective Variability
Research in affect, emotion, and mood has also extended previous research beyond
characterizing affect at the subject-level or assessing instantaneous, momentary affect.
Understanding how affect fluctuates within subjects and throughout the day, and whether these
fluctuations are associated with health behaviors or outcomes, has gained increased attention. At
any given moment, individuals are a complex configuration of characteristics, with some
personal characteristics changing from moment to moment, day to day, or week to week
(Nesselroade & Ram, 2004). For instance, emotional experiences and regulation are temporal
processes that fluctuate over time and within individuals (Gross, 2013). Understanding the
patterns of variability within individuals may help us better understand human behavior because
we may be able to see if these patterns that occur cross moments, days, or weeks precede
particular behaviors or outcomes.
Theoretical perspectives suggest that higher levels of affective variability may deplete
one’s capacity to control maladaptive behaviors in the context of urges or cues (Baumeister et
al., 1998; Weiss et al., 2018). Consistent with this perspective, the strength model of self-control
posits that self-regulation—the capacity to override and alter responses as an attempt to control
unwanted urgers or behaviors—draws upon limited resources that can be temporarily depleted
resulting in a state of ego depletion (Baumeister et al., 1998; Baumeister & Vohs, 2007). As an
individual experiences increased internal demands (e.g., negative affect, stress) or external
demands (e.g., interpersonal conflicts), the ability to self-regulate diminishes. Regulating affect
and its fluctuations has been shown to deplete self-regulatory resources given that it requires an
individual to overcome innate tendencies to display feelings in response to environmental stimuli
(Baumeister et al., 1998; Hagger et al., 2010a). Empirical evidence on affect lability, which is

10

categorized by intense affective variability, and health behaviors has demonstrated associations
of greater affect lability with increased risk for cigarette smoking (Mermelstein et al., 2010),
alcohol use (Jahng et al., 2011; Simons et al., 2014), binge eating (Anestis et al., 2009), and
substance misuse (Shadur et al., 2015; Weiss et al., 2018).
Associations between Affective Variability and Physical Activity
The concept of ego depletion may also underlie associations of affective variability and
physical activity, such that affective variability may deplete one’s self-regulatory resources and
thus make it increasingly difficult to engage in goal-directed behaviors. Regulating one’s affect
multiple times throughout the day due to fluctuations may be a type of resource depletion. While
there is considerable literature on associations of momentary affect and physical activity,
whether the overall degree of variability in affect (i.e., whether one person has more stable affect
than another person) contributes to one’s levels of physical activity is lesser known.
There is preliminary evidence for the associations between affective variability and
activity. First, (Maher et al., 2019) reported that while intraindividual variability in positive
affect was not associated with physical activity or sedentary time, greater intraindividual
variability of feeling energetic was associated with less physical activity. Second, (Kerrigan et
al., 2020) found that less variability in negative affect both at the inter- and intra-individual level
predicted physical activity among adults with overweight or obesity such that lower variability in
negative affect predicted more physical activity on the subsequent day. Third, (Dunton et al.,
2014) concluded that mean hourly leisure-time MVPA among school-aged children was
associated with less intraindividual variability in both positive and negative affect.

11

Challenges and Limitations in Examining Affective Variability and Physical Activity
Although there is preliminary evidence for the associations of affective variability and
physical activity in naturalistic settings, there are gaps in the literature and additional research is
needed. Previous research on affective determinants and health behaviors has typically combined
valence (pleasure-displeasure) and arousal (activation) when conceptualizing affect by assessing
“positive affect” and “negative affect”. However, if the two fundamental dimensions of affect
suggested by the circumplex model (i.e., valence and arousal) are mixed, it is difficult to make
conclusions given that positive and negative affect refer to valence only (Ekkekakis, 2013;
Russell, 1980). Therefore, it is important to examine affect variability differences in relation to
activity by valence and arousal.
In addition, few studies have investigated the day-level or within-subject associations of
physical activity with affective variability. Additional research is needed at the within-subject
level to see if fluctuations in affect are associated with daily changes in activity. Knowledge of
day-level associations could inform just-in-time adaptive interventions (Nahum-Shani et al.,
2015) that target variability by teaching skills for reducing fluctuations in affect across the day or
by providing coping strategies. Evaluating affective variability can provide a more nuanced
understanding of the relationships between affect and activity that may otherwise be overlooked.
Affective variability may be a key predictor of activity independent of instantaneous affect at a
given single time point as it accounts for ebbs and flows in affect across the day. Thus,
examination of affect variability may explain some of the inconsistent evidence on acute
associations found in previous literature.

12

Statistical Modeling of Variability

In order to accurately model affective variability assessed in real-world settings,
appropriate methodologies and statistical programs are needed to collect and analyze data.
MixWILD (Mixed model analysis With Intensive Longitudinal Data), an open-source and user-
friendly program (e.g., point-and-click graphical user interface), has the ability to answer unique
research questions regarding time-varying variables (e.g., momentary affect) and health
outcomes or behaviors (e.g., physical activity) which otherwise have been limited by previous
statistical models and software (https://reach-lab.github.io/MixWildGUI/). The program can test
the effects of subject-level parameters (variance and slope) of time-varying variables in studies
utilizing intensive sampling methods such as EMA (Dzubur et al., 2020). MixWILD uses a two-
stage modeling approach by combining the estimation of a Stage 1 mixed-effects location scale
(MELS) model (Hedeker et al., 2008), including an estimation of the subject-specific random
effects, with a subsequent Stage 2 linear or binary/ordinal logistic regression where values from
each subject’s random effect are used as regressors. To date, there is limited statistical software
available to conduct two-stage modeling of time-varying outcomes (Stage 1) on higher-level
outcomes (Stage 2). In example, SAS PROC NLMIXED is thus far unable to test random
intercepts and slopes as predictors, moderators, or mediators of outcome variables.
MixWILD also has the ability to model intra-individual variability differently from
previously used methods in three unique ways. First, previous methods have commonly
calculated summary statistics of variability for each person such as subject-level standard
deviations, mean square of successive differences, or probability of acute change (Solhan et al.,
2009). However, by separately computing summary statistics for each subject, these methods
ignore the fact that subjects may vastly differ in terms of the number of observations that they

13

contribute to the analysis. This may particularly be in the case in intensive sampling methods
such as EMA where subjects are repeatedly measured across numerous occasions and have
prompting schedules often tailored to their sleep-wake schedule. By assuming that subjects
contribute the same number of observations to the analyses, each summary statistic is treated as
if it is equally precise in its estimation across subjects. In contrast, MixWILD recognizes that
subjects differ in number of observations contributed to the analyses. Secondly, the commonly
used summary statistics for modeling intra-individual variability have been used in subsequent
analyses as fixed quantities, which ignores the fact that these are estimates and therefore sources
of variation, leading to small standard errors and more false positive results. Alternatively, the
two-stage modeling approach in MixWILD utilizes the plausible values re-sampling approach
(Mislevy, 1991) to account for the variation in these estimates. Lastly, the Stage 1 model of
MixWILD can characterize a subject’s data in terms of means, variances, and slopes, and also
control for other covariates in the model; this allows the subject-level variance estimates to
adjust for mean levels and trends across time. Given the distinct ability for modeling variability,
MixWILD can be used to answer unique research questions regarding affective variability and
physical activity, which otherwise would not be possible or be limited with previously available
programs.
Real-Time Data Capture Methods: Ecological Momentary Assessment (EMA)

Ecological momentary assessment (EMA) is a real-time data capture methodology that
can help address some of these limitations (Stone et al., 2007). EMA involves repeatedly
assessing participants’ current states, behaviors, and contexts in real-time and in participants’
natural environments. EMA approaches offer three main benefits: (1) reduces recall biases, (2)
increases ecological validity, and (3) achieves temporal resolution by studying processes over

14

time. EMA methods aim to reduce biases related to recall by assessing current states and
behaviors, or recent ones, rather than asking participants for a recall over long periods. Unlike
traditional retrospective study designs, EMA methods can provide information at the momentary
level. Retrospective recall and autobiographical memory are subject to several limitations that
may influence participant reports; empirical research has shown that recall is subject to
systematic bias (Stone et al., 1999). When participants are asked to recall an experience, a
variety of heuristics are used to recreate the information. In example, events that are more
salient, recent, or unusual are more likely to be recalled than others (i.e., availability heuristic),
and people are more likely to retrieve negative information when they are in a negative mood
(i.e., state biases) (Stone et al., 2007). EMA methods also seek to maximize ecological validity
by collecting data in real-world environments as participants go about their lives. Measures, such
has momentary states or behaviors, can be assessed throughout the day. Compared to laboratory
environments—which may not accurately portray the context of participants’ typical contexts
and settings—gathering information on states and experiences in naturalistic settings can yield
data that is more generalizable to real-world settings. Furthermore, understanding how processes
unfold over time, as well as proximal antecedents and consequences of behaviors, can lead to
efficacious interventions (Dunton, 2017).
Notably, EMA data are well-suited to address research questions about within-person
variation in experiences or behaviors over time. EMA studies require participants to complete
multiple assessments over time—typically several times per day—providing a nuanced picture of
how states, behaviors, and experiences unfold over time or vary across time and contexts. In
addition, these repeated assessments lead the way to collecting and analyzing intensive
longitudinal data. By employing EMA methods, models of micro-temporal processes (i.e.,

15

processes that unfold over a short period of time) can be further examined (Dunton, 2018). Time-
varying explanatory factors, such as affect, change over a short period of time, and thus require
appropriate methodologies to collect and analyze data. Given that assessing variability requires
repeated assessments in naturalistic settings, EMA and the collection of intensive longitudinal
data is optimal. EMA has been advocated for the assessment of affective variability given that
single measures tend to assess average extremity of affect rather than the frequency or degree of
fluctuations (Ebner-Priemer et al., 2009; Larsen, 1987). Furthermore, the factors influencing
physical activity may be better understood on a shorter timescale (e.g., across minutes, hours, or
days) given that these behaviors are largely driven by temporal and situational cues that occur in
everyday life. EMA methods, combined with smartphone technologies and wearable sensors,
offer unique opportunities to examine the relationships of affective variability and physical
activity. The three dissertation studies (Figure 2) extend beyond prior research by using EMA
methods coupled with novel two-stage statistical modeling approaches to capture and model
within-day variability in affect to examine associations with physical activity among young
adults.

16

Figure 2. Basic conceptual model of the three dissertation studies

17

SPECIFIC AIMS
Study 1: Examining the associations between subject-level affective variability and overall
physical activity using ecological momentary assessment: Exploring differences by trait
self-control
Aim 1a: To determine whether subject-level affective variability is associated with
overall levels of physical activity.
Hypothesis 1a: Individuals who experience more variability in affect (i.e.,
positive-activated, positive-deactivated, negative-activated, and negative-
deactivated affect) will engage in less overall physical activity (i.e., inverse
associations).
Aim 1b: To determine whether trait self-control moderates the associations between
subject-level affective variability and overall levels of physical activity.
Hypothesis 1b: The association between affective variability and physical
activity would be stronger among individuals with lower self-control. In other
words, the strength of the inverse association would be stronger (i.e., more
negative) for individuals with lower self-control. Among individuals with higher
self-control, self-control may buffer the detrimental effects of affective variability
on physical activity.
Study 2: Investigating the day-level associations between affective variability and physical
activity using ecological momentary assessment
Aim 2: To investigate the day-level associations between affective variability (i.e.,
positive-activated, positive-deactivated, negative-activated, and negative-deactivated
affect) and physical activity.

18

Hypothesis 2: Greater day-level affective variability (i.e., within-subject
variance) will be associated with less physical activity on that same day compared
to usual. Specifically, individuals will experience less affective variability (i.e.,
positive-activated affect, positive-deactivated affect, negative-activated affect,
negative-deactivated affect) on days when they engage in more physical activity
compared to their average levels of activity.
Study 3: Assessing whether subject-level associations of momentary affect and subsequent
physical activity predict future physical activity levels
Aim 3: To examine whether the subject-level association between momentary affect (i.e.,
positive-activated, positive-deactivated, negative-activated, negative-deactivated) and
subsequent physical activity (e.g., next 30 minutes) predicts future physical activity
levels.
Hypothesis 3: Individuals with a stronger subject-level association between
momentary affect and physical activity levels (i.e., in the subsequent 30 minutes)
would be less active one month later.

19

Chapter 2: Examining the associations between subject-level affective variability and
overall physical activity using ecological momentary assessment: Exploring differences by
trait self-control

Abstract
Background: Theoretical models posit higher levels of affective variability (i.e., fluctuations in
affect intensity from moment to moment) may deplete an individual’s self-regulatory capacity to
control behaviors. Having to self-regulate affect multiple times throughout the day due to daily
fluctuations may be a type of self-regulatory resource depletion. Thus, affective variability may
deplete one’s self-regulatory processes and thus decrease the likelihood of engaging in goal-
directed behaviors like physical activity. Preliminary research suggests that subject-level
variability in affect may be associated with less overall physical activity. The current study used
novel methodology to assess and model subject-level variability in affect (i.e., positive-activated,
positive-deactivated, negative-activated, negative-deactivated), and subsequently examined the
association between subject-level variability with overall levels of physical activity. This study
also explored a potential important moderator, trait self-control, which may buffer the potential
detrimental effects of affective variability.
Methods: Young adults (N=236, M=23.60+3.18 years) provided six months of smartphone-
based EMA and smartwatch-based activity data. Every two weeks, participants completed a 4-
day EMA measurement burst (M=11.2+3.2 bursts per participant). Bursts consisted of hourly
randomly-prompted EMA surveys assessing momentary positive-activated (happy, energetic),
positive-deactivated (relaxed), negative-activated (tense, stressed), and negative-deactivated
(sad, fatigued) affect. Participants continuously wore a smartwatch to measure PA across the six
months. A two-stage analytic approach tested the study aims. In the first stage, mixed-effects

20

location scale modeling decomposed mean levels and variability in affect. In the second stage, a
linear regression was tested to investigated associations between subject-level variability in
affect and overall levels of physical activity. Trait self-control was added as a moderator in
separate second stage models. Models controlled for time of day, day of the week, burst number,
sex at birth, and age.
Results: There were 99,806 completed EMA surveys (M=422.91+151.18 per participant)
included in first stage analyses among 236 participants. Variability in positive-activated affect,
positive-deactivated affect, and negative-activated affect was not associated with overall levels
of physical activity. However, subject-level variability in negative-deactivated was associated
with more overall physical activity. Trait self-control did not moderate the associations between
affective variability and physical activity.
Conclusions: Results indicate that subject-level variability in negative-deactivated affect (i.e.,
feeling sad and fatigued) was associated with greater levels of physical activity, whereas
variability in positive-activated, positive-deactivated, and negative-activated affect did not
predict physical activity. Despite the general null findings for the self-control moderation effect,
the current study extends beyond previous research in affect and physical activity by considering
the way in which affect fluctuates above or below the mean. The use of novel statistical methods
to estimate variability and subsequently predict subject-level outcomes can be applied in future
studies to examine unique research questions in health behavior and public health.

21

Introduction
Regularly engaging in physical activity has many physical and mental health benefits,
such as reducing the risk for metabolic conditions, cardiovascular disease, type 2 diabetes,
anxiety, and depression (Booth et al., 2012; McDowell et al., 2019; Pedersen & Saltin, 2015;
Piercy et al., 2018; Reed & Buck, 2009; Warburton & Bredin, 2017). Evidence also indicates a
20-30% risk reduction in premature mortality among adults who meet or exceed physical activity
recommendations (Leitzmann et al., 2007; Nocon et al., 2008). Despite the well-documented
benefits of physical activity, only one in four adults in the United States (U.S.) meet the national
physical activity guidelines (2018 Physical Activity Guidelines Advisory Committee Scientific
Report, 2018). Increasing physical activity is a public health priority that requires researchers
and practitioners to identify efficacious strategies for promoting physical activity participation.
Whether an individual engages in physical activity, and how much they participate in, is
influenced by numerous factors spanning across individual, social, and environmental factors. At
the individual-level, empirical research has largely focused on associations between cognitive
factors (e.g., motivation, self-efficacy, attitudes) and physical activity (Bauman et al., 2012;
Rhodes et al., 2019; Sheeran et al., 2016). However, physical activity researchers have
increasingly investigated affective factors as putative determinants of physical activity to
elucidate who will, or will not, maintain physical activity and under what contexts (D. E. Conroy
& Berry, 2017; Ekkekakis & Brand, 2019; Stevens et al., 2020; Williams et al., 2018). The
Affect and Health Behavior Framework (AHBF) posits that incidental affect—how one feels
outside the context of the target behavior—is a key affective determinant of health behavior
(Williams et al., 2019; Williams & Evans, 2014). Overall retrospective levels of affect have been
shown to be associated with overall levels of physical activity (Reed & Buck, 2009). In addition,

22

elevated momentary levels of incidental positive affective states (e.g., feeling happy, energetic,
calm) are associated with greater subsequent physical activity (e.g., 30 min-rest of the day),
whereas there is inconsistent evidence for the associations between incidental negative affective
states (e.g., feeling sad, stressed, tense) and subsequent physical activity (Liao et al., 2015).
Despite the empirical evidence on affective determinants of physical activity, few studies
have considered the dynamic nature of affect such as the fluctuations in affect intensity from
moment-to-moment (Ebner-Priemer et al., 2009; Nesselroade & Ram, 2004). Individuals may
vary in their average levels of affect and in their overall degree of variability (i.e., whether one
person has more stable or consistent affect than another person) beyond mean levels in affect
(Davidson, 1998; Ebner-Priemer & Trull, 2012). Theoretical models posit higher levels of
affective variability may deplete an individual’s self-regulatory capacity to control behaviors.
The strength model of self-control indicates that self-regulation (i.e., the capacity to override and
alter responses in attempt to control unwanted urges or behaviors) draws on limited resources
which can become depleted across the day as individuals draw on these resources (Baumeister et
al., 1998; Baumeister & Vohs, 2007). When an individual experiences increased internal or
external demands, the ability to self-regulate diminishes and it can become increasingly difficult
to behave in line with one’s goals and values (Hagger et al., 2010b). Having to self-regulate
affect multiple times throughout the day due to daily fluctuations may be a type of self-
regulatory resource depletion. Thus, affective variability may deplete one’s self-regulatory
processes and therefore decrease the likelihood of engaging in goal-directed behaviors like
physical activity. Emerging evidence suggests variability in affect contributes to physical activity
above and beyond mean levels of affect. Data pooled from four ambulatory assessment studies
indicated greater variability in feeling energetic was associated with lower odds of meeting

23

physical activity guidelines, whereas variability in positive affect was not associated with
physical activity (Maher et al., 2019). In a separate study, findings showed that less variability in
negative affect predicted more physical activity on the following day (Kerrigan et al., 2020).
These studies provide preliminary evidence for the relationship between affective variability and
physical activity.
Given the theoretical importance of self-control in the association between affective
variability with physical activity, individual differences in self-control may be an important
moderator of the association between affective variability and physical activity. The strength of
the relationship between affective variability and physical activity may vary due to an
individual’s ability to adjust to fluctuations in affect. Specifically, the association between
affective variability and physical activity may be stronger among individuals with overall lower
self-regulatory capacity (i.e., self-control). Trait self-control has been found to be correlated with
individual differences in self-regulatory resources (Schmeichel & Zell, 2007; Tangney et al.,
2004). Individuals with lower self-control may have less self-regulatory resources and thus be
less likely to engage in goal-oriented behaviors (e.g., physical activity) when experiencing
fluctuations in affect. On the other hand, high levels of self-control may help buffer the negative
effects of affective variability on physical activity; individuals may be able to self-regulate and
engage in physical activity even when experiencing fluctuations in affect. To our best
knowledge, prior research has yet to examine whether self-control moderates the association
between affect and physical activity. If self-control acts as a buffer between affective variability
and physical activity, physical activity interventions may benefit by developing tailored affect-
based intervention strategies for individuals who are more vulnerable to depletion of self-

24

regulatory resources (e.g., self-regulation training, coping techniques, self-reflection activities)
(Stieger et al., 2021).
To better understand the micro-temporal processes associated with physical activity, real-
time data capture methodologies such as ecological momentary assessment (EMA) are optimal
for gathering data from individuals in their natural environments (Shiffman et al., 2008). In
contrast to laboratory settings, the data collection of states, experiences, and behaviors as they
occur across the day in naturalistic settings can increase ecological validity and help provide
information for intervention development that is more generalizable to real-world contexts and
settings (Stone et al., 2007). In addition, EMA on smartphones can collect repeated
measurements and thus help elucidate intraindividual differences and capture fluctuations in
affect. EMA has been advocated for the assessment of affective variability given that single
measures tend to assess average extremity of affect rather than the frequency or degree of
fluctuations (Ebner-Priemer et al., 2009; Larsen, 1987). Intensive longitudinal data collected
from EMA also allows for more flexible modeling that can address unique research questions
involving micro-temporal processes. Elucidating whether affective variability is associated with
physical activity, and whether self-control is a moderator, can aid in the development of
intervention strategies to promote physical activity.
Therefore, the overall aim of the current study was to examine whether affective
variability is associated with overall levels of physical activity, and whether self-control
moderates these associations. The first objective of the study was to determine whether subject-
level affective variability is associated with overall levels of physical activity (Aim 1a). It was
hypothesized that individuals who experience more variability in affect (i.e., positive-activated,
positive-deactivated, negative-activated, and negative-deactivated affect) will engage in less

25

overall physical activity (i.e., inverse associations) (Dunton et al., 2014; Kerrigan et al., 2020;
Maher et al., 2019). The second objective was to determine whether trait self-control moderates
the associations between affective variability and physical activity (Aim 1b). It was hypothesized
that the association between affective variability and physical activity would be stronger among
individuals with lower self-control (Baumeister & Vohs, 2007; Muraven & Baumeister, 2000).
In other words, the strength of the inverse association would be stronger (i.e., more negative) for
individuals with lower self-control. Among individuals with higher self-control, self-control may
buffer the detrimental effects of affective variability on physical activity, such that individuals
with higher trait self-control may have greater self-regulatory resources that allow them to cope
and adapt to fluctuations in affect.
The current study will build upon the preliminary research in several ways. First, Maher
and colleagues were unable to examine variability in negative affective states due to a limited
amount of intraindividual variability in negative affect (Maher et al., 2019). In addition, Maher
and colleagues included a sample of all ages (i.e., 7-83 years old) and Kerrigan and colleagues
included a sample of participants in a weight-loss treatment program. Additional research is
required among the adult population (Kerrigan et al., 2020; Maher et al., 2019). Prior research
has also focused on moderate-to-vigorous physical activity (MVPA) as the physical activity
outcome of interest; however, this focuses on the upper end of activity intensity and ignores
other intensities (i.e., light physical activity) and movement that individuals engage in. Lastly,
utilizing a longer study period (e.g., six months) and more frequent EMA sampling (e.g., hourly)
can yield more data to estimate variability and assess physical activity.

26

Methods
Study Design
The current study used intensive longitudinal data collected from the Temporal
Influences on Movement and Exercise (TIME) study (Wang et al., 2022). The TIME study
consisted of 4-day measurement bursts every two weeks for 12 consecutive months. During each
burst, participants completed signal-contingent (i.e., randomly prompted) EMA on their personal
Android smartphone. The study took place between March 2020 and August 2022. To address
the current study’s objectives and to maximize available data, the analyses utilized data collected
in the first six months of the study. This study was conducted in accordance with the Declaration
of Helsinki and all aspects of the study were approved by the Institutional Review Board at the
University of Southern California (HS-18-00605).
Recruitment and Participants
Young adults (ages 18-29 years) living in the U.S were recruited for the TIME Study on a
rolling basis. All recruitment and study procedures were conducted remotely due to the COVID-
19 pandemic (respiratory disease caused by the SARS-CoV-2 virus) and its related restrictions.
Participants were recruited through multiple strategies: (1) individuals enrolled in the Happiness
& Health Study (a University of Southern California prospective cohort study of ninth-grade
students that began in 2013) were sent emails to addresses on file (Leventhal et al., 2015); (2)
referrals from existing participants (e.g., word of mouth); (3) online/social media advertisements
(e.g., Twitter, Facebook, Instagram); (4) emails were sent to addresses on file from other
University of Southern California studies; and (5) emails were sent to potentially eligible
participants through ResearchMatch (a national health volunteer registry that was created by

27

several academic institutions and supported by the U.S. National Institutes of Health as part of
the Clinical Translational Science Award program).
Study inclusion criteria for the study were: (1) 18-29 years old living in the United States;
(2) intend within the next 12 months to engage in, or already engage in, recommended levels of
physical activity (≥150 min/week moderate or ≥75 min/week vigorous intensity);

(3) use an
Android-based smartphone as their only primary personal mobile device with no intention to
switch to a non-Android phone; (4) able to speak and read English; and (5) plan to reside in an
area with Wi-Fi connectivity during the study period. Study exclusion criteria were: (1) physical
or cognitive disabilities that prevent participation; (2) health issues that limit physical activity;
(3) diagnosed sleep disorders; (4) unable to wear a smartwatch or answer EMA surveys at home,
work, school, or other location where the participant spends a substantial amount of time (e.g.,
participant would not be able to answer prompts more than 20% of the time); (5) spends more
than 3 hours/day on a typical weekday or weekend day driving; (6) owning an Android phone
version 6.0 (or older) or if the app will not function on the phone due to other technical issues;
(7) currently owning and wearing a smartwatch; (8) have a pay-as-you-go data plan or data plan
with less than 2 GB of data; or (9) currently pregnant. Participant eligibility criteria were
established with the intent to include participants that regularly engage in or intend to engage in
physical activity to ensure activity data would be collected in naturalistic settings, consider the
health and safety of participants, and ensure enough data would be collected from the
smartphone and smartwatch throughout the study period.
Study Procedures
Interested individuals completed an online interest form on Research Electronic Data
Capture (REDCap) hosted at the University of Southern California (Harris et al., 2009, 2019) to

28

determine initial eligibility. Study staff then contacted potential participants to answer additional
eligibility questions and provide more information about the study. Eligible participants were
asked to attend a video conference orientation on Zoom with a study staff member to complete
informed consent and receive instructions for downloading the custom TIME study smartphone
application on their personal Android smartphone. Participants were also trained on how to use
the study app to complete EMA surveys. After the orientation session, participants completed an
online baseline questionnaire on REDCap. Participants were mailed a smartwatch, and then
rained on smartwatch set-up and use during a second video conference orientation. Participants
were compensated for their participation based upon compliance with study procedures and
could receive up to $1260 USD in total for the entire 12-month study. For each 4-week period,
participants could earn up to $100 USD. In addition, participants were able to keep their
smartwatch if they completed the 12-month study period.
EMA
During each 4-day EMA burst, participants completed EMA on the custom TIME app
(https://play.google.com/store/apps/details?id=mhealth.neu.edu.microT) developed for Android
smartphones and smartwatches. The app was paired on the participant’s smartphone and
smartwatch. EMA bursts were scheduled for a block of four consecutive days, that included at
least two weekdays and two weekend days (i.e., Thursday-Sunday or Saturday-Tuesday) with at
least seven days in between each burst. During each burst, EMA surveys were randomly
prompted (i.e., signal-contingent) via push notification once every hour during the participant’s
waking hours. Prompting was restricted to between the 10
th
and 50
th
minute of each hour (e.g.,
12:10 P.M. - 12:50 P.M.) to ensure that two prompts in consecutive hours did not occur to close
to one another. EMA surveys required 1-2 minutes to complete. If no response was provided, up

29

to two reminder prompts were sent in five-minute intervals; after this point, the EMA survey was
no longer accessible. Participants were instructed to ignore any prompts that occurred during
incompatible activities (e.g., driving, napping).
Smartwatch Accelerometry
Participants were loaned a Fossil Sport Gen 4 or Gen 5 smartwatch to wear on one wrist
of their choice consistently over the 12-month period, except for during one hour per day for
charging and when the watch would be exposed to water for an extended period (e.g., showering,
swimming). Participants were asked to not install health/fitness apps on their smartwatch
because they could interfere with the sensor data collection and drain the watch battery. The
smartwatch continuously recorded participants’ activity except for when the smartwatch was
turned off.
Measures
Affective States
Affective states were measured during each EMA survey. Select items were chosen to
measure affective states and represent the two fundamental dimensions of affect suggested by the
circumplex model (i.e., valence and arousal) (Russell, 1980). Participants were asked seven
questions in the following format: “Right now, how (affect term) do you feel?” Positive-activated
affect (happy, energetic), positive-deactivated affect (relaxed), negative-activated affect (tense,
stressed), and negative-deactivated affect (sad, fatigued) were assessed (Stevens et al., 2020).
The response options were on a unipolar scale: 1 = “Not at all”; 2 = “A little”; 3 = “Moderately”;
4 = “Quite a bit”; 5 = “Extremely”. Scores were averaged to create a composite score for each
construct, respectively.

30

Physical Activity
Physical activity was passively measured through the wrist-worn smartwatches. Tri-axial
raw acceleration along the X, Y, and Z axes was measured. Monitor Independent Movement
Summary (MIMS) units were computed offline using the raw accelerometry data to summarize
movement (John et al., 2019). MIMS-units allow for measuring the amount of movement,
independent of the type of sensor (e.g., smartwatch, smartphone), and is a standardized,
nonproprietary metric that can be applied across different accelerometers (Belcher et al., 2021).
MIMS for 1-second epochs were calculated to determine the total motion at varied lengths of
times longer than one second (e.g., minutes, hours). Smartwatch non-wear was assessed through
the sleep, wear, and non-wear (SWaN) algorithm, which classifies the raw accelerometry data
into sleep, wear, and non-wear classes for a 30-second window (Arguello et al., 2018). For the
current study, daily total MIMS-units/valid smartwatch wear time (hours) was calculated for all
available days of data during the six months. The outcome, overall levels of physical activity,
was operationalized as average daily total MIMs-units/valid smartwatch wear time.
Self-Control
Self-control was assessed during the online baseline survey using the 36-item Self-
Control Scale (Tangney et al., 2004). The scale measured trait-levels of self-control over
feelings, actions, and thoughts. Example items include “I am good at resisting temptation”, “I get
carried away by my feelings”, and “I am able to work effectively toward long-term goals”
(reverse coded). Responses were on a 5-point Likert scale ranging from 1 (Not at all like me) to
5 (Very much like me). Of the 36 items, 24 were reverse coded. Higher scores indicated higher
levels of self-control (Cronbach’s alpha = 0.90).

31

Participant Characteristics
Participant characteristics were collected through a baseline electronic questionnaire.
Participants self-reported the following characteristics: 1) age; 2) sex at birth (female, male); 3)
transgender identity (yes transgender male-to-female, yes transgender female-to-male, yes
transgender gender nonconforming, no, don’t know/not sure); 4) Hispanic, Latino/a, or Spanish
origin; 5) race (check all that apply: White, Black or African American, American Indian or
Alaska Native, Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, Other Asian,
Native Hawaiian, Guamanian or Chamorro, Samoan, Other Pacific Islander); 6) educational
attainment (never attended school or only attended kindergarten, grades 1 through 8
(elementary), grades 9 through 11 (some high school), grade 12 or GED (high school graduate),
college 1 year to 3 years (some college or technical school), college 4 years or more (college
graduate)); 7) work status (check all that apply: employed for wages, self-employed, out of work
for 1 year or more, out of work for less than 1 year, homemaker, student, retired, unable to
work); 8) marital status (married, divorced, widowed, separated, never married, member of
unmarried couple). Participants were able to skip/not answer questions if they preferred to.
Statistical Analyses
Descriptive statistics (e.g., frequencies, means) were calculated to describe the study
sample and variables of interest. The stage second outcome, overall physical activity (i.e.,
average daily MIMS-units/smartwatch valid wear time, was positively skewed and therefore log-
transformed. Prior to data analysis, EMA observations were excluded from the analytic data set
if the participant’s standard deviation for an affect variable during the entire 6-month period was
less than 0.2 to ensure there was at least a minimal degree of affect variability to allow variance
modeling. A two-stage modeling approach using mixed-effects location scale modeling in

32

MixWILD (Mixed models With Intensive Longitudinal Data; https://reach-
lab.github.io/MixWildGUI/) was used (Dzubur et al., 2020; Hedeker et al., 2008). In contrast to
other statistical methods calculating variability (e.g., subject-level standard deviation, mean
square of successive differences), MixWILD recognizes that subjects can vary in terms of the
number of observations in the analysis and thus provides unbiased estimates with varying
degrees of precision. MixWILD can also test interactions between within-subject variance and
subject-level predictors such as trait self-control. In MixWILD, a mixed-effects model was
applied in the first stage (level-1 EMA observation, level-2 person) to determine whether there
are significant random subject intercept and within-subject variance in the time-varying
outcomes (i.e., positive-activated affect, positive-deactivated affect, negative-activated affect,
negative-deactivated affect). To test the study objectives, a combination of a mixed-effects
location scale (MELS) model (first stage) with a linear regression model (second stage)
simultaneously modeled variability in affect (i.e., within-subject variance) to predict overall
levels of physical activity (Hedeker & Nordgren, 2013).
The first stage models estimated subject-level mean and subject-level variability in affect
(i.e., positive-activated, positive-deactivated, negative-activated, negative-deactivated). The first
stage models also examined whether covariates (i.e., weekend day, time of day, burst number)
were associated with mean and within-subject variance of affect. For simplicity, burst number
was treated as a linear predictor in the first stage models. In the second stage, a subject’s mean
level and variability in affect (estimated in the first stage model) was used to predict overall
physical activity.
Equation 1: 𝑦 𝑖𝑗
= 𝛽 0
+ 𝜐 0 𝑖 + 𝝌 𝒊𝒋
𝜷 + 𝜖 𝑖𝑗

33

In the mean model (Equation 1), yij is the outcome for the time-varying variable of
subject i (i = 1, 2,...N) on occasion j (1,2,...ni) with the following predictors: β0 is the intercept
coefficient, ν0i is the random subject intercept (location) effect (represents the subject’s mean;
accounts for the clustering of repeated observations within subjects), Xij are the regressors in the
model (i.e., weekend day, time of day, burst number) and β are the corresponding regression
coefficients that consist of a vector of p regressors. 𝜖 𝑖𝑗
is the error term (the deviation of a
subject’s observation from the subject’s mean; the fixed part of the model).
Equation 2: 𝜎 𝜀 𝑖𝑗
2
= ex p ⁡ ( 𝒘 𝒊𝒋
′
𝝉 + 𝜔 𝑖 )
In the within-subject variance sub-model (Equation 2), 𝜎 𝜀 𝑖𝑗
2
refers to the within-subject
variance, or consistency/inconsistency within subjects (how data vary within subjects). 𝒘 𝒊𝒋
′
are
the regressors and τ is the corresponding regression coefficient vector that consists of l
regressors. ωi is the random subject scale effect, which accounts for clustering of observations
within subjects and allows within-subject variance to vary across subjects beyond the
contribution of covariates (Dzubur et al., 2020). The exponential function is used to ensure that
the within-subject variance is positive. Both random location and scale effects were assumed to
be normally distributed and allowed to be correlated.
Equation 3: 𝑦 𝑖 ∗
= 𝛽 0
∗
+ 𝛽 1
∗
𝜐 0
𝑖 ̂ + 𝛽 2
∗
𝜔 𝑖 ̂ + 𝜒 𝑖 ∗
𝛽 ∗
+ 𝜖 𝑖 ∗

In the second stage model, a linear regression model predicted overall physical activity
levels using the random subject effects (i.e., location and scale) derived from first stage model as
predictors (Aim 1a). In the second stage model (Equation 3), the outcome 𝑦 𝑖 ∗
(average daily
MIMS-units/smartwatch valid wear time) is predicted by the intercept coefficient 𝛽 0
∗
, the
estimated random location of the first stage outcome 𝜐 0
𝑖 ̂ and its corresponding coefficient 𝛽 1
∗
, the
estimated random scale of the first stage outcome 𝜔 𝑖 ̂ and its corresponding coefficient 𝛽 2
∗
, other

34

regressors (i.e., age, sex at birth) 𝑋 𝑖 ∗
.and their corresponding regression coefficients of m
regressors coefficients β, and the error term 𝜖 𝑖 ∗
.
Equation 4: 𝑦 𝑖 ∗
= 𝛽 0
∗
+ 𝛽 1
∗
𝑆𝐶
𝑖 + 𝛽 2
∗
𝜐 0
𝑖 ̂ + 𝛽 3
∗
𝜔 𝑖 ̂ + 𝛽 4
∗
( 𝜔 𝑖 ̂ × 𝑆𝐶
𝑖 ) + 𝛽 5
∗
( 𝜐 0
𝑖 ̂ × 𝑆𝐶
𝑖 ) + ⁡ 𝜒 𝑖 ∗
𝛽 ∗
+ 𝜖 𝑖
To investigate whether trait self-control moderates the association between affective
variability and overall physical activity (Aim 1b), the same first stage mean and variance models
specified above were used (Equation 1). However, in the second stage (Equation 4) trait self-
control (level-2 regressor) was added as a main effect and interactions with the estimated random
location and scale derived from the first stage model. 𝑦 𝑖 ∗
is the outcome, 𝛽 0
∗
is the intercept
coefficient, 𝜇 0
𝑖 ⁡
is the random intercept of the Stage 2 outcome, 𝛽 1
∗
SCi is the subject-level
regressor, 𝛽 2
∗
𝜐 0
𝑖 ̂ ⁡is the estimated random location of the first stage outcome, 𝛽 4
∗
( 𝜔 𝑖 ̂ × 𝑆𝐶
𝑖 ) is the
interaction of trait self-control and random scale effects, 𝛽 5
∗
( 𝜐 0
𝑖 ̂ × 𝑆𝐶
𝑖 ) is the interaction of trait
self-control and random location effects, β are the corresponding regression coefficients that
consist of a vector of s regressors and 𝜖 𝑖 ∗
is the error term. For the two second stage models
(Equation 3 and 4), β
*
(i.e., 𝛽 0
∗
,) is used to distinguish that the fixed effects in the second stage
model are different from those β (i.e., β0) in the first stage model. 𝑦 𝑖 ∗
and 𝜖 𝑖 ∗
are also used to
distinguish the outcome and error terms in the second stage model.
In total, eight separate sets of models were conducted. Four first stage outcomes (i.e.,
positive-activated affect, positive-deactivated affect, negative-activated affect, negative-
deactivated affect) and one second stage outcome (i.e., overall physical activity) were
investigated in the study; therefore, four separate sets of models were conducted to address Aim
1a. To address Aim 1b, trait self-control was tested as a moderator (e.g., within-subject variance
of positive-activated affect × self-control) in an additional four second stage models.

35

Resampling
For the second stage analyses, if either a significant effect of mean (location) or
variability (scale) was observed then resampling was conducted in MixWILD (Carsey & Harden,
2013). Given that the location and scale random effects are estimated quantities—and estimated
quantities have an amount of uncertainty in these estimates for each subject—resampling is
necessary. For each subject, 500 resampled random effects were generated from a multivariate
normal distribution using the mean and variance estimates of these random effects for each
respective subject. Similar to multiple imputation, these resampled random effects were then
used to rerun the aforementioned modeling procedure 500 times.
Results
Data Availability
A total of 332 participants were consented into the study. After a trial period (i.e., testing
the smartphone EMA questions) 86 participants were removed from the study due to issues such
as low compliance, technical issues, or the participant withdrew themselves. A total of 246
smartwatches were mailed to participants. There was originally 38,778 days of physical activity
among the 246 participants during the six-month period. Days were removed from the analytic
dataset for several reasons: 1) the time zone on the smartwatch changed (i.e., participant changed
time zones via travel or daylight savings) (n=1020 days), 2) the smartwatch sampling unit
exceeded the maximum possible for one day (i.e., 86,400 seconds) (n=38 days), 3) there was no
smartwatch data (n=7102 days), 4) data were missing for smartwatch wear time (n=667 days,
and 6) smartwatch wear time was less than two hours or greater than 24 hours (n=436 days).
After removing data there were 29,764 days (76.75% of total data; M=124.54, SD=47.13 per

36

participant) among 239 participants. Trait self-control data from the baseline questionnaire was
missing for three participants, resulting in 236 participants.
For the EMA data, there were originally 138,959 EMA observations among 245
participants (i.e., one participant did not have any EMA data). EMA observations were then
further removed if the participant’s standard deviation for an affect variable during the six month
period was less than 0.2 (n=647 for positive-activated, n=0 for positive-deactivated, n=11,696
for negative-activated [standard deviation less than 0.3], n=1,282 for negative-deactivated) or if
there was missing data on the affect variable (n=37,516 for positive-activated, n=37,655 for
positive-deactivated, n=34,686 for negative-activated, n=36,458 for negative-deactivated). The
physical activity, self-control, and EMA data were then merged together. The final four first
stage models in MixWILD included: N=99,933 EMA observations and among 235 participants
(positive-activated affect), N=99,806 EMA observations among 236 participants (positive-
deactivated affect), N=90,899 EMA observations among 217 participants (negative-activated
affect), and N=99,704 EMA observations among 234 participants (negative-deactivated affect).
Subjects were removed from the second stage analysis for inestimable random effect values; the
final models examining the associations between subject-level variability and physical activity
included: N=216 participants (positive-activate affect), N=191 participants (positive-deactivated
affect), N=172 (negative-activated affect), and N=226 participants (negative-deactivated affect).
The likelihood of answering an EMA survey prompt was unrelated to the day of week
(i.e., weekend day vs. weekday), age, or sex at birth. An EMA survey prompt was less likely to
be answered in the morning compared to the evening (p<.05) and more likely to answered in the
afternoon compared to the evening (p<.05). The final analytic sample did not significantly differ
from the initially enrolled sample by demographic characteristics.

37

Descriptive Statistics
Descriptive statistics for participant characteristics are shown in Table 1 (N=236). On
average, participants were 23.60 (SD=3.18) years old at baseline. About 55% of the sample
indicated “female” for sex at birth, and 30% self-identified as Hispanic, Latino/a, or Spanish
origin. The average number of EMA observations per participant in the final analysis was 425
(range 53-697) for positive-activated affect, 422 (range 53-693) for positive-deactivated affect,
418 (range 52-693) for negative-activated affect, and 426 (range 53-699) for negative-
deactivated affect. On average, participants contributed 11 bursts in the analyses (SD=3.15;
range 2-13). The mean positive-activated score across all EMA observations in the final analysis
was 2.78 (SD=0.97), the mean positive-deactivated score was 3.07 (SD=1.09), the mean
negative-activated score was 1.95 (SD=1.00), and the mean negative-deactivated score was 1.94
(SD=0.85). On average, participants accumulated 8,963 MIMS-units during waking hours per
day (M=8963.93, SD=3734.41). In the final analytic sample (e.g., after removing days with less
than two hours of wear time), participants had 14 hours/day of valid wear time on average
(M=14.08, SD=3.31).
Table 1. Participant Characteristics (N=236)
Demographics
1
n (%)
Age in years (M ± SD)

23.60 + 3.18
Sex

Female 131 (55.5)
Male 105 (44.5)
Transgender Identity

Yes, transgender, male-to-female 1 (0.4)
Yes, transgender, female-to-male 3 (1.3)
Yes, transgender, gender nonconforming 11 (4.7)
No 218 (92.4)
Don’t know/not sure 3 (1.3)
Hispanic, Latino/a, or Spanish Origin
Yes 71 (30.1)
No 165 (69.9)

38

Race
2,a

American Indian or Alaska Native 11 (4.7)
Asian Indian 21 (8.9)
Black or African American 30 (12.7)
Chinese 29 (12.3)
Filipino 12 (5.1)
Guamanian or Chamorro 1 (0.4)
Japanese 5 (2.1)
Korean 9 (3.8)
Native Hawaiian/Other Pacific Islander 0 (0.0)
Other Asian 13 (5.5)
Other Pacific Islander 4 (1.7)
Samoan 0 (0.0)
Vietnamese 10 (4.2)
White 116 (49.2)
Work Status
2

Employed for wages 133 (56.4)
Self-employed 12 (5.1)
Out of work for 1 year or more 7 (3.0)
Out of work for less than 1 year 27 (11.4)
Homemaker 5 (2.1)
Student 121 (51.3)
Retired 0 (0.0)
Unable to work 6 (2.5)
a
data missing for 11 participants
1
participants were able to skip/not answer any of questions
2
participants were able to select all that apply

Variability in positive-activated affect predicting physical activity
The results of the first stage model for positive-activated affect are shown in Table 2. The
first stage model examined the effects of covariates (i.e., weekend vs. weekday, time of day,
burst number) on mean levels and within-subject variability in positive-activated affect. Findings
from the first stage model indicate that mean positive-activated affect was higher on weekend
days compared to weekdays (β=0.05, p<.001). Mean positive-activated affect was lower on
mornings (e.g., 12 A.M.-12 P.M.) versus evenings (5 P.M.-12 A.M.) (β=-0.04, p<.001). Within-
subject variability in positive-activated affect was significantly greater than zero on the
exponential scale (τ=-0.80, p<.001), suggesting that on average, people varied significantly

39

within themselves on their ratings of positive-activated affect. Within-subject variance (i.e.,
variability) in positive-activated affect was greater on weekend days compared to weekdays
(τ=0.04, p<.001) and in the morning (τ=0.05, p<.001) compared to evening, whereas variability
was lower in the afternoons (12 P.M.-5 P.M.) compared to evenings (τ=-0.03, p<.001). Within-
subject variance in positive-activated affect was also associated with burst number, such that
variability decreased as the study progressed (τ=-0.05, p<.001).
The results of the second stage model are shown in Table 6, which examined the
associations between subject-level mean and variability in positive-activated affect and overall
physical activity (Aim 1a). After controlling for age and self-reported sex at birth, subject-level
mean was associated with overall physical activity, such that individuals with higher mean levels
of feeling happy and energetic had greater overall levels of physical activity (β=0.02, p<.05).
Subject-level variability in positive-activated affect was not associated with overall physical
activity (β=0.01, p=.24). Age was a significant predictor of overall physical activity with older
individuals engaging in more overall physical activity (β=0.01, p<.05). Sex at birth was also a
significant predictor with females engaging in more overall physical activity compared to males
(β=0.03, p<.05). The interaction between subject-level mean and variability in positive-activated
affect was not significant (β=0.01, p=.19).
To address Aim 1b, a separate stage 2 model was conducted to examine whether self-
control moderated the association between subject-level variability with overall physical activity.
Self-control was not associated with overall physical activity (β=-0.0003, p=.41). After
controlling for main effects and covariates (i.e., age, sex at birth), the interaction term variability
in positive-activated affect × self-control was not significant suggesting that self-control did not
moderate the effect of subject-level variability in positive-activated affect on overall physical

40

activity (β=0.0004, p=.39). Ancillary to Aim 1b, the interaction between self-control and subject-
level mean of positive-activated affect was tested in the second stage model. Self-control did not
moderate the effect of subject-level mean of positive-activated affect on overall physical activity
(β=-0.0002, p=.47).
Variability in positive-deactivated affect predicting physical activity
The first stage model examined the effects of covariates (i.e., weekend vs. weekday, time
of day) on mean levels of and within-subject variability in positive-deactivated affect; results are
shown in Table 3. Findings indicated that ratings of positive-deactivated affect tended to be
higher on weekend days compared to weekdays (β=0.08, p<.001). Mean levels of feeling relaxed
were lower in the mornings (β=-0.05, p<.05) and afternoons (β=-0.05, p<.05) compared to the
evening. The within-subject variability in positive-deactivated affect was significantly greater
than zero on the exponential scale (τ=-0.46, p<.001), suggesting that on average, people varied
within themselves on their ratings of positive-deactivated affect. There were no significant
differences in within-subject variability in positive-deactivated affect between days of the week
or time of day. Within-subject variance in positive-deactivated affect was associated with burst
number, such that variability decreased as the study progressed (τ=-0.04, p<.001).
The results of the second stage model—which examined the association between subject-
level mean and variability in positive-deactivated affect and physical activity (Aim 1a)—are
shown in Table 6. After controlling for age and sex at birth, neither subject-level mean (β=0.001,
p=.92) nor variability (β=0.01, p=.18) in positive-deactivated affect were significantly associated
with physical activity. Age was a significant predictor of overall physical activity with older
individuals engaging in more overall physical activity (β=0.01, p<.05). Sex at birth was also a
significant predictor with females engaging in more overall physical activity compared to males

41

(β=0.03, p<.05). The interaction between mean and variability in positive-deactivated was also
included in model; the interaction was significant suggesting that the magnitude of the
relationship between variability in positive-deactivated and physical activity increased (i.e.,
became more positive) for individuals with higher mean levels of positive-deactivated affect
(β=0.02, p<.05).
A separate stage 2 model examined the interaction between self-control and subject-level
variability in positive-deactivated affect (Aim 1b). The results are shown in Table 4. Self-control
was not associated with overall physical activity (β=-0.0004, p=.92). After controlling for main
effects and covariates (i.e., age, sex at birth), self-control did not significantly moderate the
association between subject-level variability in positive-deactivated affect and overall physical
activity (β=0.001, p=.16). Ancillary to Aim 1b, the interaction between self-control and subject-
level mean of positive-deactivated affect was tested in the second stage model. Self-control did
not moderate the effect of subject-level mean of positive-deactivated affect on overall physical
activity (β=-0.0002, p=.58)
Variability in negative-activated affect predicting physical activity
The results of the first stage model for negative-activated affect are shown in Table 4.
The first stage model examined the effects of covariates (i.e., weekend vs. weekday, time of day,
burst number) on mean levels of and within-subject variability in negative-activated affect.
Findings indicate that mean levels negative-activated affect were lower on weekend days
compared to weekdays (β=-0.10, p<.001). Mean levels of feeling stressed and tense were higher
in the mornings (β=0.02, p<.05) and afternoons (β=0.03, p<.05) compared to evenings. The
within-subject variability in negative-activated affect was significantly greater than zero on the
exponential scale (τ=-0.82, p<.001), suggesting that on average, people varied significantly

42

within themselves on their ratings of negative-activated affect. Variability in negative-activated
was lower on the weekends compared to weekday (τ=-0.16, p<.001), and greater in the mornings
(τ=0.03, p<.05) and afternoons (τ=0.05, p<.001) compared to evenings. Variability in negative-
activated affect decreased as the study progressed (τ=-0.03, p<.001).
Table 7 displays the results from the second stage model testing the associations between
subject-level mean and variability in negative-activated affect and physical activity (Aim 1a).
After controlling for age and sex at birth, neither subject-level mean (β=-0.01, p=.35) nor
variability (β=-0.0003, p=.98) in negative-activated affect were significantly associated with
physical activity. Age was a significant predictor of overall physical activity with older
individuals engaging in more overall physical activity (β=0.01, p<.05). Sex at birth was also a
significant predictor with females engaging in more overall physical activity compared to males
(β=0.02, p<.05). The interaction between subject-level mean and variability in negative-activated
affect was not significant (β=-0.0002, p=.99).
To address Aim 1b, a separate stage 2 model examined whether self-control moderated
the association between subject-level variability in negative-activated affect and overall physical
activity. Self-control was not associated with overall physical activity (β=-0.0001, p=.87). After
controlling for main effects and covariates (i.e., age, sex at birth), the interaction term variability
in negative-activated affect × self-control was not significant, suggesting that self-control did not
moderate the effect of subject-level variability in negative-activated affect on overall physical
activity (β=-0.00001, p=.99). Ancillary to Aim 1b, the interaction between self-control and
subject-level mean of negative-activated affect was tested in the second stage model. Self-control
did not moderate the effect of subject-level mean of negative-activated affect on overall physical
activity (β=0.0003, p=.45).

43

Variability in negative-deactivated affect predicting physical activity
The first stage model examined the effects of covariates (i.e., weekend vs. weekday, time
of day, burst number) on mean levels of and within-subject variability in negative-deactivated
affect; results are shown in Table 5. Findings indicated that ratings of negative-deactivated affect
tended to be lower on weekend days compared to weekdays (β=-0.03, p<.001). Mean levels of
negative-deactivated affect were lower in the morning (β=-0.02, p<.001). and afternoon (β=-
0.03, p<.001) compared to the evening. The within-subject variability in negative-deactivated
affect was significantly greater than zero on the exponential scale (τ=-1.06, p<.001), suggesting
that on average, people varied within themselves on their ratings of feeling sad and fatigued.
Variability in negative-deactivated affect was lower on the weekends compared to weekdays (τ=-
0.03, p<.001) and lower in afternoons compared to evenings (τ=-0.07, p<.001). Variability in
negative-deactivated affect decreased as the study progressed (τ=-0.02, p<.001).
The results of the second stage— which examined the associations between subject-level
mean and variability in negative-deactivated affect and physical activity (Aim 1a)—model are
shown in Table 7. After controlling for age and sex at birth, both subject-level mean (β=-0.02,
p<.05) and subject-level variability (β=0.02, p<.05) in negative-deactivated affect were
associated with physical activity. These findings suggest that higher mean levels of feeling sad
and fatigued were associated with less physical activity, whereas greater variability in feeling sad
and fatigued was associated with more physical activity. Age was a significant predictor of
overall physical activity with older individuals engaging in more overall physical activity
(β=0.01, p<.05). Sex at birth was also a significant predictor with females engaging in more
overall physical activity compared to males (β=0.03, p<.05). The interaction between mean and
variability in negative-deactivated was not significant (β=0.004, p=.45).

44

A separate stage 2 model examined the interaction between self-control and subject-level
variability in positive-deactivated affect (Aim 1b). The results are shown in Table 4. Self-control
was not associated with overall physical activity (β=-0.0003, p=.39). After controlling for main
effects and covariates (i.e., age, sex at birth), self-control did not significantly moderate the
associations between subject-level variability in negative-deactivated affect and overall physical
activity (β=0.0001, p=.59). Ancillary to Aim 1b, the interaction between self-control and subject-
level mean of negative-deactivated affect was tested in the second stage model. Self-control did
not moderate the effect of subject-level mean of negative-deactivated affect on overall physical
activity (β=0.0002, p=.57).
Table 2. Results from first-stage mixed-effects location scale modeling for positive-
activated affect
Positive-activated Affect
(happy, energetic)
a

Estimate (SE) p
Mean Model (β)
Intercept 2.78 (0.05) <.001
Weekend
b
0.05 (0.004) <.001
Morning
c
-0.04 (0.005) <.001
Afternoon
d
0.001 (0.004) .83
Burst -0.001 (0.001) .11
Within-subject Variance Model (τ)
Intercept -0.80 (0.04) <.001
Weekend
b
0.04 (0.01) <.001
Morning
c
0.05 (0.01) <.001
Afternoon
d
-0.03 (0.01) <.01
Burst -0.05 (0.001) <.001
a
N=235 participants and N=99,933 EMA observations;
b
Weekend day versus weekday
c
Morning
(12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Afternoon (12 P.M. – 5 P.M.) versus
evening (5 P.M. – 12 A.M.)

45

Table 3. Results from first-stage mixed-effects location scale modeling for positive-
deactivated affect
Positive-deactivated Affect
(relaxed)
a

Estimate (SE) p
Mean Model (β)
Intercept 2.94 (0.05) <.001
Weekend
b
0.08 (0.005) <.001
Morning
c
-0.05 (0.01) <.001
Afternoon
d
-0.05 (0.01) <.001
Burst 0.01 (0.001) <.001
Within-subject Variance Model (τ)
Intercept -0.46 (0.05) <.001
Weekend
b
0.01 (0.01) .29
Morning
c
-0.002 (0.01) .87
Afternoon
d
-0.001 (0.01) .91
Burst -0.04 (0.001) <.001
a
N=236 participants and N=99,806 EMA observations ;
b
Weekend day versus weekday
c
Morning
(12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Afternoon (12 P.M. – 5 P.M.) versus
evening (5 P.M. – 12 A.M.)

46

Table 4. Results from first-stage mixed-effects location scale modeling for negative-
activated affect
Negative-activated Affect
(stressed, tense)
a

Estimate (SE) p
Mean Model (β)
Intercept 1.89 (0.06) <.001
Weekend
b
-0.10 (0.005) <.001
Morning
c
0.02 (0.01) <.001
Afternoon
d
0.03 (0.01) <.001
Burst 0.01 (0.001) <.001
Within-subject Variance Model (τ)
Intercept -0.82 (0.01) <.001
Weekend
b
-0.16 (0.01) <.001
Morning
c
0.03 (0.01) .02
Afternoon
d
0.05 (0.01) <.001
Burst -0.03 (0.001) <.001
a
N=217 participants and N=90,899 EMA observations;
b
Weekend day versus weekday
c
Morning
(12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Afternoon (12 P.M. – 5 P.M.) versus
evening (5 P.M. – 12 A.M.)

47

Table 5. Results from first-stage mixed-effects location scale modeling for negative-
deactivated affect
Negative-deactivated Affect
(sad, fatigued)
a

Estimate (SE) p
Mean Model (β)
Intercept 1.90 (0.04) <.001
Weekend
b
-0.03 (0.003) <.001
Morning
c
-0.02 (0.004) <.001
Afternoon
d
-0.03 (0.004) <.001
Burst 0.01 (0.001) <.001
Within-subject Variance Model (τ)
Intercept -1.06 (0.05) <.001
Weekend
b
-0.03 (0.01) <.001
Morning
c
-0.01 (0.01) .39
Afternoon
d
-0.07 (0.01) <.001
Burst -0.02 (0.001) <.001
a
N=234 participants and N=99,704 EMA observations;
b
Weekend day versus weekday
c
Morning
(12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Afternoon (12 P.M. – 5 P.M.) versus
evening (5 P.M. – 12 A.M.)

48

Table 6. Results from second-stage mixed-effects location scale modeling for positive-
activated and positive-deactivated affect

Main Effects Models Interaction Models
Physical Activity Physical Activity
Estimate (SE) p Estimate (SE) p
Positive-activated Affect
a

Intercept 2.64 (0.05) <.001 2.68 (0.06) <.001
Age 0.01 (0.002) .01 0.01 (0.002) .01
Sex at birth (female) 0.03 (0.01) .01 0.04 (0.01) .01
Mean level of positive-activated affect 0.02 (0.01) .02 0.05 (0.04) .24
Variability in positive-activated affect 0.01 (0.01) .24 -0.04 (0.05) .48
Mean × variability of positive-activated affect 0.01 (0.01) .19 0.01 (0.01) .37
Self-control –––– –––– -0.0003 (0.001) .41
Self-control × mean level of positive-activated affect –––– –––– -0.0002 (0.001) .47
Self-control × variability in positive-activated affect –––– –––– 0.0004 (0.001) .39
Positive-deactivated Affect
b

Intercept 2.65 (0.05) <.001 2.66 (0.06) <.001
Age 0.01 (0.002) .01 0.01 (0.002) .01
Sex at birth (female) 0.03 (0.01) .03 0.03 (0.01) .02
Mean level of positive-deactivated affect 0.001 (0.01) .92 0.03 (0.04) .55
Variability in positive-deactivated affect 0.01 (0.01) .18 -0.06 (0.05) .24
Mean × variability of positive-deactivated affect 0.02 (0.01) .01 0.01 (0.01) .05
Self-control –––– –––– -0.0004 (0.0004) .92
Self-control × mean level of positive-deactivated affect –––– –––– -0.0002 (0.0003) .58
Self-control × variability in positive-deactivated affect –––– –––– 0.001 (0.0004) .16
a
N=216 participants;
b
N=191 participants

49

Table 7. Results from second-stage mixed-effects location scale modeling for negative-
activated and negative-deactivated affect

Main Effects Models Interaction Models
Physical Activity Physical Activity
Estimate (SE) p Estimate (SE) p
Negative-activated Affect
a

Intercept 2.65 (0.05) <.001 2.66 (0.06) <.001
Age 0.01 (0.002) .01 0.01 (0.002) .01
Sex at birth (female) 0.02 (0.01) .10 0.02 (0.01) .10
Mean level of negative-activated affect -0.01 (0.01) .35 -0.04 (0.04) .87
Variability in negative-activated affect -0.0003 (0.01) .98 0.001 (0.08) .99
Mean × variability of negative-activated affect -0.0002 (0.001) .99 -0.0004 (0.01) .97
Self-control –––– –––– -0.0001 (0.0004) .87
Self-control × mean level of negative-activated affect –––– –––– 0.0003 (0.0004) .45
Self-control × variability in negative-activated affect –––– –––– -0.00001 (0.001) .99
Negative-deactivated Affect
b

Intercept 2.65 (0.05) <.001 2.69 (0.06) <.001
Age 0.01 (0.002) .01 0.01 (0.002) .01
Sex at birth (female) 0.03 (0.01) .03 0.03 (0.01) .02
Mean level of negative-deactivated affect -0.02 (0.01) .01 -0.05 (0.05) .30
Variability in negative -deactivated affect 0.02 (0.01) .04 -0.01 (0.05) .84
Mean × variability of negative-deactivated affect 0.004 (0.01) .45 0.01 (0.01) .24
Self-control –––– –––– -0.0003 (0.0004) .39
Self-control × mean level of negative-deactivated affect –––– –––– 0.0002 (0.0004) .57
Self-control × variability in negative-deactivated affect –––– –––– 0.0002 (0.004) .59
a
N=172 participants;
b
N=226 participants

Discussion

The current study used novel methodology to assess and model subject-level variability
in affect (i.e., positive-activated, positive-deactivated, negative-activated, negative-deactivated
affect). We examined the associations between subject-level variability in affect and overall
levels of physical activity among a sample of young adults, and further examined whether trait
self-control moderated these associations. Results generally indicated that affective variability
was not associated with physical activity, such that individuals who experienced greater
variability in their affect did not necessarily engage in more or less physical activity. However,
the exception to this pattern of findings was that contrary to hypotheses, subject-level variability
in negative-deactivated affect (i.e., feeling sad and fatigued) was associated with more overall

50

physical activity. The results also indicate that self-control did not moderate the associations
between affective variability and physical activity. Despite the null findings for the self-control
moderation effect, the current study extends previous research in affect and physical activity,
which has largely looked at mean or momentary single time point affect, by considering the way
in which affect fluctuates above or below the mean.
The first objective of the study investigated whether subject-level variability in affect
predicated overall levels of physical activity; it was hypothesized that individuals who
experienced more variability in positive-activated, positive-deactivated, negative-activated, and
negative-deactivated affect would engage in lower levels of physical activity. Prior research in
affect dynamics indicates that greater fluctuations in intensity is associated with depression,
bipolar personality disorder, drinking, and binge eating (Anestis et al., 2009; Bos et al., 2019;
Chan et al., 2016; Mermelstein et al., 2010; Simons et al., 2014; Trull et al., 2008). The current
study did not report significant associations between subject-level variability in positive-
activated, positive-deactivated, or negative-deactivated affect and overall levels of physical
activity. The findings are partially in line with prior research which reported that greater subject-
level variability of feeling energetic, but not general positive affect per se, was associated with
overall levels of physical activity (Maher et al., 2019). In the current study, affective states were
grouped together based on the two fundamental dimensions of affect (i.e., valence and arousal),
resulting in combinations of positive/negative and activated/deactivated (Russell, 1980). While
this method aids in making more precise conclusions on affect—compared to the traditional
method of combining valence and arousal through the assessment of positive affect and negative
affect—it can be difficult to compare findings across different research studies. For example,
Maher and colleagues examined the associations between positive affect (i.e., happy, joyful,

51

cheerful, calm) and arousal (i.e., energetic) (Maher et al., 2019); it is challenging to completely
compare previous findings to the current study given that happy and energetic were assessed
together through positive-activated affect and feeling relaxed was assessed separately as
positive-deactivated affect. The field of exercise psychology may benefit from utilizing common
affective measures to better compare studies and findings; however, the conceptualization and
operationalization of affect continues to be an area of debate (Ekkekakis, 2013; Ekkekakis &
Petruzzello, 2000). One alternative method to assessing affect is to take the ‘distinct states
approach’ in which a single state is used (e.g., feeling happy) rather than averaging scores across
multiple states to create constructs like positive and negative affect (Ekkekakis, 2013). This
strategy could promote consistency across studies and enable researchers to potentially pinpoint
potential associations between affect and physical activity. For example, variability in arousal
states (e.g., energetic, fatigued) may be associated with physical activity. Arousal may be
particularly relevant for physical activity, where arousal/activation states may be an antecedent,
concomitant, or consequence of activity and exertion.
Study findings may also differ from prior research due the differences in study
populations. Preliminary evidence for affective variability (i.e., feeling energetic) predicting
physical activity has been done among samples including both children and adults, whereas the
current study sample consisted of only young adults ages 18-29 years old. Variability in
momentary affective states and regulation of affective states may differ across the lifespan
(Ebner & Fischer, 2014; Zimmermann & Iwanski, 2014). Therefore, affective variability may be
associated with physical activity only among sub-populations; additional research in young adult
and among other samples of differing ages and clinical conditions is warranted. For example,
affect variability may be an important predictor of physical activity in individuals with

52

psychiatric disorders, given the greater severity of affect dysregulation in these populations
(Aldao et al., 2010; Marwaha et al., 2014).
Attenuated associations may also stem from the fact that relationships between affective
variability and physical activity may occur at finer-grained timescales (e.g., days, hours). The
extent to which an individual’s affect fluctuates from hour-to-hour may be more predictive of
whether they decide to engage or not engage in physical activity later that day. These processes
may unfold over shorter timescales, which may not fully be captured by examining subject-level
associations (e.g., between-subject differences). Examining whether daily fluctuations in
affective variability impacts same-day physical activity may provide insight into the potential
relationships between affective variability and activity.
Contrary to the hypothesis, variability in negative-deactivated affect (i.e., feeling sad and
fatigued) was positively associated with physical activity, suggesting that individuals who
experienced more fluctuations in feeling sad and fatigued engaged in more overall physical
activity. This is conflicting with other research, which indicates that greater variability in
negative affect is associated with increased engagement in maladaptive health behaviors like
alcohol consumption, binge eating, and substance use (Anestis et al., 2009; Mermelstein et al.,
2010; Simons et al., 2014). Theoretical frameworks posit that individuals engage in maladaptive
behaviors (that provide immediate relief or something) as a coping mechanism or to regulate
their fluctuations in affect. It is possible that individuals are engaging in physical activity, while
a positive health behavior, in a similar manner to cope with these fluctuations in affect. Prior
research demonstrates that some individuals engage in physical activity to help relieve negative
feelings or to cope with emotional stress (Schultchen et al., 2019; Stults-Kolehmainen & Sinha,
2014; Wipfli et al., 2008); this may result in short term deviations from negative affective states

53

which may contribute to the observed fluctuations. The current study is unable to determine if
there is a causal relationship between affective variability and physical activity. However, this
study is the first known physical activity study to be able to model variability in negative affect,
possibly due to the study design that involves frequent assessments and a long study period.
Future research could benefit by employing similar intensive longitudinal study designs through
smartphones and smartwatches to fully capture the range of affective states.
The second objective of the study was to examine whether self-control moderates the
associations between subject-level variability in affect and physical activity. It was hypothesized
that the strength of association between affective variability and physical activity would be
stronger among individuals who have lower trait self-control; however, study results indicate that
self-control did not moderate these associations. While prior theoretical models and research
suggest that self-control predicts physical activity (Crescioni et al., 2011; Englert, 2016; Hagger
et al., 2010b), it is possible that both affective variability and self-control may play an influential
role on activity in the moment rather than at the subject-level. An individuals’ self-control and
self-regulatory capacity may differ day-to-day based on internal and environmental events.
Empirical research suggests that inhibitory control (i.e., the ability to stop, change, or delay
inappropriate behaviors) is not a stable trait, but rather fluctuates in response to internal events
(e.g., stress, depletion of self-control resources) or environmental contexts (e.g., behavior-related
cues) (Muraven et al., 2005; Muraven & Baumeister, 2000; Zack et al., 2011). For example,
short-term fluctuations in inhibitory control predicted increased alcohol consumption (Jones et
al., 2018). Future research is needed to determine whether self-control varies at the within-
subject level, and subsequently determine if self-control moderates or mediates associations
between affective variability and physical activity at the day-level. In addition, self-control is a

54

multi-dimensional construct, thus certain aspects of self-control may be more important in
regards to affective variability and physical activity (Hagger et al., 2021). Future studies may
benefit by examining different self-control dimensions and measures to fully understand the
potential associations.
Despite the null findings, additional research is warranted to determine whether affective
variability predicts physical activity, and in what circumstances. For example, there may be
associations between affective variability and physical activity among different sub-populations
that were not examined in this study, such as those with clinical conditions, those who are less
physically active, or among those who engage in habitual physical activity. Thus far, there are
few published studies exploring the associations between affect dynamics (i.e., variability) and
physical activity levels. By understanding the antecedents to physical activity, public health
programs seeking to increase physical activity can target influential correlates such as affect or
self-control. Future research is also needed to explore the moderating role of self-control on daily
physical activity, and subsequently determine if developing tailored strategies to increase
momentary self-control is efficacious. For example, individuals with low self-control may be
vulnerable to greater depletion of self-regulatory resources during challenging times (e.g.,
beginning a new job, starting a family, grief or loss, relationship break-ups), which may
adversely influence their ability to engage in regular physical activity (Hagger et al., 2010b).
Smartphone-based interventions can target momentary self-control in order to increase physical
activity through short coaching sessions, goal setting, behavioral tasks, and changing behaviors
in given situations (Allemand et al., 2020; Stieger et al., 2021).

55

Limitations
Despite the strengths of the proposed study, such as coupling smartphone and smartwatch
technology to continuously capture activity data across six months and the ability to model
variability in both positive and negative affective constructs, there are some limitations. Analyses
do not allow us to infer causality between affective variability and physical activity. Another
study limitation is that data collection occurred from March 2020 to August 2022, which
overlapped with the COVID-19 pandemic (declared a national U.S. emergency in March 2020)
and its related closures and policies; this may have drastically affected individuals’ daily lives,
including affect and physical activity. The study is also limited by using only a few items to
assess affect, which may not fully capture these constructs or variability in the constructs. In
particular, only one item (“calm”) was used to assess positive-deactivated affect. A limited
number of items were included in each EMA prompt to reduce participant burden. It should be
noted that physical activity, operationalized by MIMS-units, was lower in this sample compared
to population referenced MIMS-unit percentiles for the similar age range from National Health
and Nutrition Examination Survey data (8,963 vs. 14,309). However, the average in this sample
was not weighted. In addition, movement was likely lower among our sample given that data
collection occurred during the COVID-19 pandemic (Dunton et al., 2020; Meyer et al., 2020).
When interpreting the study findings, it is important to note that the outcome variable was
aggregated across six months to represent a person, when a person’s activity levels could change
over time. Finally, the study findings may not be generalizable to other populations, such as
adults of different age ranges, individuals who do not regularly engage in activity, or people who
are not willing or unable to partake in an intensive smartphone and smartwatch study.

56

Conclusions
The current study expanded upon prior research by considering the role of affect
dynamics (i.e., affective variability), above and beyond mean levels or momentary levels of
affect, on overall levels of physical activity. Results indicate that subject-level variability in
negative-deactivated affect (i.e., feeling sad and fatigued) was associated with greater levels of
physical activity, whereas variability in positive-activated, positive-deactivated, and negative-
activated affect did not predict physical activity. In addition, trait self-control did not moderate
the associations between subject-level variability in affect and physical activity. Despite the
majority null findings, the study contributes to the limited, but growing, literature base on
affective variability and physical activity. The use of novel statistical methods to estimate
variability and subsequently predict subject-level outcomes can be applied in future studies to
examine unique research questions in health behavior and public health.

57

Chapter 3: Investigating the day-level associations between affective variability and
physical activity using ecological momentary assessment
Abstract
Background: Understanding affect as a determinant of physical activity has gained increased
attention in health behavior research. Fluctuations in affect intensity from moment-to-moment
(i.e., affective variability) may interfere with cognitive and regulatory processes, making it
difficult to engage in goal-directed behaviors such as physical activity. Preliminary evidence
indicates that those with greater trait-level affective variability engage in lower levels of habitual
physical activity. However, the extent to which daily fluctuations in affect variability impact
same-day physical activity levels is unknown. This study used ecological momentary assessment
(EMA) to investigate day-level associations between affective variability (i.e., within-subject
variance) and physical activity.
Methods: Young adults (N=231, M=23.58+3.02 years) provided three months of smartphone-
based EMA and smartwatch-based activity data. Every two weeks, participants completed a 4-
day EMA measurement burst (M=5.17+1.28 bursts per participant). Bursts consisted of hourly
randomly-prompted EMA surveys assessing momentary positive-activated (happy, energetic),
positive-deactivated (relaxed), negative-activated (tense, stressed), and negative-deactivated
(sad, fatigued) affect. Participants continuously wore a smartwatch to measure physical activity
across the three months. Mixed-effects location scale modeling examined the day-level
associations of affective variability (i.e., positive-activated, positive-deactivated, negative-
activated, and negative-deactivated) and physical activity, controlling for covariates such as
mean levels of affect, between-subject effects of physical activity, time of day, day of week, day
in study, and smartwatch wear time.

58

Results: There were 41,546 completed EMA surveys (M=182.22+69.82 per participant)
included in the analyses. Above and beyond mean levels of affect, greater day-level variability in
positive-activated affect was associated with greater physical activity on that same day compared
to other days (τ=0.01, p<.001), whereas greater day-level variability in negative-deactivated
affect was associated with less physical activity on that same day compared to other days (τ=-
0.01, p<.001). Day-level variability in positive-deactivated affect or negative-activated affect
were not associated with day-level physical activity (ps>.05)
Conclusions: Individuals were less active on days with greater variability in feeling sad and
fatigued but more active on days with greater variability in feeling happy and energetic.
Understanding the dynamic relationships of affective variability with day-level physical activity
can strengthen physical activity interventions by considering how these processes differ within
individuals and unfold within the context of daily life. Future research should examine causal
pathways between affective variability and physical activity across the day.

59

Introduction
Regular physical activity can reduce the risk for more than 25 chronic conditions,
including cardiovascular disease, hypertension, type 2 diabetes, osteoporosis, breast cancer, and
colon cancer (Pedersen & Saltin, 2015; Piercy et al., 2018; Warburton & Bredin, 2017). Despite
the benefits of physical activity, only one in four adults in the United States (U.S.) met the
physical activity guidelines for aerobic and muscle-strengthening activities (2018 Physical
Activity Guidelines Advisory Committee Scientific Report, 2018). Among the young adult
population (e.g., ages 18-29), only 10% met national recommendations (Tucker et al., 2011).
Young adulthood is marked by substantial declines in physical activity, making it an important
public health issue (Gordon-Larsen et al., 2004; Zick et al., 2007). Given that behaviors shaped
during the formative years of young adulthood persist throughout the rest of adulthood, adopting
and maintaining healthy habits like physical activity is important for the short- and long-term
health of adults (Telama et al., 2014).
To increase physical activity, health behavior theories and empirical research have
considered affect as a key determinant of health behaviors. According to the Affect and Health
Behavior Framework (AHBF), incidental affect—how one feels throughout the day outside the
context of the target behavior—is a key affective correlate of health behaviors (e.g., physical
activity, substance use, eating) (Williams & Evans, 2014). A growing body of literature has
examined acute associations of affect with subsequent physical activity. Evidence suggests that
randomly assessed levels of incidental affect predict subsequent levels of physical activity across
short time scales (e.g., minutes, hours) and throughout the day. For instance, a systematic review
(Liao et al., 2015) on the acute relationships between affective states and physical activity in
naturalistic settings concluded that elevated positive affective states were associated with greater

60

physical activity over the next few hours (Carels et al., 2007; Dunton et al., 2009; Schwerdtfeger
et al., 2010). However, there was inconsistent evidence for the role of negative affective states as
a momentary antecedent for physical activity such that some studies found negative affect to be
inversely associated with subsequent physical activity (Niermann et al., 2016) while other studies
reported non-significant associations (Dunton et al., 2014; Kim et al., 2020; Schwerdtfeger et al.,
2010; Wichers et al., 2012). Overall, knowledge on the acute or momentary associations of
incidental affect and physical activity can have important implications for promoting physical
activity in real-time and in naturalistic settings.
While prior research has examined the impact of momentary affect on physical activity
(i.e., at a single time point), there is little knowledge on whether the dynamic nature of affect
also influences physical activity. Incidental affect can fluctuate over time and within individuals
(i.e., intraindividual variability) (Eid & Diener, 1999; Gross, 2013). Individuals not only differ
from each other in their affect intensity, or mean level of affect, but also in their affect variability
(Davidson, 1998; Ebner-Priemer & Trull, 2012). As such, two individuals may have the same
mean level of affect per day but could differ substantially from each other in the degree to which
affect fluctuates in intensity from moment-to-moment throughout the day (Ebner-Priemer et al.,
2009; Nesselroade & Ram, 2004). Theoretical perspectives suggest that higher levels of affective
variability may impact one’s capacity to control behaviors (Baumeister et al., 1998; Weiss et al.,
2018). The strength model of self-control posits that as an individual experiences increased or
changing internal demands (e.g., negative affect, stress) that require regulation, the ability to
engage in health-promoting behaviors and limit health-risk behaviors diminishes (Baumeister et
al., 1998; Baumeister & Vohs, 2007). Related to affective variability, regulating one’s affect
across the day due to rapid fluctuations may lead to altered ability to control behaviors.

61

Consistently, empirical evidence on affective variability and health behaviors has shown greater
negative affect variability predicted engagement in unhealthy behaviors (Anestis et al., 2009;
Mermelstein et al., 2010; Simons et al., 2014).
Preliminary evidence suggests that trait-level affective variability is related to physical
activity engagement, but whether day-level variability influences physical activity remains
unclear. For example, while intraindividual variability in positive affect was not significantly
associated with overall levels of physical activity, greater intraindividual variability of feeling
energetic was associated with less overall physical activity (Maher et al., 2019). While there is
preliminary evidence for associations between trait affect variability and habitual activity
(Dunton et al., 2014; Kerrigan et al., 2020; Maher et al., 2019), the extent to which daily
fluctuations in affect variability impact same-day physical activity levels is unknown. To our
best knowledge, no published studies have investigated the within-subject or day-level
associations of affective variability and physical activity. Knowledge of associations at these
finer grained timescales (e.g., day-level associations) has important theoretical and clinical
implications. Extant theoretical models of daily physical activity already discuss the role of mean
levels of affect in predicting activity (Williams & Evans, 2014), but affective variability may be
a key predictor of activity independent of mean levels of affect.
Given that an individual’s affect can fluctuate throughout the day, real-time data capture
methodologies such as ecological momentary assessment (EMA) are preferable for collecting
information from individuals in their natural environments and increasing ecological validity
(Shiffman et al., 2008; Stone et al., 2007). Additionally, repeated assessments can elucidate
intraindividual differences and capture fluctuations in affect. EMA has been advocated for the
assessment of affective variability given that single measures tend to assess average extremity of

62

affect rather than the frequency or degree of fluctuations (Ebner-Priemer et al., 2009; Larsen,
1987). By understanding the dynamic relationships of affective variability and day-level physical
activity, future health behavior theories and programming can be strengthened by considering
how these processes unfold in the context in daily life.
Therefore, the objective of the study was to examine day-level associations between
affective variability (i.e., positive-activated, positive-deactivated, negative-activated, and
negative-deactivated affect) and physical activity (Aim 2). Previous literature on affective
determinants and physical activity has also been limited by typically combining valence
(pleasure-displeasure) and arousal (activate) when conceptualizing affect (e.g., positive affect,
negative affect), which may mask or attenuate associations. If the two fundamental dimensions
of affect—suggested by the circumplex model (i.e., valence and arousal)—are mixed, it is
difficult to make conclusions given that positive and negative affect refer to valence only
(Ekkekakis, 2013; Russell, 1980). The aforementioned four affective constructs were selected to
differentiate between the valence and arousal (Russell, 1980). It was hypothesized that greater
day-level affective variability (i.e., within-subject variance) will be associated with less physical
activity on that same day compared to usual. Specifically, individuals will experience less
affective variability (i.e., positive-activated affect, positive-deactivated affect, negative-activated
affect, negative-deactivated affect) on days when they engage in more physical activity
compared to their average levels of activity.
Methods
Study Design
The current study is a secondary data analysis of an intensive longitudinal study. The
Temporal Influences on Movement and Exercise (TIME) study is a prospective within-subject

63

observational study. Data were continuously collected across 12 months through smartphone and
smartwatch technology (Wang et al., 2022). Across the 12-month period, participants completed
four-day measurements bursts every two weeks. During each burst, participants completed
signal-contingent (i.e., randomly prompted) smartphone EMA. The current analyses used data
collected in the first three months of the study to maximize available data because attrition and
missing data increased the longer individuals participated in the study) and to answer the current
study’s objectives, which do not involve examining change over longer periods of time such as
months. This study was conducted in accordance with the Declaration of Helsinki and all aspects
of the study were approved by the Institutional Review Board at the University of Southern
California (HS-18-00605).
Recruitment and Participants
Young adults (ages 18-29 years old) were recruited across the U.S. on a rolling basis
from March 2020 until August 2021. Due to health and safety concerns from the COVID-19
pandemic (respiratory disease caused by the SARS-CoV-2 virus) and its related restrictions, all
recruitment and study procedures were conducted remotely. Participants were recruited through
several strategies: (1) individuals enrolled in the Happiness & Health Study (a prospective cohort
study of ninth-grade students that began in 2013) were sent emails (Leventhal et al., 2015); (2)
referrals from existing participants (i.e., word of mouth); (3) online/social media advertisements
(e.g., Twitter, Facebook); (4) emails to addresses on file from other University of Southern
California studies; and (5) emails to individuals through ResearchMatch (a national health
volunteer registry that was created by several academic institutions and supported by the U.S.
National Institutes of Health as part of the Clinical Translational Science Award program).
Participants were eligible for the study if they were 18-29 years old, living in the United States,

64

use an Android-based smartphone, able to speak and read in English, and intend within the next
12 months to engage in, or already engage in, recommended levels of physical activity (≥150
min/week moderate or ≥75 min/week vigorous intensity). Details on the study inclusion and
exclusion criteria are published elsewhere (Wang et al., 2022). Inclusion criteria were established
to: guarantee activity data in naturalistic settings; ensure the proper devices to download the
study software warrant analysis of sleep patterns; consider the health and safety of participants;
ensure that sufficient data is collected from the smartphone and smartwatch throughout the day.
Study Procedures
Potential participants were asked to complete an online interest form; upon determining
initial eligibility, study staff contacted participants by phone to ask additional eligibility
questions and provide more information about the study. Eligible participants attended a one-on-
one video conference orientation with a study staff member to review all parts of the study and
take informed consent. Participants then completed an orientation session to download the
custom TIME study smartphone application on their personal smartphone, how to use the study
app, and how to complete EMA surveys. Participants completed a baseline self-report electronic
questionnaire on Research Electronic Data Capture hosted at the University of Southern
California (Harris et al., 2009, 2019). Participants received a study smartwatch by mail.
Compensation was based compliance; participants received up to $80/month for competing the
EMA and wearing the smartwatch.
EMA data were collected through a custom TIME app developed for Android
smartphones and smartwatches. Each EMA burst lasted four consecutive days and assessments
were randomly scheduled for a block of days, with at least seven days in between each burst and
guaranteeing two weekend days and two weekdays within each burst. During each 4-day burst,

65

participants completed EMA surveys on their smartphone about every hour during their waking
hours. The TIME app allowed for personally customizable sleep and wake scheduling that was
adjustable daily to avoid prompting during sleep. Participants were prompted by push
notifications and asked to complete a 1–2-minute EMA survey. If no response was provided, up
to two reminder prompts were sent in 5-minute intervals. After this point, the EMA survey
become inaccessible. Participants were instructed to ignore any prompts that occurred during
incompatible activities (e.g., driving). Each burst consisted of signal-contingent (i.e., randomly-
prompted) question sets triggered approximately once every hour during waking hours (e.g., 14
prompts between 8 A.M.-10 P.M.)
Participants were loaned a Fossil Sport Gen 4 or Gen 5 smartwatch for the study.
Participants were asked to wear the smartwatch on one wrist of their choice consistently over the
study period except for: (1) one hour per day when it should be charged and (2) when the watch
would be exposed to water for extended periods of time (e.g., showering, swimming).
Participants were asked to refrain from installing health/fitness apps on their smartwatch because
they could interfere with the smartwatch sensor data collection and cause battery drainage.
Measures
Affective States
To reduce participant burden (i.e., each EMA survey could be finished in under 2
minutes), select items were chosen to represent the two fundamental dimensions of affect (i.e.,
valence and arousal) (Russell, 1980). Positive-activated, positive-deactivated, negative-activated,
and negative-deactivated affect were measured through the EMA surveys (Do et al., 2021;
Stevens et al., 2020). Participants were asked seven questions in the following format: “Right
now, how (affect term) do you feel?” Positive-activated affect (happy, energetic), positive-

66

deactivated affect (relaxed), negative-activated affect (tense, stressed), and negative-deactivated
affect (sad, fatigued) were assessed. The response options were on a unipolar scale: 1 = “Not at
all”; 2 = “A little”; 3 = “Moderately”; 4 = “Quite a bit”; 5 = “Extremely”. Scores were averaged
to create a composite score for each construct, respectively.
Physical Activity
Physical activity was measured passively using the wrist-worn smartwatches. Tri-axial
raw acceleration along X, Y, and Z axes on the smartwatch was measured using the embedded
accelerometer at a sampling rate of 50Hz. Monitor Independent Movement Summary (MIMS)
units were computed offline using the raw accelerometry data to summarize movement (John et
al., 2019). MIMS-units allow for measuring the amount of movement, independent of the type of
sensor (e.g., smartwatch, smartphone), and is a standardized, nonproprietary metric that can be
applied across different accelerometers. MIMS-unit age- and sex-specific percentiles allows for
comparisons across populations and between studies that use different sensors (Belcher et al.,
2021). MIMS for 1-second epochs were calculated to determine the total motion at varied
lengths of times longer than one second (e.g., minutes, hours). Smartwatch non-wear was
assessed through the sleep, wear, and non-wear (SWaN) algorithm, which classifies the raw
accelerometry data into sleep, wear, and non-wear classes for a 30-second window (Arguello et
al., 2018). For the current study, physical activity was operationalized as MIMS-units per minute
of wear time. This variable was calculated by dividing day-total MIMS-units during waking
hours by smartwatch wear time during waking hours (in minutes). Thus, sleep time was
removed. The day needed to have at least two hours of smartwatch wear time during waking
hours to be included in the analyses.

67

Participant Characteristics
Participant characteristics and demographics were collected through a baseline electronic
questionnaire. Participants self-reported the following characteristics: 1) age; 2) sex at birth
(female, male); 3) transgender; 4) Hispanic, Latino/a, or Spanish origin; 5) race; 6) educational
attainment; 7) work status; 8) marital status. Participants were able to skip/not answer questions
if they felt uncomfortable.
Statistical Analyses
Descriptive statistics (e.g., frequencies, means) described the study sample and variables
of interest. Prior to data analysis, EMA observations were excluded from the analytic data set if
the participant’s standard deviation for an affect variable during the entire 3-month period was
less than 0.2 to ensure there was at least a minimal degree of affect variability to allow variance
modeling. The current study used mixed-effects location scale (MELS) models to determine
whether there were day-level associations between affective variability and physical activity
(Dzubur et al., 2020; Hedeker et al., 2008). Analyses were conducted in MixWILD (Mixed
model analysis With Intensive Longitudinal Data; https://reach-lab.github.io/MixWildGUI/), an
open source statistical program that can answer unique research questions regarding time-
varying variables often collected through EMA (Dzubur et al., 2020). MixWILD conducts
MELS modeling, an extension of multi-level models that allows covariates to influence the
within-subject variance and allows each subject to have their own degree of within-subject
variance, above and beyond the effects of covariates. For example, a MELS model can evaluate
whether the mean-level of and variance in momentary affect are associated with covariates (e.g.,
daily physical activity) simultaneously, and additionally include a random location effect and
random scale effect in the mean model and within-subject variance model, respectively. In

68

addition, a MELS model recognizes that subjects differ in the number of observations
contributed to the analyses so it can provide unbiased estimates with varying degrees of
precision.
The main time-varying predictor of interest, day-level physical activity, was decomposed
into within-subject (level-1, day) and between-subject (level-2, person) effects in MixWILD. The
within-subject component is created by having the regressor centered at its subject-level mean
(any given day’s deviation from the subject’s own mean across all available days) and the
between-subject component is the subject-level mean (one’s own mean across all available days)
(Curran & Bauer, 2011). MELS models tested whether within-subject (i.e., day-level) physical
activity predicted affective variability, controlling for time-varying covariates (i.e., burst number,
day of week, time of day) and subject-varying covariates (i.e., between-subject physical activity,
sex at birth). For simplicity, burst number was treated as a linear predictor. One model was run
for each outcome (i.e., positive-activated affect, positive-deactivated affect, negative-activated
affect, negative-deactivated affect), resulting in a total of four MELS models. The MELS mean
model applied in the current study is displayed.
Equation 1: 𝑦 𝑖𝑗
= 𝛽 0
+ 𝜐 0 𝑖 + 𝛽 1
𝜒 𝑖𝑗
+ 𝝌 𝒊𝒋
′
𝜷 + 𝜖 𝑖𝑗

In Equation 1, yij is outcome for the time-varying variable of subject i (i = 1, 2, …N) on
occasion j (1, 2,..ni), β0 is the intercept coefficient, 𝜐 0
𝑖 is the random subject intercept (location)
effect (represents the subject’s mean; accounts for the clustering of repeated observations within
subjects), β1 is the regression coefficient of the main predictor, and Xij is the main time-varying
predictor (i.e., physical activity). X´ij are the other regressors in the model (e.g., covariates) and β
are the corresponding regression coefficients that consist of a vector of p regressors. 𝜖 𝑖𝑗
is the

69

error term (the deviation of a subject’s observation from the subject’s mean; the fixed part of the
model).
Equation 2: 𝜎 𝜀 𝑖𝑗
2
= ex p ⁡ ( 𝒘 𝒊𝒋
′
𝝉 + 𝜔 𝑖 )
The MELS within-subject variance model applied in the current study is displayed in
Equation 2. In the within-subject variance model, 𝜎 𝜀 𝑖𝑗
2
refers to the within-subject variance, or
consistency/inconsistency within subjects (how data vary within subjects). w´ij are the regressors
and τ is the corresponding regression coefficient vector that consists of l regressors. ωi is the
random subject scale effect, which accounts for clustering of observations within subjects and
allows the within-subject variance to vary across subjects beyond the contribution of covariates
(Dzubur et al., 2020). The exponential function is used to ensure that the resulting within-subject
variance is strictly positive. Both random location and scale effects were assumed to be normally
distributed and allowed to be correlated.
An example mean sub-model (Equation 3) and within-subject variance sub-model
(Equation 4) for assessing the association between within-subject physical activity and positive-
activated affect are shown. 𝜏 𝑙 𝜐 0 𝑖 is the association of the random location and scale effects.
Equation 3: 𝑝 𝑜𝑠 𝑖𝑡 𝑣𝑒 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑎 𝑡 𝑒 𝑑 ⁡ 𝑎𝑓𝑓 𝑒 𝑐 𝑡 𝑖𝑗
= ( 𝛽 0
+ 𝜐 0 𝑖 ) + 𝛽 𝑊𝑆
( 𝑝 ℎ 𝑦 𝑠 𝑖𝑐 𝑎𝑙 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑖 𝑡 𝑦 𝑖𝑗
−
𝑝 ℎ 𝑦 𝑠 𝑖𝑐 𝑎𝑙 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑖 𝑡 𝑦 𝑖 ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅
) + 𝛽 𝐵𝑆
( 𝑝 ℎ 𝑦 𝑠 𝑖𝑐 𝑎𝑙 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑖 𝑡 𝑦 𝑖 ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅
) + 𝛽 𝑇 𝑡 𝑖𝑚𝑒 ⁡ 𝑜𝑓 ⁡ 𝑑 𝑎𝑦 𝑖𝑗
+ 𝛽 𝑊 𝑤𝑒 𝑒 𝑘 𝑒 𝑛 𝑑 𝑖𝑗
+ 𝛽 𝐷 𝑏 𝑢 𝑟𝑠 𝑡 𝑖𝑗
+
𝛽 𝑆 𝑠 𝑒 𝑥 𝑖 + 𝜀 𝑖𝑗

Equation 4: 𝜎 𝜀 𝑖 𝑗 2
= ex p ⁡ ( 𝜏 0
+ 𝜏 𝑊𝑆
( 𝑝 ℎ 𝑦 𝑠 𝑖𝑐 𝑎𝑙 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑖 𝑡 𝑦 𝑖𝑗
− 𝑝 ℎ 𝑦 𝑠 𝑖𝑐 𝑎𝑙 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑖 𝑡 𝑦 𝑖 ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅
) +
𝜏 𝐵𝑆
( 𝑝 ℎ 𝑦 𝑠 𝑖𝑐 𝑎𝑙 ⁡ 𝑎𝑐 𝑡 𝑖𝑣𝑖 𝑡 𝑦 𝑖 ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅
) + 𝜏 𝑇 𝑡 𝑖𝑚𝑒 ⁡ 𝑜𝑓 ⁡ 𝑑 𝑎𝑦 𝑖𝑗
+ 𝜏 𝑊 𝑤𝑒 𝑒 𝑘𝑒 𝑛𝑑 𝑖𝑗
+ 𝜏 𝐷 𝑏 𝑢 𝑟𝑠 𝑡 𝑖𝑗
+ 𝜏 𝑆 𝑠 𝑒 𝑥 𝑖 + 𝜏 𝑙 𝜐 0 𝑖 + 𝜔 𝑖 )

Results
Data Availability
A total of 332 participants were consented into the study. After a trial period (i.e., testing
the smartphone EMA questions) 86 participants were removed from the study due to issues such

70

as low compliance, technical issues, or the participant withdrew themselves. A total of 246
smartwatches were mailed to participants. There were initially 5,569 days (M=22.92, SD=4.88
per participant) of physical activity data available among 243 participants across three months.
Days were removed from the analytic dataset for several reasons: 1) the time zone on the
smartwatch changed (i.e., participant changed time zones via travel or daylight savings) (n=158
days), 2) the smartwatch sampling unit exceeded the maximum possible for one day (i.e., 86,400
seconds) (n=21 days), 3) there was no smartwatch data (n=855 days), 4) data were missing for
smartwatch wear time (n=667 days), 5) smartwatch wear time exceeded 24 hours (n=2 days), 7)
smartwatch wear time was less than two hours (n=177 days), and 8) only one day of physical
activity was available for the participant (n=2 days). After removing data there were 4,569 days
(82% of total data; M=19.61, SD=5.70 per participant) among 233 (96% of total sample)
participants.
For the EMA data, there were originally 71,050 EMA observations among 246
participants. EMA data were then matched to available day-level physical activity by date. After
merging the two data types, there were 56,567 EMA observations with matching day-level PA
data among 233 participants. Two participants were then excluded for having missing data on the
level-2 (i.e., person-level) covariate of sex at birth, resulting in 56,143 EMA observations among
231 participants. EMA observations were then further removed if the participant’s standard
deviation for an affect variable during the 3-month period was less than 0.2 (n=534 for positive-
activated, n=724 for positive-deactivated, n=2,891 for negative-activated, n=1,582 for negative-
deactivated) or if there was missing data on the outcome variable (n=14,063 for positive-
activated, n=14,311 for positive-deactivated, n=13,715 for negative-activated, n=13,624 for
negative-deactivated). The final four MELS models in MixWILD ranged from 39,537-41,546

71

EMA observations between 218-228 participants. The analytic sample did not significantly differ
by demographic characteristics from the entire study sample. The likelihood of answering an
EMA survey prompt was unrelated to the day of week (i.e., weekend day vs. weekday), age, or
sex at birth. An EMA survey prompt was less likely to be answered in the morning compared to
the evening (p<.05) and more likely to answered in the afternoon compared to the evening
(p<.05).
Participant Characteristics
Descriptive statistics for the analytical sample are shown in Table 8 (N=231). On
average, participants were 23.58 (SD=3.02) years old at baseline. About 55% of participants
reported female for sex at birth and 92% of participants were cisgender. Less than 50% of the
sample identified as White, and 29% indicated that they were Hispanic/Latino/a/Spanish. The
average number of EMA observations per participant in the final model was 182 (range 12-331;
SD=69) for positive-activated affect, 180 (range 10-328; SD=70) for positive-deactivated affect,
181 (range 10-328; SD=68) for negative-activated affect, and 183 (range 12-331; SD=70) for
negative-deactivated affect. The mean positive-activated score across all observations in the final
model was 2.77 (SD=0.97), the mean positive-deactivated score was 3.02 (SD=1.07), the mean
negative-activated score was 1.92 (SD=1.00), and the mean negative-deactivated score was 1.96
(SD=0.85).
Table 8. Participant Characteristics (N=231)
Demographics
1
n (%)
Age in years (M ± SD)

23.58 + 3.20
Sex

Female 127 (55.0)
Male 104 (45.0)
Transgender Identity

Yes, transgender, male-to-female 1 (0.4)
Yes, transgender, female-to-male 3 (1.3)
Yes, transgender, gender nonconforming 11 (4.8)

72

No 213 (92.2)
Don’t know/not sure 3 (1.3)
Hispanic, Latino/a, or Spanish Origin
Yes 68 (29.4)
No 163 (70.6)
Race
2,a

American Indian or Alaska Native 11 (4.8)
Asian Indian 21 (9.1)
Black or African American 30 (13.0)
Chinese 28 (12.1)
Filipino 12 (5.2)
Guamanian or Chamorro 1 (0.4)
Japanese 5 (2.2)
Korean 9 (3.9)
Native Hawaiian/Other Pacific Islander 0 (0.0)
Other Asian 13 (5.6)
Other Pacific Islander 4 (1.7)
Samoan 0 (0.0)
Vietnamese 9 (3.9)
White 114 (49.4)
Education

Never attended school or only attended
kindergarten
0 (0.0)
Grades 1-8 0 (0.0)
Grades 9-11 1 (0.4)
Grade 12 or GED (high school graduate) 24 (10.4)
Some college or technical school 95 (41.1)
College graduate 111 (48.1)
Work Status
2

Employed for wages 128 (55.4)
Self-employed 12 (5.2)
Out of work for 1 year or more 7 (3.0)
Out of work for less than 1 year 27 (11.7)
Homemaker 5 (2.2)
Student 119 (51.5)
Retired 0 (0.0)
Unable to work 6 (2.6)
a
data missing for 10 participants
b
data missing for 1 participant
1
participants were able to skip/not answer any of questions
2
participants were able to select all that apply

Day-level associations of positive-activated affective variability and physical activity
The results for the positive-activated affect model are in Table 9. The within-subject
effect of physical activity was associated with mean positive-activated affect (β=0.02, p<.001),
indicating that if a participant engaged in more physical activity on any given day compared to

73

the participant’s average levels, they had higher mean positive-activated affect on the same day.
After controlling for the effects of mean positive-activated affect, the within-subject variance in
positive-activated affect was associated with physical activity at the day-level (τ=0.01, p<.001),
suggesting that on days when a participant had greater variability in their positive-activated
affect, they engaged in more physical activity compared to usual. The between-subject effect of
physical activity was not associated with mean positive-activated affect (β=0.02, p>.05) nor
within-subject variance in positive-activated affect (τ=0.01, p>.05).
Variability in feeling happy and energetic was greater on weekend days compared to
weekdays (τ=0.03, p<.05). There was a significant random scale standard deviation, indicating
that participants differed from each other in their degree of within-subject variance in positive-
activated affect (scale standard deviation=0.60, p<.001). Within-subject variance and location
intercept were not associated with each other (estimate=-0.02, p>.05), indicating that mean levels
of positive-activated affect were not associated with variability (i.e., stability or consistency) in
positive-activated affect.
Day-level associations of positive-deactivated affective variability and physical activity
The results for the positive-deactivated affect model are shown in Table 10. The within-
subject effect of physical activity was associated with mean positive-deactivated affect (β=-
0.005, p<.001), indicating that if on any given day, a participant engaged in more physical
activity compared to the participant’s average, they would have lower mean positive-deactivated
affect on the same day. After controlling for the effects of mean positive-deactivated affect,
within-subject variance in positive-deactivated affect was not associated with physical activity at
the day level (τ=0.002, p>.05). The between-subject effect of physical activity was not
significantly associated with mean positive-deactivated affect (p>.05). However, within-subject

74

variance in positive-deactivated affect was associated with between-subject physical activity
(τ=0.04, p<.05), indicating that participants who exhibited more variability in positive-
deactivated affect engaged in more physical activity compared to other participants.
There was less variability in positive-deactivated affect in the afternoons compared to
evenings (τ=-0.05, p<.05). There was a significant random scale standard deviation, suggesting
that participants differed from each other in their degree of within-subject variance in positive-
deactivated affect (scale standard deviation=0.63, p<.001).
Day-level associations of negative-activated affective variability and physical activity
The results for the negative-activated affect model are shown in Table 11. The within-
subject effect of physical activity was inversely associated with mean negative-activated affect
(β=-0.003, p<.05), indicating that if a participant engaged in more physical activity compared to
the participant’s average, they would have lower mean negative-activated affect on the same day.
After controlling for the effects of mean negative-activated affect, the within-subject variance in
negative-activated affect was not associated with physical activity at the day-level (τ=-0.005,
p>.05). In addition, both mean negative-activated affect (β=-0.002, p>.05) and within-subject
variance in negative-activated affect (τ=0.02, p>.05) were not associated with between-subject
level of physical activity.
Variability in feeling tense and stressed was lower on weekend days compared to
weekdays (τ=-0.14, p<.001). Within-subject variance in negative-activated affect was lower in
the morning compared to evening (τ=-0.04, p<.05). There was a significant random scale
standard deviation, indicating that participants differed from each other in their degree of within-
subject variance in negative-activated affect (scale standard deviation=0.66, p<.001). Within-
subject variance and location intercept were positively associated with each other

75

(estimate=0.40, p<.001), indicating that participants with higher negative-activated affect means
exhibited more within-subject variability (i.e., less consistency) in their negative-activated affect.
Day-level associations of negative-deactivated affective variability and physical activity
The results for the negative-deactivated affect model are shown in Table 12. The within-
subject effect of physical activity was inversely associated with mean negative-deactivated affect
(β=-0.01, p<.001), indicating that if a participant engaged in more physical activity on any given
day compared to the participant’s average, they would have lower mean negative-deactivated
affect on the same day. After controlling for the effects of mean negative-deactivated affect, the
within-subject variance in negative-deactivated affect was inversely associated with physical
activity at the day-level (τ=-0.01, p<.001). These findings suggest that on days when a
participant had greater variability in negative-deactivated affect, they engaged in less physical
activity compared to usual. However, both mean negative-deactivated affect (β=-0.01, p>.05)
and within-subject variance in negative-deactivated affect (τ=-0.004, p>.05) were not associated
with between-subject level of physical activity.
Within-subject variance in negative-deactivated affect was not associated with weekend
days (p>.05). Variability in feeling sad and fatigued was lower during afternoons compared to
evenings (τ=-0.07, p<.05). There was a significant random scale standard deviation, suggesting
participants differed from each other in their degree of within-subject variance in negative-
deactivated affect (scale standard deviation=0.58, p<.001). Within-subject variance and location
intercept were positively associated with each other (estimate=0.34, p<.001), indicating that
participants with higher mean levels of feeling sad and fatigued had more within-subject
variability (i.e., less consistency) in their feelings of sad and fatigued.

76

Table 9. Mixed-effects location scale models examining day-level associations of within-
subject variance in positive-activated affect with physical activity
Positive-activated Affect
(happy, energetic)
a

Estimate (SE) p
Mean Model (β)
Intercept 2.53 (0.19) <.001
Burst number

-0.0003 (0.002) .86
Time of day (morning
b
)

-0.06 (0.01) <.001
Time of day (afternoon
c
)

-0.01 (0.01) .20
Weekend day
d
0.06 (0.01) <.001
Wear time 0.0001 (0.00002) <.001
Sex at birth
e
-0.22 (0.09) .02
Between-subject physical activity 0.03 (0.02) .10
Within-subject physical activity 0.02 (0.001) <.001
Within-subject Variance Model (τ)
Intercept -1.20 (0.17) <.001
Burst number -0.03 (0.01) <.001
Time of day (morning
b
)

0.03 (0.02) .08
Time of day (afternoon
c
) -0.03 (.02) .10
Weekend day
d
0.03 (0.01) .03
Wear time 0.00004 (0.0004) .32
Sex at birth
e
0.15 (0.08) .07
Between-subject physical activity 0.01 (0.02) .40
Within-subject physical activity 0.01 (0.002) .01
Random scale standard deviation 0.60 (0.03) <.001
Random location effect on within-subject
variance
-0.02 (0.05) .55
a
N=228 participants and N=41,546 EMA observations;
b
Morning (12 A.M. – 12 P.M.) versus
evening (5 P.M. – 12 A.M.);
c
Afternoon (12 P.M. – 5 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Weekend day versus weekday;
e
Self-reported sex at birth (1=female, 0=male)

77

Table 10. Mixed-effects location scale models examining day-level associations of within-
subject variance in positive-deactivated affect with physical activity
Positive-deactivated Affect
(relaxed)
a

Estimate (SE) p
Mean Model (β)
Intercept 3.19 (0.19) <.001
Burst number

0.03 (0.002) <.001
Time of day (morning
b
)

-0.06 (0.01) <.001
Time of day (afternoon
c
)

-0.07 (0.01) <.001
Weekend day
d
0.10 (0.01) <.001
Wear time -0.00002 (0.00002) . 34
Sex at birth
e
-0.16 (0.09) .08
Between-subject physical activity -0.01 (0.02) .62
Within-subject physical activity -0.005 (0.001) <.001
Within-subject Variance Model (τ)
Intercept -0.90 (0.18) <.001
Burst number -0.05 (0.004) <.001
Time of day (morning
b
)

-0.01 (0.02) .53
Time of day (afternoon
c
) -0.05 (0.02) .002
Weekend day
d
-0.02 (0.01) .18
Wear time 0.0002 (0.0004) <.001
Sex at birth
e
0.02 (0.09) .80
Between-subject physical activity 0.04 (0.02) .03
Within-subject physical activity 0.002 (0.003) .40
Random scale standard deviation 0.63 (0.03) <.001
Random location effect on within-subject
variance
0.004 (0.04) .92
a
N=228 participants and N=41,108 EMA observations;
b
Morning (12 A.M. – 12 P.M.) versus
evening (5 P.M. – 12 A.M.);
c
Afternoon (12 P.M. – 5 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Weekend day versus weekday;
e
Self-reported sex at birth (1=female, 0=male)

78

Table 11. Mixed-effects location scale models examining day-level associations of within-
subject variance in negative-activated affect with physical activity.
Negative-activated Affect
(tense, stressed)
a

Estimate (SE) p
Mean Model (β)
Intercept 1.92 (0.22) <.001
Burst number

-0.001 (0.002) .70
Time of day (morning
b
)

0.01 (.01) .23
Time of day (afternoon
c
)

0.02 (0.01) <.001
Weekend day
d
-0.07 (0.01) <.001
Wear time 0.0004 (0.00002) .01
Sex at birth
e
0.10 (0.11) .35
Between-subject physical activity -0.002 (0.02) .91
Within-subject physical activity -0.003 (0.001) .01
Within-subject Variance Model (τ)
Intercept -1.52 (0.22) <.001
Burst number -0.03 (0.005) <.001
Time of day (morning
b
)

-0.04 (0.02) .02
Time of day (afternoon
c
) 0.02 (0.02) .25
Weekend day
d
-0.14 (0.02) <.001
Wear time 0.0003 (0.0001) <.001
Sex at birth
e
0.19 (0.11) .08
Between-subject physical activity 0.02 (0.02) .24
Within-subject physical activity -0.005 (0.003) .09
Random scale standard deviation 0.66 (0.03) <.001
Random location effect on within-subject
variance
0.40 (0.05) <.001
a
N=218 participants and N=39,537 EMA observations;
b
Morning (12 A.M. – 12 P.M.) versus
evening (5 P.M. – 12 A.M.);
c
Afternoon (12 P.M. – 5 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Weekend day versus weekday;
e
Self-reported sex at birth (1=female, 0=male)

79

Table 12. Mixed-effects location scale models examining day-level associations of within-
subject variance in negative-deactivated affect with physical activity.
Negative-deactivated Affect
(sad, fatigued)
a

Estimate (SE) p
Mean Model (β)
Intercept 1.97 (0.17) <.001
Burst number

0.01 (0.002) <.001
Time of day (morning
c
)

-0.03 (0.01) <.001
Time of day (afternoon
d
)

-0.04 (0.01) <.001
Weekend day
e
-0.03 (0.01) <.001
Wear time 0.00003 (0.00001) .02
Sex at birth
f
0.08 (0.09) .35
Between-subject physical activity -0.01 (0.02) .66
Within-subject physical activity -0.01 (0.001) <.001
Within-subject Variance Model (τ)
Intercept -1.448 (0.19) <.001
Burst number -0.01 (0.005) .03
Time of day (morning
c
)

-0.01 (0.01) .45
Time of day (afternoon
d
) -0.07 (0.02) <.001
Weekend day
e
-0.01 (0.01) .48
Wear time 0.0002 (0.0001) <.001
Sex at birth
f
0.26 (0.10) .01
Between-subject physical activity -0.004 (0.02) .80
Within-subject physical activity -0.01 (0.003) <.001
Random scale standard deviation 0.58 (0.03) <.001
Random location effect on within-subject
variance
0.34 (0.04) <.001
a
N=224 participants and N=40,937 EMA observations;
b
Morning (12 A.M. – 12 P.M.) versus
evening (5 P.M. – 12 A.M.);
c
Afternoon (12 P.M. – 5 P.M.) versus evening (5 P.M. – 12 A.M.);
d
Weekend day versus weekday;
e
Self-reported sex at birth (1=female, 0=male)

80

Discussion
The present study extends previous research by examining the associations between
affective variability and physical activity in naturalistic settings at finer-grained timescales (day-
level vs. subject-level). Results indicated that day-level variability in positive-activated affect
(i.e., feeling happy and energetic) was associated with more same-day physical activity, whereas
variability in negative-deactivated affect (i.e., feeling sad and fatigued) was associated with less
same-day physical activity. This study extends beyond previous research by informing our
understanding of the dynamics inherent to the relationship between affect and physical activity,
as well as gives insight into how these processes differ within individuals and unfold in the
context of daily life.
Based on previous empirical research (Dunton et al., 2014; Maher et al., 2019), it was
hypothesized that day-level variability in positive-activated affect and physical activity would be
inversely associated, such that greater variability would be associated with less physical activity.
One potential explanation for observing positive associations in the current study is perhaps
individuals had increased feelings of happy and energetic after engaging in physical activity
compared to before physical activity, resulting in substantial differences in ratings. Evidence
suggests physical activity improves subsequent positive affective states (Liao et al., 2015) and
research at the momentary level reported more activity being associated with feeling more
energetic in the subsequent 15 and 30 minutes (Liao et al., 2017). Therefore, days with more
physical activity may also have more higher reports of feeling energetic. While future research is
needed to further investigate the potential causal relationships, these findings suggest positive-
activated affect variability may not necessarily be disadvantageous to physical activity;
individuals may engage in physical activity to manage their fluctuations in their affect or change

81

their affective state. These findings may differ from prior studies—which included samples of
both children and adults (ages 8-73)—because our sample consisted of on only young adults
(ages 18-29); variability in momentary affective states and emotion regulation may differ across
the lifespan (Ebner & Fischer, 2014; Zimmermann & Iwanski, 2014). By contrast, variability in
positive-deactivated affect (i.e., feeling relaxed) was not associated with day-level physical
activity. Although this finding was in line with previous research that found no associations
between subject-level variability in positive affect (e.g., happy, calm, cheerful, joyful) and
physical activity (Maher et al., 2019), future research would benefit by differentiating both
valence and arousal, and by adding more than one item to assess positive-deactivated affect.
Variability in negative-deactivated affect (i.e., sad and fatigued), but not negative-
activated affect (i.e., tense and stressed), was associated with same-day physical activity. The
current study did not find significant associations between variability in negative-activated affect
and same-day physical activity. It is possible that findings were attenuated by floor effects; the
average negative-activated affect score was 1.92, meaning that participants who were lower than
average had relatively less room to vary downward than participants who are above the average.
On the other hand, findings regarding negative-deactivated affect were consistent with the
hypothesis—greater day-level variability in feeling sad and fatigued was associated with less
day-level physical activity. Regulating fluctuations in negative affect requires overcoming innate
tendencies to display feelings in response to stimuli and has been shown to diminish self-
regulatory resources (Baumeister et al., 1998; Hagger et al., 2010a). Experiencing fluctuations in
feeling sad and fatigued may deplete self-regulatory resources involved in physical activity (e.g.,
planning, coping with barriers), making it more difficult to engage in physical activity (Buckley
et al., 2014; Van Dyck et al., 2016). Another possible explanation for the observed inverse

82

associations between negative-deactivated affective variability and physical activity is that
greater physical activity may serve to stabilize negative affect. On days when individuals are
engaging in more physical activity compared to usual, their subsequent negative affect may be
more consistent. Some prior research suggests physical activity predicts lower levels of negative
affect (Do et al., 2021; Dunton et al., 2014; Emerson et al., 2018). Future research could extend
the current work by further elucidating these associations by examining the temporality of these
processes (e.g., does physical activity stabilize affect or do fluctuations in affect influence
physical activity) as well as the mechanisms linking variability and physical activity.
Overall, the significant positive-activated and negative-deactivated findings and the null
positive-deactivated and negative-activated findings suggest that the associations between day-
level affective variability and same-day physical activity may be rooted in the extent to which
affective states involves activation and arousal. Given that physical activity requires some form
of physical exertion and movement, it may be more influenced by or have more influence on
arousal components of affect versus valence components of affect. By separating valence and
arousal through the examination of four different affect constructs, we were able to uncover
associations that may otherwise go undetected if only positive affect and negative affect were
examined (Stevens et al., 2020). The current study documented associations between negative
affective variability and physical activity; previous studies have been limited by their ability to
test these associations due to insufficient intraindividual variability in negative affect.
The findings have several implications for theory, future research, and behavior change
interventions. Theoretical frameworks seeking to explain how affect influences health behaviors
(e.g., affect and health behavior framework, affective reflective theory) should consider the roles
of both intensity and variability in incidental affect (Brand & Ekkekakis, 2018; D. E. Conroy &

83

Berry, 2017; Williams et al., 2018; Williams & Evans, 2014). For example, usual (i.e., chronic)
and momentary (i.e., acute) levels of affective determinants are thought to predict behavior;
however, the degree of stability/instability of one’s affective states over time are not considered
in theoretical frameworks on affect and physical activity. Furthermore, the potential relationship
between mean levels and variability in affect should be considered. The current study found that
mean levels of negative-activated and negative-deactivated affect were positively associated with
variability (e.g., higher mean levels associated with more variability). In terms of future research,
additional research among different samples—such as older adults and those with affect-related
mental health conditions—is needed to substantiate the evidence for associations of variability in
positive-activated affect and negative-deactivated affect with daily physical activity. Future
research can also incorporate wearable devices to continuously monitor and potentially target
fluctuations in affect and activity throughout the day. Regarding behavior change interventions,
future work could explore the potential effects of strategies delivered real-time to target
variability and increase physical activity by helping individuals self-regulate or cope with
fluctuations in affect, such as prompts for emotion regulation exercises, mindfulness practices, or
brief coaching sessions (Nahum-Shani et al., 2018; Valle et al., 2020). The current study
highlights the importance of dynamic associations occurring at the day-level which support the
use of novel ecological interventions in individuals’ daily lives that adjust to a participant’s
momentary situations (e.g., just-in-time adaptive interventions; Nahum-Shani et al., 2018),
opposed to traditional behavior change interventions that may target subject-level characteristics,
such as dispositions to experience high negative or low positive affect.
The study had several strengths including modeling within-subject variance in affective
constructs, examining the day-level associations with monitor-based physical activity, coupling

84

smartphone and smartwatch technologies to continuously capture activity data, and the robust
measure of daily affective variability through hourly EMA surveys (vs. 1-5 prompts per day).
However, there were some limitations to note. Our analyses do not allow us to infer causality
between affective variability and day-level physical activity. Future studies may expand upon
this line of research by examining the temporal relationships of affective variability and physical
activity within the day by conducting EMA studies to determine whether variability over a
shorter time frame (e.g., across hours) predicts subsequent physical activity (e.g., in the next
hour) and vice versa. Another study limitation is that data collection occurred during the
COVID-19 pandemic (beginning in March 2020), which may have drastically changed
individuals’ day-to-day lives. Varying restrictions and closures across counties may have
influenced physical activity and affective states differently; therefore, results may not be
generalizable to non-pandemic times (Courtney et al., 2021; Hamidi & Zandiatashbar, 2021).
Furthermore, although the study was able to distinguish between valence and arousal in the
conceptualization of affect, the study was limited by utilizing only a few items to assess affect. In
particular, only one item (“relaxed”) was used to assess positive-deactivated affect. This narrow
assessment of affective states may not fully capture variability within constructs. Additionally,
although each MELS model had over 40,000 observations, the analyses excluded any EMA
observations with missing data on the analytic variables. While MixWILD has strengths such as
allowing covariates to influence within-subject variance and allowing subjects to contribute
varying numbers of observations, the software does not allow any missing data on analytic
variables. It is possible that individuals who experience more fluctuations in their affect, are less
likely to respond to EMA prompts.

85

Conclusions
Results indicated young adults were more active on days when they experienced greater
variability in positive-activated affect, suggesting that moment-to-moment fluctuations in
positive-activated affect may boost or be the result of physical activity. On the other hand, young
adults were less active on days with greater variability in negative-deactivated affect;
fluctuations in negative-deactivated affect may require use of self-regulatory resources, leaving
little energy for carrying out effortful physical activity plans and behaviors. Overall, this study
underscores how variability in different affective states may differentially be associated with
daily physical activity, as well as the importance of moving beyond assessing only mean or
momentary levels of affect. Understanding the extent to which affect fluctuates throughout the
day may help us better understand how affect and activity relate and unfold in the context of
daily life. Study findings can strengthen the development of interventions seeking to increase
regular physical activity by considering an individual’s fluctuations in affective states across the
day.

86

Chapter 4: Assessing whether subject-level associations of momentary affect and
subsequent physical activity predict future physical activity levels
Abstract
Background: Despite the growing literature base on the momentary associations of affect and
physical activity, it is unclear if and how these momentary associations differ between
individuals. There may be considerable differences (between-subject heterogeneity) in the extent
to which physical activity levels are contingent upon momentary affect, such that the degree of
strength between momentary affect and physical activity may vary across people. This affect-
physical activity contingency may be predictive of future physical activity levels and could
provide useful information regarding physical activity habit development. The current study used
a novel two-stage statistical modeling strategy capable of testing whether subject-level slopes
(i.e., the association between two momentary levels) predicts a subject-level outcome. The
objective of the study was to examine whether the subject-level associations between momentary
affect and subsequent physical activity (i.e., 30 minutes later) predict future physical activity
levels.
Methods: Young adults (N=174, Mage=23.45, SD=3.18) self-reported their current positive-
activated (happy, energetic), positive-deactivated (relaxed), negative-activated (tense, stressed),
and negative-deactivated (sad, fatigued) affect through hourly ecological momentary assessment
(EMA) surveys on their personal smartphone for four measurement bursts (i.e., each lasting four
consecutive days) spanning across two months (baseline). Objective physical activity was
collected via smartwatches in the 30 minutes following the EMA surveys. Future average daily
physical activity one month after baseline was also measured through smartwatches. Two stage
mixed-effects location scale modeling tested whether the strength between momentary affect and

87

subsequent physical activity during baseline was associated with average daily physical activity
levels measured one month later.
Results: A total of 18,209 EMA observations were included in the final analyses. Results
indicate that while was subject-level heterogeneity the association between momentary affect
(i.e., positive-activated, positive-deactivated, negative-activated, negative-deactivated affect) and
physical activity in the next 30 minutes (ps<.05). However, the strength of the association
between momentary affect and subsequent physical activity did not predict young adults’ future
daily physical activity levels one month later (ps>.05).
Conclusions: While there was subject-level heterogeneity in the momentary affect contingency
of physical activity, these associations did not predict future physical activity levels one month
later. Future intensive longitudinal research is warranted to further explore these potential
associations among young adults, individuals with different mental health conditions, and as well
as with future physical activity assessed at different timescales. Overall, the study demonstrates
the application of a novel two-stage modeling approach to understand how subject-level
predictors, such as random subject slope, may explain differences in subject-level outcomes.

88

Introduction
Despite the well-known health benefits of physical activity—such as reducing the risk for
obesity, type 2 diabetes, some cancers, overall mortality, and other chronic conditions—overall
physical activity levels among adults are low (Jakicic et al., 2019; Leitzmann et al., 2007; Nocon
et al., 2008; Warburton & Bredin, 2017). About only half of adults in the United States (U.S.)
met the physical activity guidelines for aerobic (2018 Physical Activity Guidelines Advisory
Committee Scientific Report, 2018). Numerous health interventions have sought to increase
regular physical activity among adults (Michie et al., 2009; Mönninghoff et al., 2021); however,
people tend to return to their activity levels at baseline and do not maintain their new behaviors
long-term (Rhodes & Dickau, 2012; Wood & Neal, 2016). Elucidating the correlates of daily
physical activity are critical for developing efficacious strategies to promote the increased
physical activity engagement and habit formation.
Recently, researchers have considered affect as an important, putative determinant of
physical activity to better understand who will, or will not, engage in physical activity and under
what contexts in daily life. Empirical evidence suggests momentary levels of affect at any given
moment predict subsequent levels of physical activity within short time scales (e.g., minutes,
hours) (Kanning & Schoebi, 2016; Liao et al., 2015; Stevens et al., 2020). A systematic review
on acute associations between affective estates and physical activity in naturalistic settings found
that elevated positive affect was associated with greater physical activity over the next few
hours, whereas there was inconsistent evidence for the role of negative affect as a momentary
antecedent for physical activity (Liao et al., 2015).
Despite the growing literature base on the momentary associations of affect and physical
activity, it is unclear if and how these momentary associations differ between individuals. There

89

may be considerable differences in the extent to which physical activity levels are contingent
upon momentary affect. In other words, it is possible that there is between-subject heterogeneity
in the strength of associations between affect and physical activity, such that one individual’s
slope (e.g., the momentary association between affect and activity levels) may be greater than
another individual. For example, some individuals may engage in physical activity independent
of their recent and current feelings, whereas others may engage (or not engage) in any physical
activity depending on their current affective states. The degree of strength between momentary
affect and physical activity may vary across people (i.e., viewed as random effects in statistical
models). Although knowing whether one’s physical activity is contingent on their momentary
affect may be useful for the development of just-in-time adaptive interventions that seek to
promote behavior change in the moment (Nahum-Shani et al., 2018), understanding to extent to
which the affect-physical activity contingency is predictive of future physical activity levels may
provide more information in regards to physical activity habit development. A stronger
association between momentary and subsequent physical activity may reflect weaker physical
activity habits, which are typically formed through the presence of stable context cues (e.g.,
location, time) (Wood & Neal, 2016). Habit formation stems from the presence of stable context
cutes, such as like time of day, locations, people, or prior actions in a sequence. For example, a
study demonstrated that 90% of regular exercises had a cue (i.e., location or time) to exercise,
and exercising was shown to be more automatic among those who had a cue by a specific
location (Tappe et al., 2013). Therefore, individuals may be more likely to engage in physical
activity—and thus have stronger physical activity habits—if they have developed contextual
cues compared to individuals that have activity contingent on momentary affect. Understanding
how the association between affect and physical activity predicts future physical activity can

90

inform new theories and interventions by providing information on whether between-subject
heterogeneity in the momentary affect-physical activity association should be considered.
However, answering these unique research questions has been previously limited by
available statistical models and software. The recent development of a two-stage modeling
approach combining mixed-effects models (Stage 1) with a linear regression model (Stage 2)
using the MixWILD program allows for the investigation of whether subject-level slopes (i.e.,
random within-subject association between time-varying predictors) are associated with subject-
level outcomes. MixWILD (Mixed model analysis With Intensive Longitudinal Data;
https://reach-lab.github.io/MixWildGUI/), an open source statistical program that can answer unique
research questions regarding time-varying variables often collected through intensive
longitudinal data (Dzubur et al., 2020). Unlike traditional multilevel models that assume the
variances and covariances of the random effects are homogeneous across subjects, MixWILD
allowed subject-level variances (e.g., intercept, slope) of time-varying variables (e.g., affect,
physical activity) to be modeled as covariates to predict future physical activity. In addition,
MixWILD recognizes that subjects differ in the number of observations contributed to the
analyses so it can provide unbiased estimates with varying degrees of precision.
By utilizing a two-stage modeling approach, the objective of the study was to examine
whether the subject-level association between momentary affect (i.e., positive-activated,
positive-deactivated, negative-activated, negative-deactivated) and subsequent physical activity
predicts future physical activity levels (Aim 3). It was hypothesized that individuals with a
stronger subject-level association between momentary affect and physical activity levels (i.e., in
the subsequent 30 minutes) would be less active one month later; a stronger affect-activity
contingency may represent weaker physical activity habits and thus less physical activity. The

91

current study utilizes ecological momentary assessment (EMA) to gather data on momentary
affect from individuals in their natural environments (Shiffman et al., 2008). The collection of
states and experiences as they occur across the day in naturalistic settings can ecological validity
and help provide information for intervention development that is more generalizable to real-
world contexts and settings (Stone et al., 2007).
Methods
Study Design
The current study conducted a secondary data analysis of intensive longitudinal data
collected from the Temporal Influences on Movement and Exercise (TIME) study (Wang et al.,
2022). The TIME study, a prospective within-subject observational study, continuously collected
data across 12 months through smartphones and smartwatches. During the 12-month period,
participants completed 4-day measurements bursts every two weeks (i.e., up to 26 bursts total for
a maximum of 104 days). During each burst participants completed signal-contingent (i.e.,
randomly prompted) smartphone EMA. To answer the research questions of the current study,
data from the first two months of the study (i.e.. bursts 2-5) was used to estimate the association
between momentary affect and subsequent physical activity; this period is referred to as
“baseline”. Physical activity data following the seventh burst was used assess future physical
activity (i.e., one month later); this period is referred to as “follow-up”. These time periods were
selected in order to maximize available data because attrition and missing data increased the
longer individuals participated in the study. This study was conducted in accordance with the
Declaration of Helsinki and all aspects of the study were approved by the Institutional Review
Board at the University of Southern California (HS-18-00605).

92

Recruitment and Participants
The TIME Study recruited young adults (ages 18-29 years) living in the U.S. on a rolling
basis from March 2020 – August 2021. Due to the Coronavirus (COVID-19) pandemic, all
recruitment and study procedures were conducted remotely. Participants were recruited through:
1) individuals enrolled in the Happiness & Health Study (a University of Southern California
prospective cohort study of ninth-grade students that began in 2013) were sent emails to
addresses on file (Leventhal et al., 2015); 2) referrals from existing participants (i.e., word of
mouth); 3) online/social media advertisements (e.g., Twitter, Facebook, Instagram); 4) emails
were sent to addresses on file from other University of Southern California approved studies; and
5) emails were sent to potentially eligible participants through ResearchMatch (a national health
volunteer registry that was created by several academic institutions and supported by the U.S.
National Institutes of Health as part of the Clinical Translational Science Award program).
Inclusion criteria for the study were: 1) 18-29 years old living in the U.S.; 2) intend
within the next 12 months to engage in, or already engage in, recommended levels of physical
activity (≥150 min/week moderate or ≥75 min/week vigorous intensity);

3) use an Android-based
smartphone as their primary personal mobile device with no intention to switch to a non-Android
phone; 4) able to read and speak English; and 5) plan to reside in an area with Wi-Fi connectivity
during the study period. Exclusion criteria were: 1) physical or cognitive disabilities that prevent
study participation; 2) health issues that limit physical activity; 3) diagnosed sleep disorders; 4)
unable to wear a smartwatch or answer EMA surveys at home, work, school, or other location
where the participant spends a substantial amount of time (e.g., unable to answer prompts more
than 20% of the time); 5) spend more than 3 hours/day on a typical weekday or weekend day
driving; 6) own an Android phone version 6.0 (or older) or if the app will not function on the

93

phone due to other technical issues; 7) currently own and wearing a smartwatch; 8) have a pay-
as-you-go data plan or data plan with less than 2 GB of data; or 9) currently pregnant.
Study Procedures
Interested individuals completed an online interest form to determine initial eligibility.
Study staff contacted individuals to ask additional eligibility questions and provide more
information regarding the study. Eligible participants attended a video conference orientation
over Zoom with a study staff member to review all parts of the study and complete informed
consent. Participants received instructions on how to download the custom TIME smartphone
application on their personal phone, how to use the study app, and how to answer EMA surveys.
After the orientation, participants completed a baseline self-report electronic questionnaire on
Research Electronic Data Capture hosted at the University of Southern California (Harris et al.,
2009, 2019). Participants were mailed a study smartwatch and were trained on set-up and use
during a second video conference orientation. Participants were compensated for their study
participation based upon compliance; in total, participants received up to $80 per month ($960
total) for competing the EMA bursts and wearing the smartwatch.
The custom TIME app, developed for Android smartphones and smartwatches, collected
EMA data during the 12-month study period
(https://play.google.com/store/apps/details?id=mhealth.neu.edu.microT). Each EMA
measurement burst lasted for four consecutive days (i.e., Thursday-Sunday or Saturday-
Tuesday). During each burst, participants completed signal-contingent EMA surveys on their
smartphone about every hour during their waking hours. Participants inputted their sleep and
wake times in the app to ensure surveys were not prompted during sleep. A push notification
prompted participants to complete an EMA survey, which took 1-2 minutes to complete. If no

94

response was provided, two reminder prompts were sent in 5-minute intervals; the EMA survey
became inaccessible after 10 minutes. Participants were told to ignore prompts sent during
incompatible activities (e.g., driving).
Participants were loaned a Fossil Sport Gen 4 or Gen 5 smartwatch for the study.
Participants wore the smartwatch on one wrist of their choice consistently and continuously
during the study period except for one hour per day to charge the smartwatch and when the
watch would be expose to water for extended periods of time (e.g., bathing). The smartwatch
continuously recorded participants’ movement. More detailed information regarding the study
procedures are published elsewhere (Wang et al., 2022).
To estimate the random subject slope (i.e., associations between momentary affect and
physical activity in the 30 minutes following an EMA prompt) in Stage 1, data from bursts 2-5
were utilized. This was during the first months of the study; this period will be referred to as
“baseline”. Burst 1 was not included in the analysis because this burst was considered a trial and
some participants had not received their smartwatch yet. To determine whether random subject
slope predicted future levels of physical activity in Stage 2, the seven days following burst 7
were used to assess average daily physical activity. This period will be referred to as “follow-
up”. The follow-up period occurred 1 month after the baseline period.
Measures
Affective States
Select items were chosen to represent the two fundamental dimensions of affect posited
by the circumplex model (i.e., valence, arousal) (Russell, 1980). To reduce participant burden
(i.e., each EMA survey could be finished in under two minutes), limited items were asked to
measure positive-activated, positive-deactivated, negative-activated, and negative-deactivated

95

affect through the EMA surveys (Stevens et al., 2020). Participants were asked seven questions
in the following format: “Right now, how (affect term) do you feel?” Positive-activated affect
(happy, energetic), positive-deactivated affect (calm), negative-activated affect (tense, stressed),
and negative-deactivated affect (sad, fatigued) were assessed. The response options were on a
unipolar scale: 1 = “Not at all”; 2 = “A little”; 3 = “Moderately”; 4 = “Quite a bit”; 5 =
“Extremely”. Scores were averaged to create a composite score for each affect construct,
respectively.
Physical Activity
Objective physical activity data was assessed through the wrist-worn smartwatches. Tri-
axial raw acceleration along X, Y, and Z axes on the smartwatch was measured using the
embedded accelerometer at a sampling rate of 50Hz. Monitor Independent Movement Summary
(MIMS) units were computed offline using the raw accelerometry data to summarize movement
(John et al., 2019). MIMS-units allow for measuring the amount of movement, independent of
the type of sensor (e.g., smartwatch, smartphone), and is a standardized, nonproprietary metric
that can be applied across different accelerometers. MIMS for 1-second epochs were calculated
to determine the total motion at varied lengths of times longer than one second (e.g., minutes,
hours). Smartwatch non-wear was assessed through the sleep, wear, and non-wear (SWaN)
algorithm, which classifies the raw accelerometry data into sleep, wear, and non-wear classes for
a 30-second window (Arguello et al., 2018). Physical activity in the 30 minutes after an EMA
prompt during baseline (i.e., bursts 2-5) was the Stage 1 outcome. Physical activity was
operationalized as MIMS-units/smartwatch wear time. This variable was calculated by dividing
MIMS-units during the 30-minute window by smartwatch wear time during the 30-minutes
window. Consistent with current recommendations of 10 hours of wear time across waking hours

96

(i.e., two-thirds of 16 waking hours) for waist-worn accelerometers and prior studies examining
accelerometer windows following EMA prompts (Maher et al., 2017; Yang et al., 2020), the 30-
minute time window needed to have at least two-thirds of valid accelerometer wear (i.e.,
minimum of 20 minutes of valid wear) to be included in the analyses (Troiano et al., 2008). The
Stage 2 outcome, future physical activity at follow-up (i.e., one month later), was measured by
average daily total physical activity in the 7-day window after burst 7. To be included in the
analyses, the participant needed to have at least three days of data and each day needed at least
two hours wear time. This was decided up to help capture an accurate average physical activity
in the week frame and to ensure that the participant wore the smartwatch for at least some point
in the day. To account for smartwatch wear time, daily physical activity was divided by daily
smartwatch wear time. Future physical activity was operationalized as MIMS-units/smartwatch
wear time (hours). This variable was calculated by averaging each day’s MIMS-units/smartwatch
wear time (hours) for each participant. Both Stage 1 and Stage 2 outcomes were log-transformed
due to the data’s positive skew.
Participant Characteristics
Participants completed an electronic questionnaire at baseline, which assessed participant
demographic and characteristics. The following were self-reported: 1) age; 2) sex at birth
(female, male); 3) transgender identity (yes transgender male-to-female, yes transgender female-
to-male, yes transgender gender nonconforming, no, don’t know/not sure); 4) Hispanic, Latino/a,
or Spanish origin; 5) race (check all that apply: White, Black or African American, American
Indian or Alaska Native, Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, Other
Asian, Native Hawaiian, Guamanian or Chamorro, Samoan, Other Pacific Islander); 6)
educational attainment (never attended school or only attended kindergarten, grades 1 through 8

97

(elementary), grades 9 through 11 (some high school), grade 12 or GED (high school graduate),
college 1 year to 3 years (some college or technical school), college 4 years or more (college
graduate)); 7) work status (check all that apply: employed for wages, self-employed, out of work
for 1 year or more, out of work for less than 1 year, homemaker, student, retired, unable to
work); 8) marital status (married, divorced, widowed, separated, never married, member of
unmarried couple). Participants were able to skip/not answer questions if they preferred.
Statistical Analyses
A two-stage modeling approach using mixed-effects location scale modeling in
MixWILD (Mixed model analysis With Intensive Longitudinal Data; https://reach-
lab.github.io/MixWildGUI/) was conducted (Dzubur et al., 2020). In MixWILD, a mixed-effects
model in Stage 1 determined whether there are significant random subject intercept and slope in
physical activity in the 30 minutes following an EMA prompt during baseline. In Stage 2, a
single-level linear regression model predicted participants’ average daily physical activity at
follow-up using the random subject location effects (i.e., intercept and slope) derived from Stage
1 along with other covariates.
Equation 1: 𝑦 𝑖𝑗
= ( 𝛽 0
+ 𝜐 0
𝑖 ) + ( 𝛽 1
+ 𝜐 1
𝑖 ) 𝜒 𝑖𝑗
+ 𝛽𝜒
𝑖𝑗
′
+ 𝜀 𝑖𝑗

The mixed-effects mean model is shown in Equation 1. The time-varying measurement of
outcome y of subject i (i = 1, 2, …N) on occasion j (1, 2,..ni) will be predicted by: β0 is the
intercept coefficient, 𝜐 0
𝑖 the random subject intercept (location) effect (represents the subject’s
mean; accounts for the centering of repeating observations within subjects), β1 the slope
coefficient of the main predictor, 𝜐 1
𝑖 the random slope effect of the main predictor (the random
association for a participant beyond the average slope β1 of Xij predicting yij). Xij the main
predictor, β the corresponding regression coefficients for a vector of p regressors, 𝑋 𝑖𝑗
′
the

98

covariates (i.e., time of day, day of week, wear time), and 𝜖 𝑖𝑗
is the error term (the deviation of a
subject’s observation from the subject’s mean and the fixed part of the model). Here, 𝜐 0
𝑖 is the
random subject intercept (location) effect that represents each participant’s mean units of
physical activity in the 30 minutes following an EMA prompt when all other predictors are zero.
The random slope effect 𝜐 1
𝑖 represents the association of momentary affect on each participant’s
physical activity in the 30 minutes following an EMA prompt. The random subject effects (i.e.,
intercept and slope) account for the clustering of repeated observations within participants and
are also assumed to be normally distributed.
Equation 2: 𝜎 𝜀 𝑖𝑗
2
= exp ⁡( 𝒘 𝒊𝒋
′
𝜏 + 𝝊 𝒊 ′
𝝉 𝝊 + 𝜔 𝑖 )
In the within-subject variance sub-model (Equation 2), 𝜎 𝜀 𝑖𝑗
2
refers to the within-subject
variance, or consistency/inconsistency within subjects (how data vary within subjects), 𝒘 𝒊𝒋
′
are
the regressors and τ is the corresponding regression coefficient vector that consists of l
regressors. ωi is the random scale effect that accounts for clustering of observations within
subjects and allows the within-subject variance to vary across subjects beyond the contribution
of covariates (Dzubur et al., 2020). τv is the coefficient and 𝝊 𝒊 ′
, is the location random effects,
which indicates that an association between the location and scale random effects can be induced
(Hedeker & Nordgren, 2013).
Equation 3: 𝑦 𝑖 ∗
= 𝛽 0
∗
+ 𝛽 1
∗
𝜐 0
𝑖 ̂ + 𝛽 2
∗
𝜐 1
𝑖 ̂ + 𝜒 𝑖 ∗
𝛽 ∗
+ 𝜀 𝑖 ∗

In the Stage 2 single-level linear regression model (Equation 3), the continuous subject-
level outcome, future physical activity, 𝑦 𝑖 ∗
was predicted by: the intercept coefficient 𝛽 0
∗
, the
estimated random location of the Stage 1 outcome 𝜐 0
𝑖 ̂ and its corresponding coefficient 𝛽 1
∗
, the
estimated random slope of the Stage 1 outcome 𝜐 1
𝑖 ̂ and its corresponding coefficient 𝛽 2
∗
, the β
*

99

coefficients of other subject-level regressors (i.e., baseline physical activity, baseline age, sex)
𝑋 𝑖 ∗
, and the error term 𝜖 𝑖 ∗
.
Resampling of the subject-level random effects (i.e., location and slope) was conducted
in Stage 2 to account for the fact that random effects are estimated quantities (Carsey & Harden,
2013). Random effects were resampled multiple times from a normal distribution using the mean
and variance estimates of these random effects for each subject. The estimated random effects
from Stage 1 (i.e., random location and slope effects) were resampled 500 times in Stage 2.
Results
Data Availability
For the Stage 1 analyses, EMA and physical activity data from the first two months of the
study (i.e., “baseline”; bursts 2-5), was used. There 49,151 EMA observations. Of these EMA
observations, 32,709 were completed (66.55%). After removing observations where there was less
than 20 minutes of smartwatch wear time in the subsequent 30-minute window, there were 21,206
observations (64.83%) among 224 participants.
For the Stage 2 analyses, daily physical activity at follow-up (i.e., week following burst 7)
was used. There were 1,309 days of physical activity data among 205 participants (range 1-7 days
per participant). After removing days with less than two hours of smartwatch wear time, there were
1,252 days of physical activity data. After removing days when the smartwatch sampling units
exceed the maximum possible for one day (i.e., 86,400 seconds), there were 1,217 days of physical
activity data. Participants were only included in the analyses if they had a least three days of
physical activity data; this resulted in 1,206 days among 181 participants. Average physical
activity (i.e., MIMS-units/smartwatch wear time in hours) was then calculated for each participant.

100

Data for Stage 1 and Stage 2 were merged to ensure that the same participants were in both
datasets; this resulted in a total of N=174 participants in the final analyses with 18,210 EMA
observations.
Descriptive Statistics
A total of N=174 participants and N=18,209 EMA observations were included in the
final analyses (Table 13). About 52% of the participants reported female as their sex, and the
average age of participants at baseline is 23.45 years (SD=3.18). Participant characteristics for
the analytic sample are shown in Table 8. Participants contributed an average of 104 EMA
observations in the Stage 1 analyses (SD=36, range 15-185). On average, participants had 30
minutes of smartwatch wear time in the 30 minutes following an EMA prompt (range 20-30).
The final analytic sample did not significantly differ in demographic characteristics compared to
the study sample.
Table 13. Participant Characteristics (N=174)
Demographics
1
n (%)
Age in years (M ± SD)

23.45 + 3.18
Sex

Female 92 (52.0)
Male 82 (47.1)
Transgender Identity

Yes, transgender, male-to-female 1 (0.6)
Yes, transgender, female-to-male 1 (0.6)
Yes, transgender, gender nonconforming 8 (4.6)
No 162 (93.1)
Don’t know/not sure 2 (1.1)
Hispanic, Latino/a, or Spanish Origin
Yes 54 (31.0)
No 120 (69.0
Race
2,a

American Indian or Alaska Native 8 (4.6)
Asian Indian 15 (8.6)
Black or African American 21 (12.1)
Chinese 23 (13.2)
Filipino 10 (5.7)
Guamanian or Chamorro 1 (0.6)
Japanese 2 (1.1)

101

Korean 6 (3.4)
Native Hawaiian/Other Pacific Islander 0 (0.0)
Other Asian 9 (5.2)
Other Pacific Islander 3 (1.7)
Samoan 0 (0.0)
Vietnamese 7 (4.0)
White 86 (49.4)
Education

Never attended school or only attended kindergarten 0 (0.0)
Grades 1-8 0 (0.0)
Grades 9-11 1 (0.6)
Grade 12 or GED (high school graduate) 17 (9.8)
Some college or technical school 76 (43.7)
College graduate 80 (46.0)
Work Status
2

Employed for wages 95 (54.6)
Self-employed 8 (4.6)
Out of work for 1 year or more 4 (2.3)
Out of work for less than 1 year 22 (12.6)
Homemaker 4 (2.3)
Student 95 (54.6)
Retired 0 (0.0)
Unable to work 5 (2.9)
a
data missing for 8 participants
1
participants were able to skip/not answer any of questions
2
participants were able to select all that apply

Associations between momentary positive-activated affect and subsequent physical activity
predicting future physical activity
The Stage 1 results for positive-activated affect (i.e., feeling good and energetic) are
shown in Table 14. After controlling for weekend day, time of day, and burst number, greater
momentary positive-activated affect (i.e., feeling happy and energetic) predicted more physical
activity in the next 30 minutes at baseline (β=0.06, p<.001). The random location effects section
showed that there was significant random subject intercept in participants’ physical activity in
the 30 minutes following an EMA prompt (estimate=0.02, p<.001), suggesting that participants
differed in their mean physical activity when positive-activated affect was zero. The random
subject slope in participants’ physical activity in the 30 minutes following an EMA prompt was

102

also significant (estimate=0.003, p<.001), indicating that participants differed in their strength of
the within-subject association between positive-activated affect and physical activity. There was
no significant association between the two random subject location effects (i.e., random intercept
and random slope) (estimate=0.001, p=.05). This suggested that at any given EMA prompt, the
strength of a participants’ within-subject association between positive-activated affect and
physical activity in the 30 minutes following an EMA prompt was not attenuated by their mean
levels of positive-activated affect.
The results of the Stage 2 linear regression predicting participants’ average daily physical
activity (MIMS-units/smartwatch wear time in hours) 1 month later (follow-up) as a function of
the random subject intercept and random subject slope estimates generated from the Stage 1
model are shown in Table 914 After controlling for age and sex at birth, the random subject
intercept significantly predicted participants’ average daily physical activity at follow-up
(estimate=0.07, p<.001). This result suggests that participants who had higher mean physical
activity in the 30 minutes following an EMA prompt (baseline) compared to others engaged in
more daily physical activity one month later (follow-up). The random subject slope effect did not
significantly predict participants’ physical activity at follow-up, suggesting that the random slope
of positive-activated affect predicting subsequent physical activity was not related to future
physical activity levels (estimate=-0.002, p=.77). Age and sex were not significantly associated
with average daily physical activity at follow-up (ps>.05).
Associations between momentary positive-deactivated affect and subsequent physical
activity predicting future physical activity
The Stage 1 results for positive-deactivated affect (i.e., feeling relaxed) are shown in
Table 15. Greater momentary positive-deactivated affect predicted less physical activity in the

103

next 30 minutes at baseline, after controlling for weekend day, time of day, and burst number
(β=-0.03, p<.001). The random location effects section showed that there was a significant
random subject intercept in participants’ physical activity in the 30 minutes following an EMA
prompt (estimate=0.02, p<.001), suggesting that participants differed in their mean physical
activity when positive-deactivated affect was zero. The random subject slope in participants’
physical activity in the 30 minutes following an EMA prompt was also significant
(estimate=0.002, p<.001), indicating that participants differed in their strength of the association
between positive-deactivated affect and physical activity at baseline. The random intercept and
random slope were not significantly associated with each other (estimate=0.001, p=.15).
The results of the Stage 2 linear regression predicting participants’ average daily physical
activity 1 month later (follow-up) as a function of the random subject intercept and random
subject slope estimates generated from the Stage 1 model are shown in Table 15. After
controlling for age and sex at birth, the random subject intercept significantly predicted
participants’ average daily physical activity at follow-up (estimate=0.08, p<.001). This result
suggests that participants who had higher mean physical activity in the 30 minutes following an
EMA prompt (baseline) compared to others engaged in more daily physical activity one month
later (follow-up). The random subject slope effect did not significantly predict participants’
physical activity at follow-up (estimate=0.0004, p=.96), suggesting that the random slope of
positive-deactivated affect predicting subsequent physical activity was not related to future
physical activity levels. Age and sex were not significantly associated with average daily
physical activity at follow-up (ps>.05).

104

Associations between momentary negative-activated affect and subsequent physical activity
predicting future physical activity
The Stage 1 results for negative-activated affect (i.e., feeling stressed and tense) are
shown in Table 16. After controlling for weekend day, time of day, and burst number, greater
momentary negative-activated affect (i.e., feeling stressed and tense) predicted more physical
activity in the next 30 minutes at baseline (β=0.01, p<.05). The random location effects section
showed that there was significant random subject intercept in participants’ physical activity in
the 30 minutes following an EMA prompt (estimate=0.02, p<.001), suggesting that participants
differed in their mean physical activity when negative-activated affect was zero. The random
subject slope in participants’ physical activity in the 30 minutes following an EMA prompt was
also significant (estimate=0.001, p<.05), indicating that participants differed in their strength of
the within-subject association between negative-activated affect and physical activity. There was
no significant association between the two random subject location effects (i.e., random intercept
and random slope) (estimate=-0.001, p=.26). This suggested that at any given EMA prompt, the
strength of a participants’ within-subject association between negative-activated affect and
physical activity in the 30 minutes following an EMA prompt was not attenuated by their mean
levels of negative-activated affect.
The results of the Stage 2 linear regression predicting participants’ average daily physical
activity 1 month later (follow-up) as a function of the random subject intercept and random
subject slope estimates generated from the Stage 1 model are shown in Table 16. After
controlling for age and sex at birth, the random subject intercept significantly predicted
participants’ average daily physical activity at follow-up (estimate=0.08, p<.001). This result
suggests that participants who had higher mean physical activity in the 30 minutes following an

105

EMA prompt (baseline) compared to others engaged in more daily physical activity one month
later (follow-up). The random subject slope effect did not significantly predict participants’
physical activity at follow-up (estimate=-0.01, p=.43), suggesting that the random slope of
positive-activated affect predicting subsequent physical activity was not related to future
physical activity. Age and sex were not significantly associated with average daily physical
activity at follow-up (ps>.05).
Associations between momentary negative-deactivated affect and subsequent physical
activity predicting future physical activity
The Stage 1 results for negative-deactivated affect (i.e., feeling sad and fatigued) are
shown in Table 17. Greater momentary negative-deactivated affect predicted less physical
activity in the next 30 minutes at baseline, after controlling for weekend day, time of day, and
burst number (β=-0.03, p<.001). The random location effects section showed that there was a
significant random subject intercept in participants’ physical activity in the 30 minutes following
an EMA prompt (estimate=0.02, p<.001), suggesting that participants differed in their mean
physical activity when negative-deactivated affect was zero. The random subject slope in
participants’ physical activity in the 30 minutes following an EMA prompt was also significant
(estimate=0.002, p<.001), indicating that participants differed in their strength of the association
between negative-deactivated affect and physical activity at baseline. The random intercept and
random slope were not significantly associated with each other (estimate=0.001, p=.15).
The results of the Stage 2 linear regression predicting participants’ average daily physical
activity 1 month later (follow-up) as a function of the random subject intercept and random
subject slope estimates generated from the Stage 1 model are shown in Table 17. After
controlling for age and sex at birth, the random subject intercept significantly predicted

106

participants’ average daily physical activity at follow-up (estimate=0.08, p<.001). This result
suggests that participants who had higher mean physical activity in the 30 minutes following an
EMA prompt (baseline) compared to others engaged in more daily physical activity one month
later (follow-up). The random subject slope effect did not significantly predict participants’
physical activity at follow-up (estimate=0.0004, p=.96), suggesting that the random slope of
negative-deactivated affect predicting subsequent physical activity was not related to future
physical activity levels. Age and sex were not significantly associated with average daily
physical activity at follow-up (ps>.05).

107

Table 14. Results of Stage 1 mixed-effects model (positive-activated affect) and Stage 2
linear regression model
Physical Activity
Estimate (SE) p
Stage 1 Model
Regression coefficients
Intercept 0.94 (0.01) <.001
Positive-activated affect

0.06 (0.004) <.001
Time of day (morning
a
)

0.01 (0.01) .04
Time of day (afternoon
b
)

0.03 (0.01) <.001
Weekend day
c
0.002 (0.002) .38
Burst number
Random location effect variance and covariances
Random intercept 0.02 (0.002) <.001
Random slope 0.003 (0.0004) <.001
Covariance (random intercept and slope)

0.001 (0.001) .05
Stage 2 Model
Intercept 2.76 (0.05) <.001
Random intercept 0.07 (0.01) <.001
Random slope

-0.002 (0.01) .77
Age 0.001 (0.002) .82
Sex at birth
d
0.01 (0.01) .32
N of observations for Stage 1 model=18,210. N of subjects for Stage 2 model=174.
The dependent variable for the Stage 1 mixed-effects model is physical activity (log-transformed
MIMS-units/smartwatch wear time) in the 30-minute window following the EMA assessment of
affect during baseline (i.e., first two months of the study).
The dependent variable for the Stage 2 linear regression model is mean daily physical activity
(log-transformed MIMS-units/smartwatch wear time) during follow-up (i.e., one month later).
Positive-activated affect is mean centered.
a
Morning (12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
b
Afternoon (12 P.M. – 5
P.M.) versus evening (5 P.M. – 12 A.M.);
c
Weekend day versus weekday;
d
Self-reported sex at
birth (1=female, 0=male)

108

Table 15. Results of Stage 1 mixed-effects model (positive-deactivated affect) and Stage 2
linear regression model
Physical Activity
Estimate (SE) p
Stage 1 Model
Regression coefficients
Intercept 0.92 (0.02) <.001
Positive-deactivated affect

-0.03 (0.004) <.001
Time of day (morning
a
)

0.004 (0.01) .43
Time of day (afternoon
b
)

0.02 (0.01) <.001
Weekend day
c
-0.0001 (0.004) .98
Burst number 0.002 (0.002) .24
Random location effect variance and covariances
Random intercept 0.02 (0.002) <.001
Random slope 0.002 (0.004) <.001
Covariance (random intercept and slope)

0.001 (0.001) .15
Stage 2 Model
Intercept 2.77 (0.05) <.001
Random intercept 0.08 (0.01) <.001
Random slope

0.004 (0.01) .96
Age 0.0001 (0.002) .97
Sex at birth
d
0.01 (0.01) .29
N of observations for Stage 1 model=18,210. N of subjects for Stage 2 model=174.
The dependent variable for the Stage 1 mixed-effects model is physical activity (log-transformed
MIMS-units/smartwatch wear time) in the 30-minute window following the EMA assessment of
affect during baseline (i.e., first two months of the study).
The dependent variable for the Stage 2 linear regression model is mean daily physical activity
(log-transformed MIMS-units/smartwatch wear time) during follow-up (i.e., one month later).
Positive-deactivated affect is mean centered.
a
Morning (12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
b
Afternoon (12 P.M. – 5
P.M.) versus evening (5 P.M. – 12 A.M.);
c
Weekend day versus weekday;
d
Self-reported sex at
birth (1=female, 0=male)

109

Table 16. Results of Stage 1 mixed-effects model (negative-activated affect) and Stage 2
linear regression model
Physical Activity
Estimate (SE) p
Stage 1 Model
Regression coefficients
Intercept 0.93 (0.01) <.001
Negative-activated affect

0.01 (0.004) .01
Time of day (morning
a
)

0.01 (0.01) .17
Time of day (afternoon
b
)

0.02 (0.01) <.001
Weekend day
c
-0.003 (0.004) .57
Burst number 0.001 (0.002) .64
Random location effect variance and covariances
Random intercept 0.02 (0.002) <.001
Random slope 0.001 (0.0003) .002
Covariance (random intercept and slope)

-0.001 (0.001) .26
Stage 2 Model
Intercept 2.77 (0.05) <.001
Random intercept 0.08 (0.01) <.001
Random slope

-0.01 (0.01) .43
Age 0.0001 (0.002) .97
Sex at birth
d
0.01 (0.01) .44
N of observations for Stage 1 model=18,210. N of subjects for Stage 2 model=174.
The dependent variable for the Stage 1 mixed-effects model is physical activity (log-transformed
MIMS-units/smartwatch wear time) in the 30-minute window following the EMA assessment of
affect during baseline (i.e., first two months of the study).
The dependent variable for the Stage 2 linear regression model is mean daily physical activity
(log-transformed MIMS-units/smartwatch wear time) during follow-up (i.e., one month later).
Negative-activated affect is mean centered.
a
Morning (12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
b
Afternoon (12 P.M. – 5
P.M.) versus evening (5 P.M. – 12 A.M.);
c
Weekend day versus weekday;
d
Self-reported sex at
birth (1=female, 0=male)

110

Table 17. Results of Stage 1 mixed-effects model (negative-deactivated affect) and Stage 2
linear regression model
Physical Activity
Estimate (SE) p
Stage 1 Model
Regression coefficients
Intercept 0.93 (0.01) <.001
Negative-deactivated affect

-0.05 (0.01) <.001
Time of day (morning
a
)

0.01 (0.01) .27
Time of day (afternoon
b
)

0.02 (0.01) <.001
Weekend day
c
-0.004 (0.004) .30
Burst number 0.002 (0.002) .29
Random location effect variance and covariances
Random intercept 0.02 (0.002) <.001
Random slope 0.002 (0.001) <.001
Covariance (random intercept and slope)

0.001 (0.001) .38
Stage 2 Model
Intercept 2.75 (0.05) <.001
Random intercept 0.08 (0.01) <.001
Random slope

-0.01 (0.01) .20
Age 0.001 (0.002) .57
Sex at birth
d
0.01 (0.01) .58
N of observations for Stage 1 model=18,210. N of subjects for Stage 2 model=174.
The dependent variable for the Stage 1 mixed-effects model is physical activity (log-transformed
MIMS-units/smartwatch wear time) in the 30-minute window following the EMA assessment of
affect during baseline (i.e., first two months of the study).
The dependent variable for the Stage 2 linear regression model is mean daily physical activity
(log-transformed MIMS-units/smartwatch wear time) during follow-up (i.e., one month later).
Negative-deactivated affect is mean centered.
a
Morning (12 A.M. – 12 P.M.) versus evening (5 P.M. – 12 A.M.);
b
Afternoon (12 P.M. – 5
P.M.) versus evening (5 P.M. – 12 A.M.);
c
Weekend day versus weekday;
d
Self-reported sex at
birth (1=female, 0=male)

111

Discussion
The current study examined whether the momentary affect-physical activity relationship
was associated with future daily physical activity levels among young adults using smartphone
and smartwatches. The study demonstrates the application of a novel two-stage modeling
approach to understand how subject-level predictors, such as random subject slope, may explain
differences in subject-level outcomes. Results from the study suggest that while there was
subject-level heterogeneity the association between momentary affect (i.e., positive-activated,
positive-deactivated, negative-activated, negative-deactivated affect) and physical activity in the
next 30 minutes, the strength of the association between momentary affect and subsequent
physical activity did not predict young adults’ future daily physical activity levels one month
later.
The first stage in the novel two stage modeling approach examined the associations
between momentary affect and physical activity in subsequent 30-minute window, and whether
these associations differed between individuals by estimating random subject slopes. The
findings indicated that greater positive-activated affect (i.e., feeling happy and energetic) and
greater negative-activated affect (i.e., feeling stressed and tense) predicted more physical activity
in the next 30 minutes, whereas greater positive-deactivated affect (i.e., feeling relaxed) and
greater negative-deactivated affect (i.e., feeling sad and fatigued) predicted less subsequent
physical activity. These findings are partially in line with prior research; momentary positive
affect was associated with more subsequent physical activity (Emerson et al., 2018; Kanning &
Schoebi, 2016; Schwerdtfeger et al., 2010), whereas negative affect either predicted less physical
activity (Niermann et al., 2016) or had non-significant associations (Schwerdtfeger et al., 2010;
Wichers et al., 2012). Among studies that assessed valence and arousal separately, feeling calm

112

was associated with less subsequent physical activity (Kanning & Schoebi, 2016; Reichert et al.,
2016), which is line with the current study findings. Our findings indicate that negative-activated
affect predicted more subsequent physical activity, which could indicate that individuals use
activity as a coping mechanism for their feelings of stressed and tense. It is also possible that the
situations or contexts in which a person reports feeling more stressed or tense also involve more
bodily movement (e.g., rushing to work, errands), which was captured by the smartwatch as
MIMS-units. While the operationalization of physical activity through MIMS-units has benefits
beyond this research such as utilizing a standardized non-proprietary metric, MIMS-units capture
all movement and not just intentional physical activity. Additionally, it is possible that
individuals may be anticipating and dreading their planned, upcoming physical activity, resulting
in higher levels of momentary negative-activated affect. These findings add to the mixed
literature on whether pre-activity negative affect predicts subsequent physical activity.
Results from the first stage model also indicate that participants differed from each other
in the strength of their within-subject association between momentary affect and physical activity
in the following 30-minute window. These findings underscore the importance of considering the
between-subject heterogeneity of associations between two time-varying variables. These
differences characterized by idiographic, person-specific (i.e., random) effects are important to
consider in both health behavior theories and intervention development. Furthermore, future
research can elucidate whether person-level characteristics predict this between-subject
heterogeneity. For example, demographics (e.g., age, employment status/type, mental health and
well-being), emotion regulation, physical activity enjoyment, self-efficacy for physical activity,
affective attitudes towards activity, motivation, or habits may explain differences in the affect-

113

physical activity relationships (Anderson & Cychosz, 1994; Kendzierski & DeCarlo, 1991;
McAuley et al., 1989; Mulero-Portela et al., 2013; Verplanken & Orbell, 2003).
The main objective of the current study was to investigate whether the strength of
association between momentary affect and subsequent physical activity predicted future physical
activity levels. It was hypothesized that participants with a stronger within-subject association
between affect and physical activity would be less physically active at the one-month follow-up
point. The findings did not support the hypothesis: the strength of affect-physical activity
association did not significantly predict future physical activity levels. One potential explanation
for these results is that future physical activity (e.g., average daily physical activity one month
later) may more closely represent a change in physical activity in the analyses rather than future
physical activity because the model controls for baseline physical activity through the inclusion
of random subject intercept. The association between the random intercept and future physical
activity (the Stage 2 outcome) was substantial (i.e., 2.75 for positive-activated affect), suggesting
individuals who had higher mean levels of physical activity in the 30 minutes following an EMA
prompt compared to others also had more future physical activity; additionally, there may have
been little variation left in the outcome variable for the random slope to contribute to. By
including baseline physical activity in the model, the outcome of future physical activity may
represent a change in activity rather than solely future levels of activity. The follow-up timepoint
of one month later (i.e., burst 7) was selected a priori to maximize available data among
participants given attrition. Future longitudinal research examining physical activity further into
the future is warranted to better understand these potential associations.
Although the study hypotheses were not supported, the study has important implications
for future research. Thus far, this two stage modeling approach has only been applied to a study

114

assessing whether subject-level slopes between context (i.e., indoor versus outdoors) and
physical activity predicts subject-level physical activity six months later (Yang et al., 2021). The
current provides another empirical example for this modeling approach that extends beyond
traditional multi-level models my taking into consideration that each subject contributes a
different number of observations to the analysis. Future research study can apply this two-stage
modeling approach to test other research questions about whether the strength of the within-
subject association between two time-varying factors predicts a subject-level outcome. To better
understand the development of physical activity habits, additional research may explore whether
associations between time-varying factors (e.g., contextual factors, affective attitudes) and
subsequent physical activity is related to future physical activity. Understanding affective
determinants and the temporality of associations with physical activity an help strengthen
theoretical frameworks and further improve the development of intervention strategies (Stevens
et al., 2020; Williams et al., 2018) In regard to research on the affect-physical activity
relationship, future research should consider examining affect constructs that are distinguished
by their valence (pleasure-displeasure) and arousal (activation) by assessing positive-activated,
positive-deactivated, negative-activated, negative-deactivated affect rather than just positive and
negative affect (Russell, 1980). Not only is there heterogeneity in the within-subject associations
between affect and physical activity, but there is also heterogeneity in relationships between
affect and activity. The findings that people differ in their strength of association between affect
and subsequent physical activity may also have implications for future research and
interventions. Future work could determine the efficacy of tailored interventions, such as
whether targeting momentary affect to increase physical activity is more effective among certain
individuals (e.g., those with a stronger within-subject association). For those individuals whose

115

physical activity is contingent on their affect, mobile interventions could deliver content to help
individuals regulate their affect. Furthermore, tailored programming could help encourage the
development of physical activity habits by incorporating strategies such as setting goals and
plans, and methods to follow through with those plans even negative affect is high.
Future research could also expand upon this work by utilizing MixWILD to explore
whether variability in momentary physical activity is also associated with future physical
activity. MixWILD also has the capability to model random scale effects (i.e., within-subject
variability) of a Stage 1 time-varying variable, in addition to modeling random location effects
(i.e., random intercept and slopes). The current study did not estimate the within-subject variance
of physical activity in the 30-minute window following an EMA prompt in Stage 1 nor the effect
of the random scale of physical activity in the 30-minute window (i.e., the degree of within-
subject variability) on average daily physical activity in Stage 2 given that this was not part of
the a priori hypotheses. Future studies should explore the option of examining whether the
magnitude of subject-level variability in time-varying factors predicts an individual’s overall
levels of health behaviors or outcomes.
Limitations
Despite the strengths of the proposed study, such as the novel modeling approach applied
to test whether the strength of the within-subject association between two time-varying factors
(i.e., affect and physical activity) is associated with subject-level physical activity, there are
some limitations to note. First, only the 30-minute window following an EMA prompt was
analyzed for subsequent physical activity. This decision was made a priori based on previous
evidence indicated associations between affect and physical activity in the subsequent 30-
minutes window (Liao et al., 2015). However, assessing varying time windows (e.g., 15, 60, 120

116

minutes) can be important for investigating temporal relationships. Future studies could examine
varying time-windows to investigate if there are differences in the affect-physical activity
relationship. Second, the current study examined future physical activity one month later. As
mentioned previously, there may be associations between random subject slope and physical
activity further into the future (e.g., 3 or 6 months) that this study did not examine. One month
was selected to maximize available data at the subject-level. Future studies may benefit by
examining activity further into the future to be able to understand the potential long-term effects.
Another limitation is that data collection took place between March 2020 and August 2022, a
unique time in history where the COVID-19 pandemic may have influenced daily affective states
and physical activity (Courtney et al., 2021; Hamidi & Zandiatashbar, 2021). In addition,
participants resided in different jurisdictions across the U.S., with varying COVID-19 related
regulations and closures for common places for activity (e.g., fitness centers, parks). The study
findings are limited in their generalizability to populations that differ from the current study
sample such as children, older adults, individuals who are not currently physically active or
intend to be physically active (a study inclusion criteria), or individuals who are unable to
participate in year-long intensive study (e.g., individuals with occupations or responsibilities that
do not allow them to answer EMA prompts frequently throughout the day). Lastly, the strength
of the affect-physical activity association may differ among individuals with mental health
conditions such as major depressive disorder, generalized anxiety disorder, or other conditions
that are marked by affective states (Aldao et al., 2010; Marwaha et al., 2014). Future studies
could examine whether certain conditions or individual characteristics moderate the associations
between random subject slope in affect and physical activity with future overall physical activity
levels.

117

Conclusion
The current study employed a novel two-stage modeling approach to investigate whether
the strength of association between momentary affect and subsequent physical activity in the
next 30-minute window predicted future overall physical activity levels one month later. The
findings indicated that while there was subject-level heterogeneity in the momentary affect
contingency of physical activity, these associations did not predict future physical activity levels.
Future intensive longitudinal research is warranted to further explore these potential associations
among young adults, individuals with different mental health conditions, and as well as with
future physical activity assessed at different timescales. Overall, the study demonstrates the
importance of and ability to examine whether subject-level slopes predict subject-level
outcomes.

118

Chapter 5: Discussion

The overall goal of this dissertation was to investigate the associations between affective
variability (i.e., moment-to-moment fluctuations in affect) and physical activity among a sample
of young adults (ages 18-29 years old) using ecological momentary assessment (EMA) on
smartphones and smartwatches. The current dissertation addressed several methodological
limitations found in the existing literature, including the limited conceptualization of affective
states that typically combines both valence and arousal, the inability to examine associations of
affective variability and physical activity at finer-grained time-scales (i.e., day-level), the limited
ability to model within-subject variance (i.e., intraindividual variability) in both positive and
negative affective constructs, and the limited capacity to investigate whether subject-level slopes
in two time-varying constructs predicts subject-level outcomes. By collecting intensive
longitudinal data in naturalistic settings through smartphone and smartwatch technology and the
use of novel two-stage modeling approaches in MixWILD (Dzubur et al., 2020), the dissertation
addresses unique three research questions related to affective variability and physical activity.
The research contributes to the limited—yet growing—literature base on variability in affect and
physical activity.
Summary of Findings
The first study of the dissertation assessed whether subject-level variability in affect (i.e.,
positive-activated, positive-deactivated, negative-activated, negative-deactivated) was associated
with overall levels of physical activity. This study also explored a potential important moderator,
trait self-control, which was hypothesized to buffer the potential detrimental effects of affective
variability. Results indicate that subject-level variability in negative-deactivated affect (i.e.,
feeling sad and fatigued) was associated with greater levels of physical activity, whereas

119

variability in positive-activated, positive-deactivated, and negative-activated affect did not
predict physical activity. Also contrary to the hypothesis, trait self-control did not moderate
associations between subject-level variability in affect. Overall, the current study extends beyond
previous research in affect and physical activity by considering the way in which affect
fluctuates above or below the mean and calls for additional research at the momentary or day-
level.
The second study of the dissertation examined the associations between affective
variability and physical activity in naturalistic settings at finer-grained timescales (day-level vs.
subject-level). This study investigated day-level associations between affective variability (i.e.,
within-subject variance) and physical activity. It was hypothesized that greater day-level
affective variability (i.e., within-subject variance) would be associated with less physical activity
on that same day compared to usual. Specifically, it was hypothesized that individuals would
experience less affective variability (i.e., positive-activated affect, positive-deactivated affect,
negative-activated affect, negative-deactivated affect) on days when they engaged in more
physical activity compared to their average levels of activity. Results indicated that greater day-
level variability in positive-activated affect was associated with greater physical activity on that
same day compared to other days, whereas greater day-level variability in negative-deactivated
affect was associated with less physical activity on that same day compared to other days. This
study extends beyond previous research by informing our understanding of the dynamics
inherent to the relationship between affect and physical activity, as well as gives insight into how
these processes differ within individuals and unfold in the context of daily life.
The third and final study of the dissertation utilized a novel two-stage modeling approach
to examine whether the subject-level association between momentary affect (i.e., positive-

120

activated, positive-deactivated, negative-activated, negative-deactivated) and subsequent
physical activity predicted future physical activity levels. Results from the study suggest that
while there was subject-level heterogeneity the association between momentary affect (i.e.,
positive-activated, positive-deactivated, negative-activated, negative-deactivated affect) and
physical activity in the next 30 minutes, the strength of the association between momentary
affect and subsequent physical activity did not predict young adults’ future daily physical
activity levels one month later.
Overall, the dissertation findings suggest that affective variability may not necessarily be
disadvantageous to engaging in physical activity among young adults. In line with theoretical
models suggesting that experiencing fluctuations in affect may deplete an individual’s self-
regulatory resources and therefore make it increasingly difficult to engage in behaviors in line
with goals (e.g., engaging in physical activity), it was hypothesized that affective variability
would be associated with less physical activity (Baumeister & Vohs, 2004; Muraven &
Baumeister, 2000). The second study’s findings do partially support the overall dissertation
hypothesis, such that fluctuations in negative-deactivated affect (i.e., feeling sad and fatigued)
were associated with less physical activity on the same-day compared to other days. However,
the first study reported that subject-level variability in negative-deactivated affect was associated
with greater overall levels of physical activity. These findings suggest that the relationships
between affective variability and physical activity may differ across different time-scales (e.g.,
days vs. months). The second study also reported that variability in positive-activated affect (i.e.,
feeling happy and energetic) was associated with more physical activity on the same day
compared to other days. While future research is needed to further investigate the potential
causal relationships, these findings suggest affect variability may not necessarily be

121

disadvantageous to physical activity; individuals may engage in physical activity to manage their
fluctuations in their affect or change their affective state. Furthermore, the different directions of
association between variability in negative-deactivated affect and physical activity in the first
two studies highlights the importance of examining associations at the day-level, which may
provide more insight into nuanced associations of time-varying constructs. The null findings
across the three studies also suggest that, as assessed in this study, affective variability may not
play a significant influential role on physical activity among young adults. Future research is
warranted to confirm these findings and further elucidate the dynamic relationships between
affective states and physical activity.
Overall Strengths
This dissertation had several strengths that future research studies on affect and physical
activity could build upon. First, the frequent EMA assessments (i.e., hourly prompts during burst
assessments) and the long study period captured intensive longitudinal data that was able to
provide a robust measure of variability in affect. Previous research has been limited by the
inability to successfully model variability in negative affective states; these dissertation studies
are among the first known physical activity studies to model variability in negative affect.
Second, the software program MixWILD estimated within-subject variability by taking into
consideration that individuals contribute a different number of observations in the analyses
unlike standard methods for calculating variability (e.g., subject-level standard deviation, mean
square of successive differences). MixWILD recognizes that subjects can vary in terms of the
number of observations in the analysis and thus provides unbiased estimates with varying
degrees of precision (Dzubur et al., 2020). This is particularly important in EMA studies when
repeated assessments in naturalistic settings results in varying number of observations. Third, the

122

use of mixed-effects location scale modeling allowed important covariates (e.g., time of day, day
of week, age, sex at birth) to influence within-subject variance and allowed each subject to have
their own degree of within-subject variance above and beyond the effects of covariates. Fourth,
the dissertation was able to explore unique research questions that have previously been limited
by statistical methods and software. For example, the third study examined whether subject-level
slopes between two time-varying constructs (i.e., momentary affect and subsequent physical
activity) was associated with subject-level physical activity. Unlike traditional multilevel models
that assume the variances and covariances of the random effects are homogeneous across
subjects, MixWILD allowed subject-level variances (e.g., intercept, slope) of time-varying
variables (affect and physical activity) to be modeled as covariates to predict future physical
activity.
Overall Limitations
The dissertation also had several limitations that should be considered when interpreting
the findings and for future research studies. First, physical activity was operationalized by
MIMS-units, which measures the amount of movement and not necessarily “physical activity”.
MIMS-units are a standardized, nonproprietary metric that can be applied across different
accelerometers and measures the amount of movement independent of the type of sensor (e.g.,
smartwatch, smartphone) (John et al., 2019). However, unlike other device-based measures that
convert raw accelerometry data into metabolic equivalents (METs) or minutes in moderate-to-
vigorous physical activity, MIMS-units have yet to be converted or compared to values that have
meaningful public health implications (e.g., Physical Activity Guidelines for Americans). Given
that MIMS-units represents movement, it is possible that non-intentional physical activity was
captured on the smartwatch. Therefore, the findings should take into consideration that the

123

physical activity assessed in this dissertation was measured through MIMS-units. Second, the
dissertation was limited by utilizing only a few times to assess affect as an attempt to reduce
participant burden. In particular, only one item (“relaxed”) was used to assess positive-
deactivated affect. The narrow assessment of affective states may not fully capture variability
within constructs. Third, data for this dissertation came from the TIME Study, which was
collected during the COVID-19 pandemic (i.e., data collection occurred from March 2020 –
August 2022). The pandemic may have drastically changed individuals’ day-to-day lives.
Varying restrictions and closures across counties may have influenced physical activity and
affective states differently; therefore, results may not be generalizable to non-pandemic times
(Courtney et al., 2021; Hamidi & Zandiatashbar, 2021). Fourth, the study included a general
sample of young adults who were currently active or had intentions to be active. To address the
TIME Study’s overall aims, inclusion criteria included currently engaging in at least 150 minutes
of moderate-to-vigorous physical activity or intending to engage in physical activity over the
next year (Wang et al., 2022). While this criterion captured sufficient physical activity data to
address the larger study’s aims, the specific associations between affective variability and
physical activity may be different among individuals who regularly engage in activity versus
those who are not physically active. For example, the strength of association between variability
in affect and physical activity may be stronger among individuals who are not regularly
active/engage in habitual activity (Maher et al., 2019; Wood & Neal, 2016). In addition, the
study findings may not be generalizable to different age groups or those with varying clinical
conditions. Variability in momentary affective states and regulation of affective states may differ
across the lifespan (Ebner & Fischer, 2014; Zimmermann & Iwanski, 2014). Therefore, affective
variability may be associated with physical activity among certain sub-populations; additional

124

research among other samples of differing ages and clinical conditions is warranted. For
example, affect variability may be an important predictor of physical activity in individuals with
psychiatric disorders, given the greater severity of affect dysregulation in these populations
(Aldao et al., 2010; Marwaha et al., 2014).
Implications
Despite the noted study limitations, the dissertation work has implications for theory,
research methods, and behavior change interventions. Theoretical frameworks aiming to explain
how affective states influence health behaviors (e.g., Affect and Health Behavior Framework,
Affective Reflective Theory) should consider average levels of affect, momentary affect, and
variability in affect (Brand & Ekkekakis, 2018; Conroy & Berry, 2017; Williams et al., 2018;
Williams & Evans, 2014). For example, usual (i.e., chronic) and momentary (i.e., acute) levels of
affective determinants are thought to influence and be influenced by behavior; however, the
degree of stability/instability of one’s affective states over time have yet to be considered in
theoretical frameworks regarding affect and physical activity. Taken together with the published
literature on affective variability and other health behaviors (Anestis et al., 2009; Bos et al.,
2019; Jahng et al., 2011; Maher et al., 2019; Mermelstein et al., 2010; Weiss et al., 2018), this
dissertation substantiates the need to consider the affect dynamics above and beyond mean
levels. While this dissertation focused on incidental affect—how one feels throughout the day
and is unrelated to any specific target behavior—assessing interindividual variability in other
affective determinants may also provide valuable insight into the affect-physical activity
relationship.
In addition to theoretical implications, this dissertation also has methodological
implications for future studies on affect and physical activity. The dissertation provides empirical

125

evidence for the feasibility of coupling smartwatch and smartphone technology to assess states
and behaviors across a year. The TIME Study utilized hourly signal-contingent EMA to assess
momentary affect; this dense sampling scheme gathered hundreds to thousands of EMA
observations per participant that allowed us to estimate within-subject variance in affective
states. While hourly EMA may feel burdensome to some study participants, this dense sampling
only occurred during measurement bursts (i.e., every two weeks). Future EMA studies could
implement similar study designs to gather intensive longitudinal data in naturalistic settings. The
study design also permitted modeling variability in negative-activated and negative-deactivated
affect; prior work has been limited in the ability to estimate variability in negatively-valenced
affective states (Maher et al., 2019). The findings suggest that there are both day-level and
subject-level associations between negative-deactivated affect and physical activity. The
dissertation also provides a recent example of using MIMS-units to operationalize physical
activity (John et al., 2019). MIMS-units provides a unique opportunity to capture continuous
bodily movement through a non-proprietary metric that can be utilized across different devices to
compare findings across distinct research studies. Future work can compare activity to
population referenced MIMS-unit percentiles, and eventually compare these metrics to activity
intensity categories that map onto public health recommendations (Belcher et al., 2021). The
coupling of frequent EMA measurements and continuous activity monitoring should be
leveraged in future research to fully understand the dynamic relationships between affect and
physical activity in naturalistic settings. Lastly, our work provides three empirical examples of
applying mixed-effects location scale modeling in MixWILD. Future health behavior research
can benefit by using MixWILD to test (1) predictors of variability in time-varying outcomes and

126

(2) associations between subject-level parameters (i.e., variances and slopes) with time-varying
or subject-level outcomes.
The dissertation work also has implications for the development of behavior change
interventions. While additional research is warranted to validate the findings and elucidate the
mechanisms underlying the associations between affective variability and physical activity, the
dissertation work lays the foundation for the development of interventions seeking to increase
engagement in physical activity. The second study of the dissertation found that day-level
associations of variability in positive-activated affect (i.e., feeling happy and energetic) and
negative-deactivated affect (i.e., feeling sad and fatigued) with physical activity. Given these
findings, future research could explore the potential effects of strategies delivered in real-time to
target fluctuations in affect and increase physical activity by helping individual self-regulate or
cope with fluctuations in affect, such as prompts for emotion regulation exercises, mindfulness
practices, or brief coaching sessions (Nahum-Shani et al., 2018; Valle et al., 2020). The findings
highlight the importance of examining associations occurring at the day-level, which support the
application of novel ecological interventions in individuals’ daily lives that adjust to a
participant’s momentary situations (e.g., just-in-time adaptive interventions; Nahum-Shani et al.,
2018), opposed to traditional behavior change interventions that may only target subject-level
characteristics, such as dispositions to experience high negative or low positive affect. The third
dissertation study findings—that young adults differ in their strength of association in affect and
subsequent physical activity (i.e., between-subject heterogeneity)— also have implications for
future research in intervention development. Future work could determine the efficacy of tailored
interventions, such as whether targeting momentary affect to increase physical activity is more
effective among certain individuals (e.g., those with a stronger within-subject association). For

127

those individuals whose physical activity is contingent on their affect, mobile interventions could
deliver content to help individuals regulate their affect. Furthermore, tailored programming for
young adults could help encourage the development of physical activity habits by incorporating
strategies such as setting goals and plans, and methods to follow through with those plans even
negative affect is high.
Future Research Directions
There are several opportunities for future research to build upon the findings of the
dissertation work. One opportunity for future work is to examine the temporal relationships of
affective variability and physical activity within the day by conducting additional EMA studies.
To our best knowledge, the second study is among the first to examine associations of affective
variability and physical activity at the day-level. Future studies may expand upon this line of
research to determine whether variability over a shorter time frame (e.g., across hours) predicts
subsequent physical activity (e.g., over the next hour) and vice versa. Similar to momentary
affect, there may be bi-directional associations such that affective variability may predict
subsequent activity and physical activity stabilize affect or result in fluctuations (Kim et al.,
2020; Liao et al., 2015).
Another opportunity for future research is to investigate whether momentary self-control
moderates or mediates associations between affective variability and physical activity at the day-
level. Results from the second study indicate that trait self-control did not moderate associations
between subject-level affective variability and overall physical activity; however, both affective
variability and self-control may influence physical activity in the moment rather than at the
subject-level, as evidenced in the first study. An individuals’ self-control and self-regulatory
capacity may differ day-to-day based on internal and environmental events. Empirical research

128

suggests that inhibitory control (i.e., the ability to stop, change, or delay inappropriate behaviors)
is not a stable trait, but rather fluctuates in response to internal events (e.g., stress, depletion of
self-control resources) or environmental contexts (e.g., behavior-related cues) (Muraven et al.,
2005; Muraven & Baumeister, 2000; Zack et al., 2011). For example, short-term fluctuations in
inhibitory control predicted increased alcohol consumption (Jones et al., 2018). By
understanding the potential mechanisms underlying the relationship between affective variability
and activity, efficacious interventions can be developed.
Future research on affect and physical activity should consider the different
conceptualizations and operationalization of affective states. In this dissertation, affective states
were grouped together based on the two fundamental dimensions of affect (i.e., valence and
arousal), resulting in combinations of positive/negative and activated/deactivated (Russell,
1980). While this assists in making more precise conclusions on affect—compared to the
traditional method of combining valence and arousal through the assessment of positive affect
and negative affect—it can be difficult to compare findings across different research studies. For
example, Maher and colleagues examined the associations between positive affect (i.e., happy,
joyful, cheerful, calm) and arousal (i.e., energetic) (Maher et al., 2019); it is challenging to
completely compare previous findings to the current work given that happy and energetic were
assessed together through positive-activated affect and feeling relaxed was assessed separately as
positive-deactivated affect. Overall, the field of exercise psychology may benefit from utilizing
common affective measures to better compare studies and findings; however, the
conceptualization and operationalization of affect continues to be an area of debate (Ekkekakis,
2013; Ekkekakis & Petruzzello, 2000). Another alternative method to measuring affect is to take
a ‘distinct states approach’, where a single state is examined (e.g., feeling happy) rather than

129

averaging scores across multiple states to create constructs like positive and negative affect
(Ekkekakis, 2013). This strategy could promote consistency across studies and enable
researchers to potentially pinpoint potential associations between affect and physical activity. For
example, variability in arousal states (e.g., energetic, fatigued) may be more strongly associated
with physical activity. Arousal may be particularly relevant for physical activity, where
arousal/activation states may be an antecedent, concomitant, or consequence of activity and
exertion.
To better understand how fluctuations in affect influence physical activity, future
research should also consider emotion regulation and mental health. Similar to self-control,
emotion regulation may moderate the strength of association between affective variability and
physical activity, such that positive emotion regulation strategies may buffer the negative effects
of affective variability on physical activity (Blanke et al., 2020; Burr et al., 2021; Koval et al.,
2013). Future research may benefit by using EMA to assess use of emotion regulation strategies
and assessing whether they moderate day-level associations between affective variability and
physical activity. Considering mental health characteristics may also be important for future
research. Variability in affect has largely been examined in psychopathology research given that
affect dysregulation is symptom of some psychiatric disorders (Aldao et al., 2010; Marwaha et
al., 2014). Conducting similar research among different samples, such as those with diagnosed
clinical conditions, may provide insight whether affective variability is an important predictor of
physical activity for certain sub-populations. While the 25% of the current study sample self-
reported a diagnosis of a depressive disorder at some time in their life, we did not have sufficient
information to their current diagnoses. The overall associations between affective variability and
physical activity may also be moderated by varying levels of mental health characteristics.

130

Conclusions
This dissertation used real-time data capture techniques such as EMA to repeatedly assess
affective states among young adults in naturalistic settings and examine the relationship between
variability in affect (i.e., within-subject variance) and physical activity. By extending beyond
previous literature, which has largely focused on usual or momentary levels of affect predicting
physical activity, this dissertation investigated fluctuations in affect intensity as a determinant of
physical activity. Findings revealed that greater variability in positive-activated affect was
associated with greater same-day physical activity compared to other days, while greater
variability in negative-deactivated affect was associated with less same-day physical activity
compared to other days. The findings also indicated that individuals who had more overall
variability in negative-deactivated affect engaged in more overall levels of physical activity.
Taken together, the results underscore how variability in different affective states may
differentially be associated with physical activity depending on the valence, arousal/activation,
and temporality of constructs and associations. In addition, the findings highlight the importance
of moving beyond assessing only mean or momentary levels of affect. This dissertation provides
a strong foundation for future studies in the field of affect and physical activity research by
providing preliminary evidence for day-level associations between affective variability and
physical activity and empirical examples of utilizing novel two-stage modeling that applies
mixed-effects location scale modeling to estimate within-subject variance in affect. Results from
this dissertation support the use of EMA to assess affective variability, the application of novel
statistical methods to model variability, and consideration of affect dynamics in the development
of future physical activity interventions.

131

REFERENCES

2018 Physical Activity Guidelines Advisory Committee Scientific Report. (2018). 779.
Aldao, A., Nolen-Hoeksema, S., & Schweizer, S. (2010). Emotion-regulation strategies across
psychopathology: A meta-analytic review. Clinical Psychology Review, 30(2), 217–237.
https://doi.org/10.1016/j.cpr.2009.11.004
Allemand, M., Keller, L., Gmür, B., Gehriger, V., Oberholzer, T., & Stieger, M. (2020).
MindHike, a Digital Coaching Application to Promote Self-Control: Rationale, Content,
and Study Protocol. Frontiers in Psychiatry, 11, 575101.
https://doi.org/10.3389/fpsyt.2020.575101
Anestis, M. D., Peterson, C. B., Bardone-Cone, A. M., Klein, M. H., Mitchell, J. E., Crosby, R.
D., Wonderlich, S. A., Crow, S. J., le Grange, D., & Joiner, T. E. (2009). Affective
lability and impulsivity in a clinical sample of women with bulimia nervosa: The role of
affect in severely dysregulated behavior. The International Journal of Eating Disorders,
42(3), 259–266. https://doi.org/10.1002/eat.20606
Arguello, D., Andersen, K., Morton, A., Freedson, P. S., Intille, S. S., & John, D. (2018).
Validity of proximity sensor-based wear-time detection using the ActiGraph GT9X.
Journal of Sports Sciences, 36(13), 1502–1507.
https://doi.org/10.1080/02640414.2017.1398891
Arnett, J. J. (2015). Emerging adulthood: The winding road from the late teens through the
twenties, 2nd ed (pp. xvi, 394). Oxford University Press.
https://doi.org/10.1093/oxfordhb/9780199795574.013.9
Arnett, J. J., Žukauskienė, R., & Sugimura, K. (2014). The new life stage of emerging adulthood
at ages 18–29 years: Implications for mental health. The Lancet Psychiatry, 1(7), 569–
576. https://doi.org/10.1016/S2215-0366(14)00080-7
Baker, T. B., Piper, M. E., McCarthy, D. E., Majeskie, M. R., & Fiore, M. C. (2004). Addiction
Motivation Reformulated: An Affective Processing Model of Negative Reinforcement.
Psychological Review, 111(1), 33–51. https://doi.org/10.1037/0033-295X.111.1.33
Bandura, A. (1989). Human agency in social cognitive theory. The American Psychologist,
44(9), 1175–1184. https://doi.org/10.1037/0003-066x.44.9.1175
Bauman, A. E., Reis, R. S., Sallis, J. F., Wells, J. C., Loos, R. J., & Martin, B. W. (2012).
Correlates of physical activity: Why are some people physically active and others not?
The Lancet, 380(9838), 258–271. https://doi.org/10.1016/S0140-6736(12)60735-1
Baumeister, R. F., Bratslavsky, E., Muraven, M., & Tice, D. M. (1998). Ego depletion: Is the
active self a limited resource? Journal of Personality and Social Psychology, 74(5),
1252–1265. https://doi.org/10.1037/0022-3514.74.5.1252

132

Baumeister, R. F., & Vohs, K. D. (2007). Self-Regulation, Ego Depletion, and Motivation.
Social and Personality Psychology Compass, 1(1), 115–128.
https://doi.org/10.1111/j.1751-9004.2007.00001.x
Belcher, B. R., Wolff-Hughes, D. L., Dooley, E. E., Staudenmayer, J., Berrigan, D., Eberhardt,
M. S., & Troiano, R. P. (2021). US Population-referenced Percentiles for Wrist-Worn
Accelerometer-derived Activity. Medicine and Science in Sports and Exercise, 53(11),
2455–2464. https://doi.org/10.1249/MSS.0000000000002726
Booth, F. W., Roberts, C. K., & Laye, M. J. (2012). Lack of exercise is a major cause of chronic
diseases. Comprehensive Physiology, 2(2), 1143–1211.
https://doi.org/10.1002/cphy.c110025
Bos, E. H., de Jonge, P., & Cox, R. F. A. (2019). Affective variability in depression: Revisiting
the inertia–instability paradox. British Journal of Psychology, 110(4), 814–827.
https://doi.org/10.1111/bjop.12372
Brand, R., & Ekkekakis, P. (2018). Affective–Reflective Theory of physical inactivity and
exercise. German Journal of Exercise and Sport Research, 48(1), 48–58.
https://doi.org/10.1007/s12662-017-0477-9
Buckley, J., Cohen, J. D., Kramer, A. F., McAuley, E., & Mullen, S. P. (2014). Cognitive control
in the self-regulation of physical activity and sedentary behavior. Frontiers in Human
Neuroscience, 8. https://doi.org/10.3389/fnhum.2014.00747
Burke, G. L., Bild, D. E., Hilner, J. E., Folsom, A. R., Wagenknecht, L. E., & Sidney, S. (1996).
Differences in weight gain in relation to race, gender, age and education in young adults:
The CARDIA study. Ethnicity & Health, 1(4), 327–335.
https://doi.org/10.1080/13557858.1996.9961802
Carels, R. A., Coit, C., Young, K., & Berger, B. (2007). Exercise makes you feel good, but does
feeling good make you exercise?: An examination of obese dieters. Journal of Sport &
Exercise Psychology, 29(6), 706–722. https://doi.org/10.1123/jsep.29.6.706
Carsey, T. M., & Harden, J. J. (2013). Monte Carlo Simulation and Resampling Methods for
Social Science. SAGE Publications.
Caspersen, C. J., Powell, K. E., & Christenson, G. M. (1985). Physical activity, exercise, and
physical fitness: Definitions and distinctions for health-related research. Public Health
Reports, 100(2), 126–131.
Chan, D. K. C., Zhang, X., Fung, H. H., & Hagger, M. S. (2016). Affect, Affective Variability,
and Physical Health: Results from a Population-Based Investigation in China.
International Journal of Behavioral Medicine, 23(4), 438–446.
https://doi.org/10.1007/s12529-015-9510-2

133

Conroy, D. E., & Berry, T. R. (2017). Automatic Affective Evaluations of Physical Activity.
Exercise and Sport Sciences Reviews, 45(4), 230–237.
https://doi.org/10.1249/JES.0000000000000120
Conroy, M. B., Simkin-Silverman, L. R., Pettee, K. K., Hess, R., Kuller, L. H., & Kriska, A. M.
(2007). Lapses and Psychosocial Factors Related to Physical Activity in Early
Postmenopause. Medicine & Science in Sports & Exercise, 39(10), 1858–1866.
https://doi.org/10.1249/mss.0b013e318137388d
Courtney, J., Nuss, K., Wang, S., Do, B., & Dunton, G. (2021). Using a daily diary approach to
examine the early effects of COVID-19 on daily physical activity bouts and contexts
among residents of Colorado and California. Translational Behavioral Medicine, 11(9),
1771–1781. https://doi.org/10.1093/tbm/ibab066
Crescioni, A. W., Ehrlinger, J., Alquist, J. L., Conlon, K. E., Baumeister, R. F., Schatschneider,
C., & Dutton, G. R. (2011). High trait self-control predicts positive health behaviors and
success in weight loss. Journal of Health Psychology, 16(5), 750–759.
https://doi.org/10.1177/1359105310390247
Curran, P. J., & Bauer, D. J. (2011). The Disaggregation of Within-Person and Between-Person
Effects in Longitudinal Models of Change. Annual Review of Psychology, 62, 583–619.
https://doi.org/10.1146/annurev.psych.093008.100356
Davidson, R. J. (1998). Affective Style and Affective Disorders: Perspectives from Affective
Neuroscience. Cognition and Emotion, 12(3), 307–330.
https://doi.org/10.1080/026999398379628
Daw, J., Margolis, R., & Wright, L. (2017). Emerging Adulthood, Emergent Health Lifestyles:
Sociodemographic Determinants of Trajectories of Smoking, Binge Drinking, Obesity,
and Sedentary Behavior. Journal of Health and Social Behavior, 58(2), 181–197.
https://doi.org/10.1177/0022146517702421
De Bourdeaudhuij, I., Sallis, J. F., & Saelens, B. E. (2003). Environmental Correlates of Physical
Activity in a Sample of Belgian Adults. American Journal of Health Promotion, 18(1),
83–92. https://doi.org/10.4278/0890-1171-18.1.83
Do, B., Wang, S. D., Courtney, J. B., & Dunton, G. F. (2021). Examining the day-level impact of
physical activity on affect during the early months of the COVID-19 pandemic: An
ecological momentary assessment study. Psychology of Sport and Exercise, 56, 102010.
https://doi.org/10.1016/j.psychsport.2021.102010
Dowd, J. B., & Zajacova, A. (2014). Long-Term Obesity and Cardiovascular, Inflammatory, and
Metabolic Risk in U.S. Adults. American Journal of Preventive Medicine, 46(6), 578–
584. https://doi.org/10.1016/j.amepre.2014.01.016
Dunton, G. F. (2017). Ecological Momentary Assessment in Physical Activity Research.
Exercise and Sport Sciences Reviews, 45(1), 48–54.
https://doi.org/10.1249/JES.0000000000000092

134

Dunton, G. F. (2018). Sustaining Health-Protective Behaviors Such as Physical Activity and
Healthy Eating. JAMA, 320(7), 639–640. https://doi.org/10.1001/jama.2018.6621
Dunton, G. F., Atienza, A. A., Castro, C. M., & King, A. C. (2009). Using ecological momentary
assessment to examine antecedents and correlates of physical activity bouts in adults age
50+ years: A pilot study. Annals of Behavioral Medicine: A Publication of the Society of
Behavioral Medicine, 38(3), 249–255. https://doi.org/10.1007/s12160-009-9141-4
Dunton, G. F., Huh, J., Leventhal, A. M., Riggs, N., Hedeker, D., Spruijt-Metz, D., & Pentz, M.
A. (2014). Momentary assessment of affect, physical feeling states, and physical activity
in children. Health Psychology, 33(3), 255–263. https://doi.org/10.1037/a0032640
Dunton, G. F., Wang, S. D., Do, B., & Courtney, J. (2020). Early effects of the COVID-19
pandemic on physical activity locations and behaviors in adults living in the United
States. Preventive Medicine Reports, 20, 101241.
https://doi.org/10.1016/j.pmedr.2020.101241
Dzubur, E., Ponnada, A., Nordgren, R., Yang, C.-H., Intille, S., Dunton, G., & Hedeker, D.
(2020). MixWILD: A program for examining the effects of variance and slope of time-
varying variables in intensive longitudinal data. Behavior Research Methods, 52(4),
1403–1427. https://doi.org/10.3758/s13428-019-01322-1
Ebner, N. C., & Fischer, H. (2014). Emotion and aging: Evidence from brain and behavior.
Frontiers in Psychology, 5.
https://www.frontiersin.org/articles/10.3389/fpsyg.2014.00996
Ebner-Priemer, U. W., Eid, M., Kleindienst, N., Stabenow, S., & Trull, T. J. (2009). Analytic
strategies for understanding affective (in)stability and other dynamic processes in
psychopathology. Journal of Abnormal Psychology, 118(1), 195–202.
https://doi.org/10.1037/a0014868
Ebner-Priemer, U. W., & Trull, T. J. (2012). Investigating temporal instability in psychological
variables: Understanding the real world as time dependent. In Handbook of research
methods for studying daily life (pp. 423–439). The Guilford Press.
Eid, M., & Diener, E. (1999). Intraindividual variability in affect: Reliability, validity, and
personality correlates. Journal of Personality and Social Psychology, 76(4), 662–676.
https://doi.org/10.1037/0022-3514.76.4.662
Ekkekakis, P. (2013). The Measurement of Affect, Mood, and Emotion: A Guide for Health-
Behavioral Research. Cambridge University Press.
Ekkekakis, P., & Brand, R. (2019). Affective responses to and automatic affective valuations of
physical activity: Fifty years of progress on the seminal question in exercise psychology.
50 Years of FEPSAC: Current and Future Directions to Sport and Exercise Psychology
Research, 42, 130–137. https://doi.org/10.1016/j.psychsport.2018.12.018

135

Ekkekakis, P., & Petruzzello, S. J. (2000). Analysis of the affect measurement conundrum in
exercise psychology: I. Fundamental issues. Psychology of Sport and Exercise, 1(2), 71–
88. https://doi.org/10.1016/S1469-0292(00)00010-8
Ekkekakis, P., & Petruzzello, S. J. (2002). Analysis of the affect measurement conundrum in
exercise psychology: IV. A conceptual case for the affect circumplex. Psychology of
Sport and Exercise, 3(1), 35–63. https://doi.org/10.1016/S1469-0292(01)00028-0
Emerson, J. A., Dunsiger, S., & Williams, D. M. (2018). Reciprocal within-day associations
between incidental affect and exercise: An EMA study. Psychology & Health, 33(1),
130–143. https://doi.org/10.1080/08870446.2017.1341515
Englert, C. (2016). The Strength Model of Self-Control in Sport and Exercise Psychology.
Frontiers in Psychology, 7.
https://www.frontiersin.org/articles/10.3389/fpsyg.2016.00314
Gordon-Larsen, P., Nelson, M. C., & Popkin, B. M. (2004). Longitudinal physical activity and
sedentary behavior trends: Adolescence to adulthood. American Journal of Preventive
Medicine, 27(4), 277–283. https://doi.org/10.1016/j.amepre.2004.07.006
Gross, J. J. (2013). Handbook of Emotion Regulation, Second Edition. Guilford Publications.
Hagger, M. S., Wood, C., Stiff, C., & Chatzisarantis, N. L. D. (2010a). Ego depletion and the
strength model of self-control: A meta-analysis. Psychological Bulletin, 136(4), 495–525.
https://doi.org/10.1037/a0019486
Hagger, M. S., Wood, C. W., Stiff, C., & Chatzisarantis, N. L. D. (2010b). Self-regulation and
self-control in exercise: The strength-energy model. International Review of Sport and
Exercise Psychology, 3(1), 62–86. https://doi.org/10.1080/17509840903322815
Hagger, M. S., Zhang, C.-Q., Kangro, E.-M., Ries, F., Wang, J. C. K., Heritage, B., & Chan, D.
K. C. (2021). Trait self-control and self-discipline: Structure, validity, and invariance
across national groups. Current Psychology, 40(3), 1015–1030.
https://doi.org/10.1007/s12144-018-0021-6
Hales, C. M., Carroll, M. D., Fryar, C. D., & Ogden, C. L. (2020). Prevalence of Obesity and
Severe Obesity Among Adults: United States, 2017–2018. 360, 8.
Hall, P. A., Fong, G. T., & Lowe, C. J. (2018). Affective Dynamics in Temporal Self-Regulation
Theory: Social Forces Meet Neurobiological Processes. In Affective Determinants of
Health Behavior. Oxford University Press.
https://doi.org/10.1093/oso/9780190499037.003.0006
Hamidi, S., & Zandiatashbar, A. (2021). Compact development and adherence to stay-at-home
order during the COVID-19 pandemic: A longitudinal investigation in the United States.
Landscape and Urban Planning, 205, 103952.
https://doi.org/10.1016/j.landurbplan.2020.103952

136

Harris, P. A., Taylor, R., Minor, B. L., Elliott, V., Fernandez, M., O’Neal, L., McLeod, L.,
Delacqua, G., Delacqua, F., Kirby, J., & Duda, S. N. (2019). The REDCap consortium:
Building an international community of software platform partners. Journal of
Biomedical Informatics, 95, 103208. https://doi.org/10.1016/j.jbi.2019.103208
Harris, P. A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., & Conde, J. G. (2009). Research
electronic data capture (REDCap)—A metadata-driven methodology and workflow
process for providing translational research informatics support. Journal of Biomedical
Informatics, 42(2), 377–381. https://doi.org/10.1016/j.jbi.2008.08.010
Hedeker, D., Mermelstein, R. J., & Demirtas, H. (2008). An Application of a Mixed-Effects
Location Scale Model for Analysis of Ecological Momentary Assessment (EMA) Data.
Biometrics, 64(2), 627–634. https://doi.org/10.1111/j.1541-0420.2007.00924.x
Hedeker, D., & Nordgren, R. (2013). MIXREGLS: A Program for Mixed-Effects Location Scale
Analysis. Journal of Statistical Software, 52(12), 1–38.
https://doi.org/10.18637/jss.v052.i12
Jahng, S., Solhan, M. B., Tomko, R. L., Wood, P. K., Piasecki, T. M., & Trull, T. J. (2011).
Affect and Alcohol Use: An Ecological Momentary Assessment Study of Outpatients
With Borderline Personality Disorder. Journal of Abnormal Psychology, 120(3), 572–
584. https://doi.org/10.1037/a0024686
Jakicic, J. M., Kraus, W. E., Powell, K. E., Campbell, W. W., Janz, K. F., Troiano, R. P., Sprow,
K., Torres, A., & Piercy, K. L. (2019). Association between Bout Duration of Physical
Activity and Health: Systematic Review. Medicine and Science in Sports and Exercise,
51(6), 1213–1219. https://doi.org/10.1249/MSS.0000000000001933
Jakicic, J. M., Powell, K. E., Campbell, W. W., Dipietro, L., Pate, R. R., Pescatello, L. S.,
Collins, K. A., Bloodgood, B., Piercy, K. L., & 2018 PHYSICAL ACTIVITY
GUIDELINES ADVISORY COMMITTEE*. (2019). Physical Activity and the
Prevention of Weight Gain in Adults: A Systematic Review. Medicine and Science in
Sports and Exercise, 51(6), 1262–1269. https://doi.org/10.1249/MSS.0000000000001938
John, D., Tang, Q., Albinali, F., & Intille, S. (2019). An Open-Source Monitor-Independent
Movement Summary for Accelerometer Data Processing. Journal for the Measurement of
Physical Behaviour, 2(4), 268–281. https://doi.org/10.1123/jmpb.2018-0068
Jones, A., Tiplady, B., Houben, K., Nederkoorn, C., & Field, M. (2018). Do daily fluctuations in
inhibitory control predict alcohol consumption? An ecological momentary assessment
study. Psychopharmacology, 235(5), 1487–1496. https://doi.org/10.1007/s00213-018-
4860-5
Kaczynski, A. T., & Henderson, K. A. (2007). Environmental Correlates of Physical Activity: A
Review of Evidence about Parks and Recreation. Leisure Sciences, 29(4), 315–354.
https://doi.org/10.1080/01490400701394865

137

Kanning, M., & Schoebi, D. (2016). Momentary Affective States Are Associated with
Momentary Volume, Prospective Trends, and Fluctuation of Daily Physical Activity.
Frontiers in Psychology, 7. https://doi.org/10.3389/fpsyg.2016.00744
Kerrigan, S. G., Schumacher, L., Manasse, S. M., Loyka, C., Butryn, M. L., & Forman, E. M.
(2020). The association between negative affect and physical activity among adults in a
behavioral weight loss treatment. Psychology of Sport and Exercise, 47, 101507.
https://doi.org/10.1016/j.psychsport.2019.03.010
Kim, J., Conroy, D. E., & Smyth, J. M. (2020). Bidirectional Associations of Momentary Affect
with Physical Activity and Sedentary Behaviors in Working Adults. Annals of Behavioral
Medicine, 54(4), 268–279. https://doi.org/10.1093/abm/kaz045
Larsen, R. J. (1987). The stability of mood variability: A spectral analytic approach to daily
mood assessments. Journal of Personality and Social Psychology, 52(6), 1195–1204.
https://doi.org/10.1037/0022-3514.52.6.1195
Lazarus, R. S. (1991). Emotion and Adaptation. Oxford University Press.
Lazarus, R. S. (1993). Coping theory and research: Past, present, and future. Psychosomatic
Medicine, 55(3), 234–247.
Lee, I.-M., Djoussé, L., Sesso, H. D., Wang, L., & Buring, J. E. (2010). Physical Activity and
Weight Gain Prevention. JAMA : The Journal of the American Medical Association,
303(12), 1173–1179. https://doi.org/10.1001/jama.2010.312
Leitzmann, M. F., Park, Y., Blair, A., Ballard-Barbash, R., Mouw, T., Hollenbeck, A. R., &
Schatzkin, A. (2007). Physical activity recommendations and decreased risk of mortality.
Archives of Internal Medicine, 167(22), 2453–2460.
https://doi.org/10.1001/archinte.167.22.2453
Leventhal, A. M., Strong, D. R., Kirkpatrick, M. G., Unger, J. B., Sussman, S., Riggs, N. R.,
Stone, M. D., Khoddam, R., Samet, J. M., & Audrain-McGovern, J. (2015). Association
of Electronic Cigarette Use With Initiation of Combustible Tobacco Product Smoking in
Early Adolescence. JAMA, 314(7), 700–707. https://doi.org/10.1001/jama.2015.8950
Liao, Y., Chou, C.-P., Huh, J., Leventhal, A., & Dunton, G. (2017). Examining acute bi-
directional relationships between affect, physical feeling states, and physical activity in
free-living situations using electronic ecological momentary assessment. Journal of
Behavioral Medicine, 40(3), 445–457. https://doi.org/10.1007/s10865-016-9808-9
Liao, Y., Shonkoff, E. T., & Dunton, G. F. (2015). The Acute Relationships Between Affect,
Physical Feeling States, and Physical Activity in Daily Life: A Review of Current
Evidence. Frontiers in Psychology, 6. https://doi.org/10.3389/fpsyg.2015.01975
Lloyd-Richardson, E. E., Bailey, S., Fava, J. L., & Wing, R. (2009). A prospective study of
weight gain during the college freshman and sophomore years. Preventive Medicine,
48(3), 256–261. https://doi.org/10.1016/j.ypmed.2008.12.009

138

Maher, J. P., Dzubur, E., Nordgren, R., Huh, J., Chou, C.-P., Hedeker, D., & Dunton, G. F.
(2019). Do fluctuations in positive affective and physical feeling states predict physical
activity and sedentary time? Psychology of Sport and Exercise, 41, 153–161.
https://doi.org/10.1016/j.psychsport.2018.01.011
Marwaha, S., He, Z., Broome, M., Singh, S. P., Scott, J., Eyden, J., & Wolke, D. (2014). How is
affective instability defined and measured? A systematic review. Psychological
Medicine, 44(9), 1793–1808. https://doi.org/10.1017/S0033291713002407
Mata, J., Thompson, R. J., Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Gotlib, I. H. (2012).
Walk on the bright side: Physical activity and affect in major depressive disorder. Journal
of Abnormal Psychology, 121(2), 297–308. https://doi.org/10.1037/a0023533
McDowell, C. P., Dishman, R. K., Gordon, B. R., & Herring, M. P. (2019). Physical Activity and
Anxiety: A Systematic Review and Meta-analysis of Prospective Cohort Studies.
American Journal of Preventive Medicine, 57(4), 545–556.
https://doi.org/10.1016/j.amepre.2019.05.012
McKee, S. A., Wall, A.-M., Hinson, R. E., Goldstein, A., & Bissonnette, M. (2003). Effects of an
implicit mood prime on the accessibility of smoking expectancies in college women.
Psychology of Addictive Behaviors: Journal of the Society of Psychologists in Addictive
Behaviors, 17(3), 219–225. https://doi.org/10.1037/0893-164X.17.3.219
Mermelstein, R., Hedeker, D., & Weinstein, S. (2010). Ecological momentary assessment of
mood-smoking relationships in adolescent smokers. In Substance abuse and emotion (pp.
217–236). American Psychological Association. https://doi.org/10.1037/12067-009
Merten, M. J. (2010). Weight Status Continuity and Change From Adolescence to Young
Adulthood: Examining Disease and Health Risk Conditions. Obesity, 18(7), 1423–1428.
https://doi.org/10.1038/oby.2009.365
Meyer, J., McDowell, C., Lansing, J., Brower, C., Smith, L., Tully, M., & Herring, M. (2020).
Changes in physical activity and sedentary behavior in response to covid-19 and their
associations with mental health in 3052 us adults. International Journal of Environmental
Research and Public Health, 17(18). https://doi.org/10.3390/ijerph17186469
Mislevy, R. J. (1991). Randomization-based inference about latent variables from complex
samples. Psychometrika, 56(2), 177–196. https://doi.org/10.1007/BF02294457
Mokdad, A. H., Ford, E. S., Bowman, B. A., Dietz, W. H., Vinicor, F., Bales, V. S., & Marks, J.
S. (2003). Prevalence of Obesity, Diabetes, and Obesity-Related Health Risk Factors,
2001. JAMA, 289(1), 76–79. https://doi.org/10.1001/jama.289.1.76
Muraven, M., & Baumeister, R. F. (2000). Self-regulation and depletion of limited resources:
Does self-control resemble a muscle? Psychological Bulletin, 126(2), 247–259.
https://doi.org/10.1037/0033-2909.126.2.247

139

Muraven, M., Collins, R. L., Shiffman, S., & Paty, J. A. (2005). Daily Fluctuations in Self-
Control Demands and Alcohol Intake. Psychology of Addictive Behaviors, 19, 140–147.
https://doi.org/10.1037/0893-164X.19.2.140
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. https://doi.org/10.1037/hea0000306
Nahum-Shani, I., Smith, S. N., Spring, B. J., Collins, L. M., Witkiewitz, K., Tewari, A., &
Murphy, S. A. (2018). 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. https://doi.org/10.1007/s12160-016-9830-8
Nesselroade, J. R., & Ram, N. (2004). Studying Intraindividual Variability: What We Have
Learned That Will Help Us Understand Lives in Context. Research in Human
Development, 1(1–2), 9–29. https://doi.org/10.1080/15427609.2004.9683328
Niermann, C. Y. N., Herrmann, C., von Haaren, B., van Kann, D., & Woll, A. (2016). Affect and
Subsequent Physical Activity: An Ambulatory Assessment Study Examining the Affect-
Activity Association in a Real-Life Context. Frontiers in Psychology, 7, 677.
https://doi.org/10.3389/fpsyg.2016.00677
Nocon, M., Hiemann, T., Müller-Riemenschneider, F., Thalau, F., Roll, S., & Willich, S. N.
(2008). Association of physical activity with all-cause and cardiovascular mortality: A
systematic review and meta-analysis. European Journal of Cardiovascular Prevention &
Rehabilitation, 15(3), 239–246. https://doi.org/10.1097/HJR.0b013e3282f55e09
Pedersen, B. K., & Saltin, B. (2015). Exercise as medicine—Evidence for prescribing exercise as
therapy in 26 different chronic diseases. Scandinavian Journal of Medicine & Science in
Sports, 25 Suppl 3, 1–72. https://doi.org/10.1111/sms.12581
Piercy, K. L., Troiano, R. P., Ballard, R. M., Carlson, S. A., Fulton, J. E., Galuska, D. A.,
George, S. M., & Olson, R. D. (2018). The Physical Activity Guidelines for Americans.
JAMA, 320(19), 2020–2028. https://doi.org/10.1001/jama.2018.14854
Prochaska, J. O., & Velicer, W. F. (1997). The Transtheoretical Model of Health Behavior
Change. American Journal of Health Promotion, 12(1), 38–48.
https://doi.org/10.4278/0890-1171-12.1.38
Reed, J., & Buck, S. (2009). The effect of regular aerobic exercise on positive-activated affect: A
meta-analysis. Psychology of Sport and Exercise, 10(6), 581–594.
https://doi.org/10.1016/j.psychsport.2009.05.009
Reichert, M., Tost, H., Reinhard, I., Zipf, A., Salize, H.-J., Meyer-Lindenberg, A., & Ebner-
Priemer, U. W. (2016). Within-Subject Associations between Mood Dimensions and
Non-exercise Activity: An Ambulatory Assessment Approach Using Repeated Real-Time
and Objective Data. Frontiers in Psychology, 7, 918.
https://doi.org/10.3389/fpsyg.2016.00918

140

Rhodes, R. E., & Gray, S. M. (2018). Affect in the process of action control of health protective
behaviors. In Affective Determinants of Health Behavior. Oxford University Press.
Rhodes, R. E., Mark, R. S., & Temmel, C. P. (2012). Adult sedentary behavior: A systematic
review. American Journal of Preventive Medicine, 42(3), e3-28.
https://doi.org/10.1016/j.amepre.2011.10.020
Rhodes, R. E., McEwan, D., & Rebar, A. L. (2019). Theories of physical activity behaviour
change: A history and synthesis of approaches. Psychology of Sport and Exercise, 42,
100–109. https://doi.org/10.1016/j.psychsport.2018.11.010
Rothman, A. J. (2000). Toward a theory-based analysis of behavioral maintenance. Health
Psychology, 19(1, Suppl), 64–69. https://doi.org/10.1037/0278-6133.19.Suppl1.64
Russell, J. A. (1978). Evidence of convergent validity on the dimensions of affect. Journal of
Personality and Social Psychology, 36(10), 1152–1168. https://doi.org/10.1037/0022-
3514.36.10.1152
Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social
Psychology, 39(6), 1161–1178. https://doi.org/10.1037/h0077714
Russell, J. A. (2009). Emotion, core affect, and psychological construction. Cognition and
Emotion, 23(7), 1259–1283. https://doi.org/10.1080/02699930902809375
Russell, J. A., & Barrett, L. F. (1999). Core affect, prototypical emotional episodes, and other
things called emotion: Dissecting the elephant. Journal of Personality and Social
Psychology, 76(5), 805–819. https://doi.org/10.1037/0022-3514.76.5.805
Russell, J. A., Lewicka, M., & Niit, T. (1989). A cross-cultural study of a circumplex model of
affect. Journal of Personality and Social Psychology, 57(5), 848–856.
https://doi.org/10.1037/0022-3514.57.5.848
Saelens, B. E., Sallis, J. F., Frank, L. D., Cain, K. L., Conway, T. L., Chapman, J. E., Slymen, D.
J., & Kerr, J. (2012). Neighborhood Environment and Psychosocial Correlates of Adults’
Physical Activity. Medicine & Science in Sports & Exercise, 44(4), 637–646.
https://doi.org/10.1249/MSS.0b013e318237fe18
Sallis, J. F., Hovell, M. F., Hofstetter, C. R., Elder, J. P., Faucher, P., Spry, V. M., Barrington, E.,
& Hackley, M. (1990). Lifetime history of relapse from exercise. Addictive Behaviors,
15(6), 573–579. https://doi.org/10.1016/0306-4603(90)90059-7
Schmeichel, B. J., & Zell, A. (2007). Trait self-control predicts performance on behavioral tests
of self-control. Journal of Personality, 75(4), 743–755. https://doi.org/10.1111/j.1467-
6494.2007.00455.x
Schultchen, D., Reichenberger, J., Mittl, T., Weh, T. R. M., Smyth, J. M., Blechert, J., &
Pollatos, O. (2019). Bidirectional relationship of stress and affect with physical activity

141

and healthy eating. British Journal of Health Psychology, 24(2), 315–333.
https://doi.org/10.1111/bjhp.12355
Schwerdtfeger, A., Eberhardt, R., Chmitorz, A., & Schaller, E. (2010). Momentary affect
predicts bodily movement in daily life: An ambulatory monitoring study. Journal of
Sport & Exercise Psychology, 32(5), 674–693. https://doi.org/10.1123/jsep.32.5.674
Seymour, R. B., Hughes, S. L., Ory, M. G., Elliot, D. L., Kirby, K. C., Migneault, J., Patrick, H.,
Roll, J. M., & Williams, G. (2010). A Lexicon for Measuring Maintenance of Behavior
Change. American Journal of Health Behavior, 34(6), 660–668.
Shadur, J. M., Hussong, A. M., & Haroon, M. (2015). Negative affect variability and adolescent
self-medication: The role of the peer context. Drug and Alcohol Review, 34(6), 571–580.
https://doi.org/10.1111/dar.12260
Sheeran, P., Gollwitzer, P. M., & Bargh, J. A. (2013). Nonconscious processes and health.
Health Psychology, 32(5), 460–473. https://doi.org/10.1037/a0029203
Sheeran, P., Maki, A., Montanaro, E., Avishai-Yitshak, A., Bryan, A., Klein, W. M. P., Miles,
E., & Rothman, A. J. (2016). The impact of changing attitudes, norms, and self-efficacy
on health-related intentions and behavior: A meta-analysis. Health Psychology, 35, 1178–
1188. https://doi.org/10.1037/hea0000387
Sheeran, P., Webb, T. L., Gollwitzer, P. M., & Oettingen, G. (2018). Self-Regulation of Affect–
Health Behavior Relations. In Affective Determinants of Health Behavior. Oxford
University Press. https://doi.org/10.1093/oso/9780190499037.003.0005
Shiffman, S., Stone, A. A., & Hufford, M. R. (2008). Ecological Momentary Assessment.
Annual Review of Clinical Psychology, 4(1), 1–32.
https://doi.org/10.1146/annurev.clinpsy.3.022806.091415
Simkin, L. R., & Gross, A. M. (1994). Assessment of coping with high-risk situations for
exercise relapse among healthy women. Health Psychology: Official Journal of the
Division of Health Psychology, American Psychological Association, 13(3), 274–277.
https://doi.org/10.1037//0278-6133.13.3.274
Simons, J. S., Wills, T. A., & Neal, D. J. (2014). The Many Faces of Affect: A Multilevel Model
of Drinking Frequency/Quantity and Alcohol Dependence Symptoms Among Young
Adults. Journal of Abnormal Psychology, 123(3), 676–694.
https://doi.org/10.1037/a0036926
Solhan, M. B., Trull, T. J., Jahng, S., & Wood, P. K. (2009). Clinical assessment of affective
instability: Comparing EMA indices, questionnaire reports, and retrospective recall.
Psychological Assessment, 21(3), 425–436. https://doi.org/10.1037/a0016869
Stevens, C. J., Baldwin, A. S., Bryan, A. D., Conner, M., Rhodes, R. E., & Williams, D. M.
(2020). Affective Determinants of Physical Activity: A Conceptual Framework and

142

Narrative Review. Frontiers in Psychology, 11.
https://doi.org/10.3389/fpsyg.2020.568331
Stieger, M., Allemand, M., & Lachman, M. E. (2021). Targeting self-control as a behavior
change mechanism to increase physical activity: Study protocol of a randomized
controlled trial. Contemporary Clinical Trials, 100, 106236.
https://doi.org/10.1016/j.cct.2020.106236
Stone, A., Shiffman, S., Atienza, A., & Nebeling, L. (2007). The Science of Real-Time Data
Capture: Self-Reports in Health Research. Oxford University Press.
Stone, A., Shiffman, S. S., & DeVries, M. W. (1999). Ecological momentary assessment. In
Well-being: The foundations of hedonic psychology (pp. 26–39). Russell Sage
Foundation.
Stults-Kolehmainen, M. A., & Sinha, R. (2014). The Effects of Stress on Physical Activity and
Exercise. Sports Medicine, 44(1), 81–121. https://doi.org/10.1007/s40279-013-0090-5
Tangney, J. P., Baumeister, R. F., & Boone, A. L. (2004). High self-control predicts good
adjustment, less pathology, better grades, and interpersonal success. Journal of
Personality, 72(2), 271–324. https://doi.org/10.1111/j.0022-3506.2004.00263.x
Telama, R., Yang, X., Leskinen, E., Kankaanpää, A., Hirvensalo, M., Tammelin, T., Viikari, J.
S. A., & Raitakari, O. T. (2014). Tracking of Physical Activity from Early Childhood
through Youth into Adulthood. Medicine & Science in Sports & Exercise, 46(5), 955–
962. https://doi.org/10.1249/MSS.0000000000000181
Tice, D. M., Bratslavsky, E., & Baumeister, R. F. (2001). Emotional distress regulation takes
precedence over impulse control: If you feel bad, do it! Journal of Personality and Social
Psychology, 80(1), 53–67. https://doi.org/10.1037/0022-3514.80.1.53
Tremblay, M. S., Colley, R. C., Saunders, T. J., Healy, G. N., & Owen, N. (2010). Physiological
and health implications of a sedentary lifestyle. Applied Physiology, Nutrition, and
Metabolism, 35(6), 725–740. https://doi.org/10.1139/H10-079
Truesdale, K. P., Stevens, J., Lewis, C. E., Schreiner, P. J., Loria, C. M., & Cai, J. (2006).
Changes in risk factors for cardiovascular disease by baseline weight status in young
adults who maintain or gain weight over 15 years: The CARDIA study. International
Journal of Obesity, 30(9), 1397–1407. https://doi.org/10.1038/sj.ijo.0803307
Trull, T. J., Solhan, M. B., Tragesser, S. L., Jahng, S., Wood, P. K., Piasecki, T. M., & Watson,
D. (2008). Affective instability: Measuring a core feature of borderline personality
disorder with ecological momentary assessment. Journal of Abnormal Psychology, 117,
647–661. https://doi.org/10.1037/a0012532
Tucker, J. M., Welk, G. J., & Beyler, N. K. (2011). Physical Activity in U.S. Adults: Compliance
with the Physical Activity Guidelines for Americans. American Journal of Preventive
Medicine, 40(4), 454–461. https://doi.org/10.1016/j.amepre.2010.12.016

143

Valle, C. G., Nezami, B. T., & Tate, D. F. (2020). Designing in-app messages to nudge behavior
change: Lessons learned from a weight management app for young adults.
Organizational Behavior and Human Decision Processes, 161, 95–101.
https://doi.org/10.1016/j.obhdp.2020.10.004
Van Dyck, D., Plaete, J., Cardon, G., Crombez, G., & De Bourdeaudhuij, I. (2016). Effectiveness
of the self-regulation eHealth intervention ‘MyPlan1.0.’ on physical activity levels of
recently retired Belgian adults: A randomized controlled trial. Health Education
Research, 31(5), 653–664. https://doi.org/10.1093/her/cyw036
Wang, S., Intille, S., Ponnada, A., Do, B., Rothman, A., & Dunton, G. (2022). Investigating
Microtemporal Processes Underlying Health Behavior Adoption and Maintenance:
Protocol for an Intensive Longitudinal Observational Study. JMIR Research Protocols,
11(7), e36666. https://doi.org/10.2196/36666
Warburton, D. E. R., & Bredin, S. S. D. (2017). Health benefits of physical activity: A
systematic review of current systematic reviews. Current Opinion in Cardiology, 32(5),
541–556. https://doi.org/10.1097/HCO.0000000000000437
Weiss, N. H., Bold, K. W., Contractor, A. A., Sullivan, T. P., Armeli, S., & Tennen, H. (2018).
Trauma exposure and heavy drinking and drug use among college students: Identifying
the roles of negative and positive affect lability in a daily diary study. Addictive
Behaviors, 79, 131–137. https://doi.org/10.1016/j.addbeh.2017.12.015
WHO guidelines on physical activity and sedentary behaviour. (2020).
https://apps.who.int/iris/bitstream/handle/10665/336656/9789240015128-
eng.pdf?sequence=1&isAllowed=y
Wichers, M., Peeters, F., Rutten, B. P. F., Jacobs, N., Derom, C., Thiery, E., Delespaul, P., & van
Os, J. (2012). A time-lagged momentary assessment study on daily life physical activity
and affect. Health Psychology, 31(2), 135–144. https://doi.org/10.1037/a0025688
Williams, D. M., & Evans, D. R. (2014). Current Emotion Research in Health Behavior Science.
Emotion Review, 6(3), 277–287. https://doi.org/10.1177/1754073914523052
Williams, D. M., Rhodes, R. E., & Conner, M. T. (2018). Affective Determinants of Health
Behavior. Oxford University Press.
Williams, D. M., Rhodes, R. E., & Conner, M. T. (2019). Conceptualizing and intervening on
affective determinants of health behaviour. Psychology & Health, 34(11), 1267–1281.
https://doi.org/10.1080/08870446.2019.1675659
Williamson, D. F. (1990). The 10-Year Incidence of Overweight and Major Weight Gain in US
Adults. Archives of Internal Medicine, 150(3), 665.
https://doi.org/10.1001/archinte.1990.00390150135026

144

Williamson, D. F., Kahn, H. S., Remington, P. L., & Anda, R. F. (1990). The 10-year incidence
of overweight and major weight gain in US adults. Archives of Internal Medicine, 150(3),
665–672.
Wipfli, B. M., Rethorst, C. D., & Landers, D. M. (2008). The Anxiolytic Effects of Exercise: A
Meta-Analysis of Randomized Trials and Dose–Response Analysis. Journal of Sport and
Exercise Psychology, 30(4), 392–410. https://doi.org/10.1123/jsep.30.4.392
Yi, L., Wilson, J. P., Mason, T. B., Habre, R., Wang, S., & Dunton, G. F. (2019). Methodologies
for assessing contextual exposure to the built environment in physical activity studies: A
systematic review. Health & Place, 60, 102226.
https://doi.org/10.1016/j.healthplace.2019.102226
Zack, M., Woodford, T. M., Tremblay, A. M., Steinberg, L., Zawertailo, L. A., & Busto, U. E.
(2011). Stress and Alcohol Cues Exert Conjoint Effects on Go and Stop Signal
Responding in Male Problem Drinkers. Neuropsychopharmacology, 36(2), Article 2.
https://doi.org/10.1038/npp.2010.177
Zheng, Y., Manson, J. E., Yuan, C., Liang, M. H., Grodstein, F., Stampfer, M. J., Willett, W. C.,
& Hu, F. B. (2017). Associations of Weight Gain From Early to Middle Adulthood With
Major Health Outcomes Later in Life. JAMA, 318(3), 255.
https://doi.org/10.1001/jama.2017.7092
Zick, C. D., Smith, K. R., Brown, B. B., Fan, J. X., & Kowaleski-Jones, L. (2007). Physical
Activity During the Transition from Adolescence to Adulthood. Journal of Physical
Activity and Health, 4(2), 125–137. https://doi.org/10.1123/jpah.4.2.125
Zimmermann, P., & Iwanski, A. (2014). Emotion regulation from early adolescence to emerging
adulthood and middle adulthood: Age differences, gender differences, and emotion-
specific developmental variations. International Journal of Behavioral Development,
38(2), 182–194. https://doi.org/10.1177/0165025413515405

Asset Metadata

Creator Do, Bridgette (author)

Core Title Investigating the associations of affective variability and physical activity among young adults using ecological momentary assessment

Contributor Electronically uploaded by the author (provenance)

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Preventive Medicine

Degree Conferral Date 2023-05

Publication Date 04/27/2025

Defense Date 03/27/2023

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

Tag affect,affective states,digital health,ecological momentary assessment,Exercise,OAI-PMH Harvest,physical activity

Format theses (aat)

Language English

Advisor Dunton, Genevieve F. (committee chair), Belcher, Britni R. (committee member), Hedeker, Donald (committee member), Mason, Tyler B. (committee member), Miller, Kimberly A. (committee member)

Creator Email bridgetd@usc.edu,bridgettemdo@gmail.com

Unique identifier UC113088947

Identifier etd-DoBridgett-11728.pdf (filename)

Legacy Identifier etd-DoBridgett-11728

Document Type Dissertation

Format theses (aat)

Rights Do, Bridgette

Internet Media Type application/pdf

Type texts

Source 20230427-usctheses-batch-1032 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

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

Repository Email uscdl@usc.edu

Abstract (if available)

Abstract This dissertation is comprised of three unique studies that leveraged real-time data capture methods (i.e., ecological momentary assessment [EMA]) and novel statistical modeling strategies to examine the influence of affective variability (i.e., fluctuations in affect intensity) on physical activity among young adults. The overarching objective of this dissertation was to increase our scientific understanding on the potential associations between affective variability (i.e., positive-activated, positive-deactivated, negative-activated, negative-deactivated affect) and physical activity among young adults by coupling smartphone-based EMA with smartwatch technology. The specific aims of this dissertation were to (1) determine whether subject-level affective variability is associated with overall levels of physical activity, (2) determine whether trait self-control moderates the associations between subject-level variability and overall levels of physical activity, (3) investigate the day-level associations between affective variability and physical activity, and (4) examine whether the subject-level association between momentary affect and subsequent physical activity (e.g., next 30 minutes) predicts future physical activity levels. Findings suggest that (1) subject-level variability in negative-deactivated affect was associated with greater overall levels of physical activity, while variability in positive-activated, positive-deactivated, and negative-activated affect were not associated with overall levels of physical activity, (2) trait self-control did not moderate the associations between subject-level affective variability and overall physical activity, (3) greater day-level variability in positive-activated affect was associated with greater physical activity on that same day compared to other days, whereas greater day-level variability in negative-deactivated affect was associated with less physical activity on that same day compared to other days, and (4) the strength of association between momentary affect and subsequent physical activity (i.e., 30 minutes later) did not predict future daily physical activity levels one month later. Taken together, the findings highlight how variability in different affective states may be differently associated with physical activity depending on the affective valence, arousal/activation state, and the temporality of the associations examined. In addition, the dissertation underscores the importance of assessing affective variability above and beyond mean or momentary levels. Future research can expand upon this work by elucidating mechanisms underlying the associations of affective variability and physical activity, considering important mental health characteristics, and developing EMA studies to further explore the temporal relationships at the day-level and longitudinally. Overall, this dissertation provides a foundation for future studies to unique role that affective variability may play in physical activity engagement both within- and between-subjects. Ultimately, knowledge on affect dynamics and momentary affective processes in free-living situations can help optimize future intervention strategies by targeting fluctuations in affect to help increase physical activity among young adults.