University of Southern California Dissertations and Theses

The formation and influence of online health social networks on social support, self-tracking behavior and weight loss outcomes

(USC Thesis Other)

The formation and influence of online health social networks on social support, self-tracking behavior and weight loss outcomes

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

THE FORMATION AND INFLUENCE OF ONLINE HEALTH SOCIAL NETWORKS ON
SOCIAL SUPPORT, SELF-TRACKING BEHAVIOR AND WEIGHT LOSS OUTCOMES

By
Jingbo Meng

August 2014

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfilment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(COMMUNICATION)
Jingbo Meng Copyright 2014

Dedication
To my family.
To all the great mentors over the years.

i

Acknowledgements
This dissertation is driven by my research interests across health communication,
organizational communication and social networks. The completion of this project would not
have been possible without all the help and support from many people. First of all, I would like
to gratefully and sincerely thank my advisor Dr. Margaret McLaughlin for her unremitting
support, guidance and understanding during my graduate studies at Annenberg School. She did
not only teach me how to conduct scientific research, but also encouraged me to pursue the
implications of scholarly work in public service.
I also wish to extend my thanks and appreciation to my other committee members. I am
indebted and grateful to Dr. Janet Fulk for her guidance on projects over my years at Annenberg.
Janet inspired my research interest in teams and I have learned a great deal from her on
multilevel research. Dr. Peter Monge inspired my research interest in social networks and I have
profited greatly from his methods classes. Dr. Maryalice Jordan-Marsh always provided valuable
suggestions from an interdisciplinary perspective of health care and health technologies. In
addition to committee members, I want to thank Dr. Dimitri Williams and Dr. Andrea
Hollingshead, with whom I have collaborated on different projects. Those experiences all
influence my dissertation in substantive ways.
I would like to thank my Annenberg colleagues for providing wonderful companionship,
social support, and feedback on my projects over the years at Annenberg. My thanks extend to
Young Ji Kim, Amanda Beacom, Nancy Chen, Zheng An, Wenlin Liu, Jinghui Hou, Lu Li Rong
Wang and Yujung Nam. I also want to thank Annenberg alumni, including Cindy Cuihua Shen,
Helen Hua Wang, Katya Ognyanova, Drew Margolin, Robby Ratan and Wei Peng for their great
support.
ii

Finally, I must acknowledge the incredible love and support provided by my husband,
Mi, who has spent countless hours discussing and critiquing research ideas and methodologies.
He always stood by me with his persistent patience, positivity and unconditional love. Thank
you.

iii

Table of Contents

Dedication

i
Acknowledgements ii
List of Tables

vi
List of Figures

vii
Abbreviations

viii
Abstract

ix
Chapter 1 Introduction 1
Chapter 2 Understanding Online Health Social Networks 10

Current Research on OHSNs 10

Services Provided by OHSNs 15
Chapter 3 Study I: OHSNs and Individual Health Outcomes 22

Multi-Theoretical Multilevel Framework 22

Formation Mechanisms of Online Health Buddy Networks 26

Multi-dimensionality in homophily 29

The paradox of social exchange 38

Endogenous factors 41

Social Influence of Online Health Buddy Networks 43
Chapter 4 Study II: Ego Network Structure, Social Support and Self-Tracking
Behavior
47
Social Support and Social Networks

Ego Network and Perceived Social Support

Triadic Closure in Ego Networks

Perceived Social Support and Self-Tracking Behavior 61
Self-Tracking Behavior and Health Progress 66
Chapter 5 Methods 70
Test Bed: FatSecret 71
Data Collection 74
Data Cleaning 79
Measurement 80

Measures from Web Crawling Data 80
iv

Measures in Online Survey 83

Measures in Content Coding 85
Analysis 86
Chapter 6 Results 89

Study I 89

Preliminary Analysis 89

Hypotheses Testing 93
Study II

Preliminary Analysis 98

Hypotheses Testing 101

Post-hoc Analysis 104
Chapter 7 Discussion and Conclusion 110
Discussion 110

Study I 111

Study II 118
Limitations and Future Research 121
Conclusion 125
References

128

v

List of Tables
Table 1. A Summary of Hypotheses and Research Questions 69
Table 2. Timeline of Data Collection and Variables Summary 78
Table 3. Descriptive Statistics of Self-Tracking, Groups, Health Progress and Weight
Outcome across Four Time Points of Data Collection (N = 709) 90
Table 4. Descriptive Statistics of Degree, Density and the Network Change (N = 709) 93
Table 5. Results of SIENA Analysis for Study I 97
Table 6. Descriptive Statistics of Survey Sample (N = 357) 99
Table 7. Zero-Correlation of Variables 100
Table 8. A Summary of SEM Model Comparisons 107
Table 9. A Summary of Results of Hypothesis Testing and Research Questions 109

vi

List of Figures

Figure 1. Social Networking in OHSNs 16
Figure 2. Quantified Self-Tracking in OHSNs 19
Figure 3. Triadic Closure in Sociometric Networks 56
Figure 4. Triadic Closure in Egocentric Networks 57
Figure 5. Hypothesized Model in Study II 68
Figure 6. An Example of Weight Tracking on FatSecret 73
Figure 7. Degree Distribution for the Final Subsample (N = 709) 91
Figure 8. Visualization of the Network Using Gelphi (N = 709) 92
Figure 9. A Description of Difference between Start Weight and Goal Weight (N = 305) 99
Figure 10. Parameter Estimation for the Hypothesized Model 102
Figure 11. Parameter Estimation for Final Modified Model 108

vii

Abbreviations

CMC Computer-Mediated Communication

CHESS Comprehensive Health Enhancement Support System
CMSS Computer-Mediated Social Support

ITCs Information and Communication Technologies
MTML Multi-Theoretical Multilevel

MMORPGs Massively Multiplayer Online Role Playing Games
OHSNs Online Health Social Networks

OSGs Online Support Groups

PHRs Personal Health Records

TTM The Transtheoretical Model

SINEA Simulation Investigation for Empirical Network Analysis

SNSs Social Network Sites

viii

Abstract
The current dissertation provides an examination of online health social networks in
intentionally designed health-related social media. It investigated the theoretical mechanisms that
drive the formation of online health social networks, examined the joint dynamics of network
selection and network influence on individual health outcomes, and tested a model of ego
network structure, social support and self-tracking behavior. The dissertation is situated in an
entrepreneurial online weight management social networking site, FatSecret. It employed a four-
month longitudinal study design, and collected unobtrusive behavioral data extracted from the
site and self-reported data from an online survey with the users of FatSecret.
Drawing on MTML framework, the first study found that demographic homophily
including age and gender similarity, inbreeding homophily in the form of group affiliation, and
health-related homophily including initial health status and health progress significantly
predicted the selection of online health buddies. Similarity in health goal was not a significant
predictor due to its strong correlation with initial health status in the specific health issue of
weight loss. A person’s frequency of updating personal health information was marginally
significant in predicting tie formation. By conducting a SIENA analysis, the study also found
empirical support for a social influence effect among health buddies, such that an individual’s
weight outcome tended to become similar to the average of the individual’s health buddies’
weight outcomes. The second study drew from literature on structural and functional social
support, the buffering model of social support and self-regulation depletion studies. By
conducting SEM analyses, the study found that both the size and triadic closure of an
individual’s ego network predicted perceived social support for weight loss in FatSecret. Then,
perceived social support predicted more active self-tracking behavior, and self-tracking behavior
predicted improved health progress. Post hoc analysis showed that there was a negative and
ix

significant effect of health progress on self-tracking behavior, such that improved health progress
at an earlier time point would reduce the amount of self-tracking at a later time point. Finally, the
implications of designing effective web-based weight loss interventions by better organizing and
engineering peer-to-peer social networks were discussed.

Keywords: Online health social networks, Homophily, Personal health information, Social
influence, Self-tracking, Social support
x

CHAPTER 1
Introduction
The Internet has become an indispensable part of the health communication landscape.
About 72% of adult American Internet users searched health information online in the past year
(Fox & Duggan, 2013). In particular, social media are playing an increasingly prominent role in
public health. Among those online health seekers, 35% have read or watched someone else’s
experience about health issues posted and shared by other social media users (Fox & Duggan,
2013). A large-scale survey shows that 32% of Americans have used SNS for health-related
activities (Thackeray, Crookston, & West, 2013). From health information seeking via Twitter
and Facebook (Hawn, 2009) to online buddy networks to support coaching and counseling for
smoking cessation and weight loss, social media provide a valuable resource for consumers of
health services.
The increasing popularity of health applications of the Internet in general and social
media specifically is driven by a revolution in the current health care system from a reactive and
“disease-centered” to a proactive and “patient-centered” medical approach (Lejbkowicz, Caspi &
Miller, 2012). Coined as personalized and participatory healthcare, this change in the health care
system aims to leverage advanced healthcare technologies, and encourage patients to become full
partners rather than passive recipients in the healthcare decision making process (Hood &
Auffray, 2012). This personalized and participatory healthcare model helps to reduce medical
visits, hospital stays and healthcare costs overall (Sohn, Helms, Pelleter et al., 2012). More
importantly, health consumers become more empowered and efficacious to perform activities to
achieve their health goals (Oh & Lee, 2011). At the same time, empowered health consumers
tend to take a proactive role and more actively engage in doctor-patient communication (Oh &
Lee, 2011). As a reciprocal process, the increasingly public use of social media technologies and
1

the revolution of personalized and participatory healthcare systems reinforce and support each
other.
Given the importance and popularity of social media applications in healthcare, this
dissertation aims to contribute to the current literature in the broad area of social media and
health. In particular, this dissertation is inspired and motivated by the juxtaposition of two
primary seminal works in social networks and health. The first ground breaking work was
conducted by medical sociologists and network scientists Nicholas Christakis and James Fowler
(2009): Connected, the Surprising Power of Our Social Networks and How They Shape Our
Lives. The book presents the results of a series of studies on social influence of people’s real-life
social networks. In general, those studies show that our world is governed by the three degree of
influence rule: we influence and are influenced by people up to three degrees away from us.
Specifically in health-related social influence, the authors show that both positive and negative
emotions are contagious and both healthy and unhealthy behaviors can spread through our
contacts in society. For example, your friend’s friend’s friend has more impact on how much
smiling you do and how happy you are than $5,000 in your pocket; similarly, your friend’s
friend’s friend can make you fat – or thin. The three degree influence in social networks
underlies unhealthy behaviors such as eating disorders, smoking, substance abuse, and suicide
clusters, as well as positive and pro-social behaviors such as happiness and altruism.
Christakis-Fowler’s three degree influence model stimulates thousands of studies in
investigating peer influence on health behaviors and outcomes. In a natural experiment
(Yakusheva, Kapinos, & Eisenberg, 2014), first-year college students were assigned roommates
based on the heterogeneity of their weights, while keeping other physical and socio-demographic
factors equivalent. After one year, the researchers found that there was indeed peer influence on
2

college student’s weight status. The influence appeared to be heterogeneous as heavier and
thinner students were affected by roommates more than average-weight students. Another series
of studies examined peer influence in adolescents’ friendship networks. For instance, by
analyzing 50 cohort networks with 13,214 individuals, one study found that while controlling for
the tendency of selection of homophilous friends, adolescents became more similar in their
alcohol consumption behavior (Osgood, Ragan, Wallace, Gest, Feinberg & Moody, 2012).
While these works focus on social influence of real-life social networks on individual
health behaviors and outcomes, the other seminal work examined the spread of health behaviors
in online social networks (Centola, 2010, 2011). Unlike naturally occurring real-life social
networks such as family, friends, co-workers and neighbors, these online social networks
consisted of strangers who were anonymous to one another. As the goal of the study was to
investigate the effects of different structures of online social networks in spreading health
behaviors, Centola (2010) deliberately kept a strict anonymity of participants in his experiment.
In an artificially created online community, each participant had an anonymous online profile,
and the researcher matched them with other participants referred to as “health buddies”. The
researcher manipulated the rule of buddy matching so that different network structures were
generated. The result of the study revealed that different network structures had varying social
influence on spreading health behaviors. The conclusion was that the clustered network structure
was more effective in the diffusion of healthy behavior. In another experiment, Centola (2011)
manipulated individual attributes in an online health network. What he found was that putting a
healthy seed into an existing online health social network is effective in spreading positive health
outcomes, assuming that the healthy seed shared some similar socio-demographic traits with the
members in that network.
3

Centola’s works (2010, 2011) are certainly valuable to understanding the effective
network structure in spreading healthy behaviors and outcomes. However, the work also raises a
critical question: how do individuals choose their health buddies and form buddy networks in a
naturalistic online health community? Unlike real-life social networks in Christakis and Fowler’s
studies which are not formed with a purpose for health promotion, or online social networks in
Centola’s studies which are manipulated into specific structures, naturally occurring online
health buddy networks are created with a goal of improved health (Welbourne, Blanchard, &
Wadsworth, 2013). However, we have little knowledge about the ways in which online health
social networks form and develop structure. This is a basic yet critical research question in that it
lays the foundation for us to investigate and understand the effects of online health social
networks. According to Christakis and Fowler (2009), social networks are neutral in terms of
health implications; they are powerful but only channels that help to transmit and spread health
behaviors and outcomes. It is whom you are connected to that would impact the kinds of social
influence exerted through social networks. Without knowing the ways that individuals form
online health social networks, the conclusion of social influence of online health social networks
is conceptually incomplete and methodologically flawed.
To date, only a couple of studies examined social contagion of health-related outcomes in
online health social networks (e.g., Ma, Chen & Xiao, 2010), and no published work examined
the mechanisms that drive the formation of online health social networks
1
. Most scholarly works
on social media and health focus on using social media as a tool for health interventions. For
example, Greene, Sacks, Piniewski et al. (2012) conducted a randomized controlled study to test
the effectiveness of an online social network – iWell OSN – in increasing participants’ physical
1
Except that computer science scientists did some work on community detection in online networks with diabetes
(Chomutare, Arsand & Hartvigsen, 2013)
4

activity and weight outcomes. McLaughlin, Nam, Gould et al. (2011) designed and implemented
a videosharing social networking intervention for young adult cancer survivors to exchange
social support with one another. A more recent intervention involves the use of Facebook as a
platform to deliver the intervention to groups and combines mobile applications to promote
healthy diet and physical activity (Patrick, Marshall, Fowler, et al., 2014). These works either use
online social networks as existing infrastructure to deliver intervention messages, or include
online social networks as a component in their intervention designs. While these works are
valuable, more scholarly attention is needed to directly examine the patterns, characteristics and
effects of online health social networks.
The current dissertation represents such an effort to investigate the underlying
mechanisms that drive the formation of online health social networks, and empirically test the
social influence of online health social networks on individual health outcomes in a naturalistic
setting. To achieve this goal, the first study of the dissertation builds on the Multi-theoretical
Multilevel Framework (Monge & Contractor, 2003) and draws from literature on motivations to
participate in online health social networks to derive hypotheses about the ways that individuals
select their online health partners or buddies. In the meanwhile, thanks to the advancement of
social network analysis techniques, this study performs an empirical test of the joint dynamics of
network selection and social influence simultaneously (Snijders, van de Bunt, Steglich, 2010).
The first study of my dissertation contributes to the existing literature on social media and health
in that it takes a relational perspective and a network approach to study individual networking
behaviors and the corresponding health consequences. Rather than simply treating social
networks in social media as a tool for health promotion, this study provides a critical and direct
5

examination of online health social networks. The findings will shed light on the real value of
online health social networks in enhancing public health.
An additional interest of the current dissertation involves a key technological innovation:
wearable devices for personalized and participatory healthcare. The advent of wearable devices
has given a rapid rise to self-tracking online health communities wherein health conscious
consumers can generate, store and analyze information about their health conditions and statuses
(Paton, et al., 2012). Although the technological feature of self-tracking has just recently been
incorporated in health-related social media, the idea of self-tracking is not new. Prior to the age
of the Internet or computers, patients were encouraged by doctors to keep a diary of their daily
food intake, exercise activities and other health conditions and behaviors. Self-tracking behavior
is very helpful in monitoring and understanding one’s health experiences. The majority of studies
use it as an independent variable to investigate the effectiveness of self-tracking behavior in
promoting physical activity and fostering more effective self-management for diabetic patients
(McCorkle, Ercolano, Lazenby et al., 2011). Randomized trials and other empirical studies
consistently revealed that self-tracking can positively influence patients’ health-related self-
efficacy and lead to more favorable health outcomes (Olander, Fletcher, Williams, Atkinson,
Turner & Frech, 2013).
The second study of the current dissertation represents an effort to investigate the effect
of ego network structures in influencing individual self-tracking behaviors, and their impacts on
ultimate individual health outcomes. Different from previous works, this study treats self-
tracking as both an independent and a dependent variable at different processes of the theoretical
model. The finding will contribute to existing literature by providing additional empirical
evidence of the effect of self-tracking on health outcomes. More importantly, its contribution lies
6

in a discovery of the mechanism in which adherence to self-tracking behavior can be enhanced
through self-organized personal online health social networks.
Overall, this dissertation is situated in entrepreneurial online heath social networks for
weight management. Entrepreneurial online health social networks are early adopters of
innovative technologies for health care. They provide a lower barrier to entry and exit, and a
more persistent virtual community than many online health communities designed and
implemented by health researchers and professionals. Funded health intervention projects have
experienced difficulties in continuing operating and scaling up due to the termination of funding.
As a result, participants enrolled in those funded health intervention programs, which may be
very successful, have to quit the service. In contrast, popular entrepreneurial online health
communities rely on their own business models to self-sustain in the market. They are more
convenient and accessible to general health consumers anywhere and anytime as long as they
have a computer and the Internet. However, the design of these popular entrepreneurial online
health social networks is not driven by theories or typically moderated by medical experts.
Patient participation in those sites may be ad hoc and highly dependent on self-help and self-
organizing. The context of entrepreneurial sites even highlights the importance and value of the
current dissertation, whose purpose is to study the organizing logics of online health buddy
networks and their effects on healthy behavior (i.e., self-tracking) and health outcome (i.e.,
weight loss). I believe this dissertation represents a unique niche in the subject of social media
and health.
Chapter Summaries
This dissertation is organized as follows. Chapter 2 introduces online health social
networks. It provides conceptualizations of social media and online health social networks,
7

focusing on several featured services provided by online health social networks. Then, it reviews
current literature on online support groups. After building the basic understanding of online
health social networks, Chapter 3 proposes the theoretical framework and hypotheses in the first
study. It begins with introducing the Multi-theoretical Multilevel framework to study the
generating mechanisms of online health social networks. Having been mainly used in
organizational networks and virtual team sciences, the Multi-theoretical Multilevel framework is
justified as an appropriate theoretical framework in the present context. By comparing online
health social networks and teams in virtual worlds, and drawing from literature on motivations to
participate in online health communities, two theoretical perspectives – homophily and social
exchange theory – were used to derive hypotheses about individual networking behaviors.
Specifically, a new dimension of health homophily is proposed to add into demographic and
inbreeding homophily in the specific context of online health community. Then, the chapter
reviews studies of social influence via social networks in both online and offline settings, and
proposes a test of joint dynamics of network selection and network influence on individual
weight loss outcomes.
Chapter 4 proposes the theoretical model and hypotheses in the second study. It begins
with an introduction of structural and functional social support, and attempts to disentangle the
relationship between structural support in the form of ego network structure and functional
support in the form of perceived social support. By conceptualizing self-tracking as a type of
self-regulation behavior, it then explicates the positive role of perceived social support in
enhancing self-tracking behavior by buffering psychological stress.
Chapter 5 describes the details of the methods employed in this dissertation. It includes
selection and description of the test bed, data collection process, sampling strategies, data
8

cleaning steps, measurement in an online survey, data transformation of extracted web data, as
well as data analysis. Chapter 6 provides results for each hypothesis and research question
proposed in both studies. Finally, Chapter 7 includes a discussion of the findings and the
contributions of those findings to theories and existing literature. Limitations and future research
are also discussed.

9

CHAPTER 2
Understanding Online Health Social Networks
Current Research of Online Health Social Networks
Social Media and Health. Social media is “a group of Internet-based applications that
build on the ideological and technological foundations of Web 2.0, and that allow the creation
and exchange of User Generated Content” (p. 61, Kaplan & Haenlein, 2010). The essence of
Web 2.0 ideology and technology is that content and applications on the Internet are no longer
created by individuals, but are constantly published and modified by all ordinary users
collectively (Kaplan & Haenlein, 2010). Social media have various Web-based applications,
including wikis, forums and message boards, review and opinion sites, social network sites,
blogs and microblogs, virtual game worlds, bookmarking, and media sharing sites (Kaplan &
Haenlein, 2010; Neiger, Thackeray & Wagenen et al., 2012). Each category of social media has
corresponding health applications (Korda & Itani, 2011). The convergence of different social
media forms on one web service is a more prominent trend (Korda & Itani, 2011; Orizio, Schulz,
Gasparotti et al., 2010). For example, in an online health community, it is typical now to have
wikis, message boards, blogs, review and social networking features integrated on one platform.
On such a platform, patients can retrieve health information from wikis, publish their own blogs
about disease experiences, review each other’s health tips, ask and answer questions on message
boards and network with one another by becoming friends or health buddies.
Online health social networks inhabit two primary types of social media: open social
media and intentionally designed health social media (Centola, 2013). Open social media are
“large-scale virtual communication infrastructure that are designed for social interactions across
many substantive domains” (p. 2136, Centola, 2013). Examples of open social media are
Facebook and Twitter, which are not intentionally designed for health-related interactions. Yet,
10

health consumers can make use of such platforms for their particular interests by creating special
interest groups, pages and channels. Recent work on open social media and health has provided
important insights into the dynamics of information dissemination and opinion propagation on
health topics. For instance, studies have shown Facebook and Twitter are used widely to educate
patients about cancer and diabetes (De la Torre-Díez, et al., 2012; Shaw & Johnson, 2011).
Other studies pointed out that sentiments about vaccines can be propagated on Twitter through
Retweets (Love, Himelboim, Holton, & Stewart, 2013). Large-scale open social media may offer
an unprecedented opportunity to access public sentiments and attitudes towards health issues.
However, they do not provide context-specific observations for examining health-related
networking behaviors and effects. Therefore, I need to turn to intentionally designed health
social media.
Intentionally designed health social media are designed with a clear goal of promoting
health by aggregating individuals with similar health interests (Centola, 2013). These social
networks can serve a variety of health goals, ranging from counseling for smoking cessation,
support for stress management to encouraging and monitoring exercise and diet routines.
Examples of intentionally designed health social media are PatientsLikeMe, QuitNet and
FatSecret. PatientsLikeMe is a site that was originally created for patients with Amyotrophic
lateral sclerosis (ALS). PatientsLikeMe provides a space where ALS patients can share their
symptoms, conditions and medications through patient networks. QuitNet is a site for smokers
who intend to quit smoking and prevent relapse. FatSecret is a site for people who are trying to
diet and lose weight. Both QuitNet and FatSecret allow for social networking so that their users
can acquire health buddies to quit smoking or lose weight together.
11

For a researcher who is interested in the social dynamics of health, intentionally designed
health social media provide platforms to examine the relationships between individual health
behaviors (such as exercise frequency per week), networking behaviors (such as with whom to
become friends or health buddies), and health outcomes (such as quitting smoking, or losing
weight). Many of these sites are equipped with self-tracking tools or technically support
uploading digitally recorded real-time personal health records. They usually provide social
networking features to connect their users for information sharing and social interaction. The
data extracted from these sites can be very useful in understanding the associations between
people’s online social networks, commitment to self-monitoring behavior and changes in their
health outcomes.
Online support groups. One major stream of research on intentionally designed health
social media involves studies of online support groups (OSGs). There are two types of OSGs.
One is the formal computer-mediated social support (CMSS) group, which often refers to
intervention groups with closed membership enrollment, fixed duration and medical experts
involved as leaders in group communication (Rains & Young, 2009). Those CMSS groups are
typically designed and organized by health professionals and researchers. They usually include
an educational component and supportive communication system. Researchers in the fields of
public health, medicine and healthcare conduct extensive research to test the effectiveness of
CMSS intervention groups for various health concerns. These studies mostly use randomized
control design methodology (e.g., Joseph et al., 2013; Jones et al., 2008). The other type of OSG
is a self-help network, which refers to self-organized online health networks without a clear
boundary or medical experts’ moderation. There are an increasing number of studies focusing on
this type of self-organized OSG (e.g., Chuang & Yang, 2012; Chung, 2013).
12

Although the majority of the research efforts concentrate on using OSGs as a tool for the
delivery of public health services and interventions, communication scholars view OSGs as peer-
to-peer social support communities. From a perspective of communication in OSGs, there are
three primary lines of study. The first focuses on the content of social support messages
exchanged between community members (e.g., Chuang & Yang, 2012). Sillence (2013) collected
messages over a one-month period from an online breast cancer support forum. She analyzed
messages for advice exchange in particular, and found that messages presenting problem
disclosure and personal experience were major ways of advice solicitation. Coursaris and Liu
(2009) randomly collected messages from a HIV/AIDS self-help group. The authors found that
information support and emotional support were exchanged most frequently. Group interactions
facilitated social support exchanges as ways of sharing personal experience and expression of
gratitude.
The second line of research involves members’ participation and engagement in OSGs
(e.g., Ye, 2013). The central theme of these studies is to identify factors that influence and could
potentially increase members’ participation in OSGs. Several major factors discovered so far are
individual socio-demographic variables, offline social support resources, and characteristics of
supportive messages posted in CMSS groups. McLaughlin et al. (2012) designed and
implemented a web-based social networking and videosharing intervention program tailored for
young adult cancer survivors to share social support with one another. The authors found that
participants with little social support from friends and family participated more in the online
intervention program. Similarly, Han et al. (2011) studied user participation in the
Comprehensive Health Enhancement Support System (CHESS) designed by communication
scholars and health professionals. The study found that lurkers in an online breast cancer
13

community had a significantly greater level of offline social support than posters. Other
individual attributes, including age and information competence, also played a role in the level of
members’ participation in the community (Hu, Bell, Kravitz & Orrange, 2012). Somewhat
differently, Pfeil, Zaphiris and Wilson (2010) examined the sequence of social support messages,
and concluded that message sequence played an important role in sustaining online support
communities. Specifically, they found that mutual exchange of personal information and
receiving support after talking about personal problems were basic components for community
members to engage in continued social support.
The third line of research focuses on the effect of participation in OSGs. Studies have
examined the role of insightful and emotional disclosure on outcome variables such as health
self-efficacy, emotional well-being and functional well-being in CHESS (Shaw et al., 2007;
Shim, Cappella & Han, 2011), the effect of activities in OSGs on patient empowerment and
doctor-patient communication (Barak, Boniel-Nissim & Suler, 2008; Bartlett & Coulson, 2011;
Mo & Coulson, 2010; Oh & Lee, 2011), and the relationships between specific usage of OSG
features and perceived information and emotional support (Chung, 2013). More recent studies on
CHESS investigated the processes through which participation in CMSS groups contributes to
the above-mentioned outcome variables. For example, Namkoong et al. (2012) found that
participation in CHESS enhanced users’ well-being through the mediation of an increased degree
of perceived bonding.
Among the studies described above, the majority studied formal CMSS groups. Even
though formal CMSS groups are important venues for health intervention, self-help or self-
organized groups found on online health networks such as Daily Strength and PatientsLikeMe
deserve more scholarly attention than they have currently received. These informal self-
14

organized online health social networks can attract more general health consumers and patients
than formal CMSS intervention groups (Fox, 2011). Without the lead of health experts as in
formal CMSS intervention groups, the process and effect of social support in informal self-
organized CMSS groups is in need of more systematic study.
As we are now clearer about what has been studied in social media and health, the
following sections will discuss featured services that online health social networks provide as a
basic background to understand the scope of the present dissertation. Then, I will define online
health social networks in this specific study, and explicate how the concept has been treated in
previous research.
Services Provided by Online Health Social Networks
Information and emotional support. The basic service that the majority of online health
social networks offer are information and emotional support provided by registered site users.
Information support comes in the form of advice, factual input, and feedback (Walther & boyd,
2002). Such messages may discuss disease symptoms, conditions, treatments and medications,
etc. Information support helps people to learn about medical procedures (Wicks et al., 2010),
make decisions about medical visits (Hu et al., 2012) and make choices of physicians and
medications etc. (Wicks et al., 2010). Emotional support comes through “expressions of caring,
concern, empathy, and sympathy” (Walther & boyd, 2002). Such messages are less about
knowledge or decision making, but more about psychological well-being, stress release and
encouragement. Previous studies have mostly concentrated on the exchange of information and
emotional support messages on message boards. They show that information and emotional
support are frequently found in both open social media (e.g., Frohlich & Zmyslinski-Seelig,
2012) and intentionally designed health social media (Sillence, 2013).
15

Social networking: Health buddy. The popularity of open social media sites like
Facebook has given rise to the proliferation of social networking functions in various social
media platforms. Social networking features are characterized by (1) a public or semi-public
profile within a bounded system, (2) a list of users with whom a person is connected to, and (3)
the capability to view and traverse lists of connections made by others within the bounded
system (boyd & Ellison, 2007; 2012). Intentionally designed health social media has quickly
picked up the social networking feature, and now users are able to create profiles and become
friends or health buddies with one another (Paton et al; 2012). Just like on Facebook, members of
an OHSN can add other members as friends, or health buddies, through the process of “friend
requests”. The connection will be established if two relational parties confirm the friendship or
health buddy relationship (Figure 1). Once two members are connected, they can then view each
other’s full profiles, receive news updates and more. By incorporating social networking
features, OHSNs enable their members to “gather, learn from, and give support to others in
need” (Boulos & Wheelert, 2007).
Figure 1. Social Networking in OHSNs

16

The incorporation of social networking features has made OHSNs different from OSGs,
which are mostly in the form of message boards. Because of the presence of profiles and friends’
news feeds, the virtual environment in an OHSN is less anonymous than in a purely text-based
online discussion forum. By browsing other members’ profiles, more social information is
visible and therefore it’s possible to build more direct, meaningful and deeper peer relationships
(Phua, 2010). Peer-to-peer interaction has a long history in health care, back to its inception of
offline social support groups for alcohol abstention and weight control (Kiesler, 1985). With
social networking features, online health buddy relationships may resemble offline social support
relationships to a greater extent than message exchange relationships in a typical OSG. Recent
web-based social support services have adopted peer-to-peer networking tools to create health
buddy systems for coaching and counseling in order to promote physical activity (Gotsis, Wang,
Spruijt-Metz, Jordan-Marsh & Valente, 2013) and assist smoking cessation (Cobb, Graham, &
Abrams, 2010).
Moreover, distinct from a typical OSG, in OHSNs “connections come before content”
(Rau, Gao & Ding, 2008). Social networking features give an additional route to transmit
information and emotional support beyond message exchanges on discussion forums. Public
profiles are an equally important venue to post health information, resources (Yan, Tan & Peng,
2011) and provide emotional support such as posting a hug on other members’ profile walls
(Swan, 2009). For health resources located on members’ profiles, it is important to establish
connections first to get access to the content. Of course, the access to profile content may depend
on individual users’ privacy settings and it is sometimes possible to view full information
without becoming friends. Yet, in general, it is more convenient and realistic to access full
personal information by making the actual connections.
17

Quantified self-tracking. Mirroring the concept of Personal Health Records (PHRs) in
formal healthcare organizations, quantified self-tracking is a service that becomes increasingly
prominent in intentionally designed health social media. Quantified self-tracking is a behavior
that regularly collects “any data that can be measured about the self such as biological, physical,
behavioral or environmental information” (p. 509, Swan, 2009). Usually, additional services are
provided together with quantified self-tracking tools, including time-stamped graphical
visualization of the health records, and “a feedback loop of introspection and self-
experimentation” (p. 509, Swan, 2009) (Figure 2). Equipped with those tools, an individual user
is able to keep a continuous record of his or her behaviors, treatments, health regimens and
physical and mental condition. Through retrospective review of one’s history of health-related
records, the user can learn and understand the correlations among those health data. As a result,
the user may become more convinced and feel more efficacious to perform healthy behaviors,
and/or be more informed in the decision making process (Paton, Hanse, Fernandez-Luque &
Lau, 2012).

18

Figure 2. Quantified Self-Tracking in OHSNs

19

In addition to introspection and self-experimentation, social networking can help to scale
up the benefit of quantified self-tracking. The value of self-tracking of health records is not
restricted within an individual, but can spread to other people by sharing personal health data
(Wicks, Massagli, Frost et al., 2010). Social epidemiological studies have shown that knowledge
about and access to medical treatments and technologies are localized (Kawachi, & Berkman,
2003). Consequently, spatial and social constraints can greatly impact individual health
outcomes because of the disproportional availability of health information and knowledge
(Berkman & Kawachi, 2003). By reading others’ health records on online social networks,
individuals can learn from each other about novel health treatments, regimens and resources.
As a relatively new service in OHSNs, quantified self-tracking has received increasing
scholarly attention. One study on PatientsLikeMe found that among 1323 survey participants,
57% agreed that they learned about a symptom, side effects of their treatments and what it was
like to take a specific treatment for their condition through sharing self-tracking of health records
(Wicks, et al., 2010). Another qualitative study analyzed forum posts on PatientsLikeme, and
found that users reviewed and discussed personal health records on forums. Interestingly,
patients actually referenced personal health data and located others with particular experiences to
answer specific health-related questions on the forum (Frost & Massagli, 2008). These studies
show that quantified self-tracking of health records is a valuable information resource to online
health social network users.
Defining Online Health Social Networks
Broadly speaking, OHSNs refer to social networks of individuals who communicate and
interact with one another through social media for a shared common health-related interest or
goal. OHSNs can be used interchangeably with online health communities (Kraut, Resnick,
20

Kiesler et al., 2012; Rainie & Wellman, 2012; Wellman, Boase & Chen, 2002). The traditional
difference between online social networks and online communities is that online social networks
are organized around each individual, whereas online communities are organized around topics
or interests (Rau et al., 2008). However, as different web applications converge on a single social
media platform, online health social networks have the duality of networked individualism
(Wellman, Quan-Haase, Boase et al., 2006) as well as sharing common health-related interests.
As mentioned earlier in this chapter, OHSNs can be technically supported by open social
media or intentionally designed health social media. Intentionally designed health social media
include both formal CMSS groups and informal self-organized networks. The current
dissertation will focus on informal self-organized online health social networks. Many of these
informal self-organized OHSNs are entrepreneurial or commercial. Internet startups are the early
adopters of innovative services as new approaches to using social media for improved health
(Centola, 2013). For example, companies such as PatientsLikeMe, CureTogether, Medhelp,
SugarStats, FatSecret and SparkPeople have offered extensive social media platforms with
services including online health profiles, quantified self-tracking, historical display of patients’
health information, and social networking tools that afford members the opportunity to interact
and share personal health records. The current dissertation will focus on these informal self-
organized OHSNs who provide social media services of information and emotional support,
health buddy systems afforded by social networking features, and quantified self-tracking.

21

CHAPTER 3
Study I: Online Health Social Networks and Individual Health Outcomes
As discussed in the introduction, the purposes of the first study are to examine the
mechanisms through which members in OHSNs select their health buddies, and test the joint
dynamics of network selection and social influence on individual health outcomes
simultaneously. To achieve these research goals, the first study will adopt the Multi-Theoretical
Multilevel Framework as a theoretical guide. Specifically, it draws from literature on motivation
to participate in online health communities and derives hypotheses from two primary
motivations: looking for similar others and looking for diverse information resources. Then, it
reviews social influence theories and empirical evidence to derive hypotheses on social influence
through online social networks.
Multi-Theoretical Multilevel (MTML) Framework
Monge and Contractor (2003) proposed a Multi-Theoretical Multilevel (MTML)
framework to explain the emergence of organizational networks. Specifically, their work
summarized nine families of theoretical mechanisms at different levels of analysis that drive the
creation, maintenance, dissolution and reconstruction of organizational networks (Contractor,
Wasserman & Faust, 2006). There are three primary merits of the MTML framework that
contribute to our understanding of network formation. First, the emergence of networks is not
some phenomenon that can be fully explained by one single theory, or at one single level of
analysis (Robins, Pattison & Woolcock, 2005). The MTML framework allows for the
juxtaposition of several theoretical mechanisms across different levels of analysis to untangle the
unique influence that each theory claims. Second, the MTML framework allows researchers to
examine both the endogenous and exogenous factors that influence network formation, using
22

hybrid network attribute models (Monge & Contractor, 2003). Using the MTML framework
helps to understand the roles of both structural properties and individual attributes in driving the
global structure of networks. Third, the MTML framework helps network research move from
“exploratory and descriptive techniques to confirmatory and inferential techniques” (Contractor
et al., 2006, p. 683).
As originally developed to explain the emergence of organizational networks, the MTML
framework has been applied to study interorganizational networking patterns (Shumate, Fulk &
Monge, 2005). For example, using the MTML analytic framework, Atouba and Shumate (2010)
investigated collaboration networks among NGOs and INGOs. The results showed that
homophily in organization type and funding sources, and structural signatures (i.e., reciprocity,
transitivity and popularity) accounted for the interorganizational networks.
In the past three years, the MTML framework has been extended to study the assembly of
teams to advance the science of team (Börner, Contractor, Falk-Krzesinki et al., 2010). The
framework views the research problem of team assembly from a motivational perspective. It
explains individuals’ motivations to create, maintain, dissolve or reconstitute a team linkage with
another individual (Contractor & Su, 2011). The motivations are based on a variety of individual
attributes (e.g., age, gender, resources) and the extant links among individuals (e.g.,
communication, group affiliation) (Contractor, 2013). The nine families of theoretical
mechanisms summarized in the MTML framework help to refine a specific set of individual
attributes and certain characteristics of the extant links that reflect individual motivations in a
specific team assembly context. For example, Lungeanu, Huang and Contractor (2014) applied
the MTML model in studying the assembly of interdisciplinary scientific research teams. Their
work furthered our understanding of important mechanisms (i.e., high levels of prior co-
23

authorship, but low levels of prior citation relationships) of successful interdisciplinary
collaboration.
More interestingly, using the MTML framework, a series of studies has been devoted to
studying the assembly of ad hoc and transient teams in virtual communities (Contractor, 2013).
This work has been done in the context of Massively Multiplayer Online Role Playing Games
(MMORPGs), in which teaming up with other players is an important mechanism to win and
succeed as an individual player. Being the very first series of studies, they contribute to the
knowledge of how people organize their personal relationships and collaborations in ICT-
enabled virtual worlds. Specifically, they examined important mechanisms, such as homophily,
proximity and social exchange etc., in driving the building of collaborative relations in virtual
gaming teams (Huang, Shen & Contractor, 2013; Zhu, Huang & Contractor, 2013).
Teams in Virtual Worlds and Online Health Buddy Networks
Online health buddy networks are web-based social networks that individuals form to
work together to achieve a common health goal. Health buddies are sources of health
information, practices and social support in an OHSN. As an overarching theoretical framework,
the MTML model provides a useful lens to understand the mechanisms that drive the formation
of online health buddy networks. Online health buddy networks resemble teams in virtual
worlds, such as MMORPG teams (Zhu, et al., 2013) and project teams in open source online
communities (Shen & Monge, 2011).
One major similarity lies in the dynamic, emergent and voluntary nature of organizing in
teams and networks. In OHSNs, online gaming communities and open source communities,
individuals are embedded in social contexts and are able to self-organize into teams or networks
to accomplish challenging tasks and fulfill common goals. In both online gaming and health
24

social networks, for example, the discussion forum is an important place to recruit potential team
members (Williams et al., 2006) and health buddies. Individuals who are in need of teammates or
buddies often post a call on the discussion forum in the community to meet potential relational
partners. Particularly in OHSNs, individuals may browse other users’ profiles or search users
using buddy finders available in the community. Through various channels, individuals are able
to self-organize or voluntarily form their personal buddy networks to fulfill a shared health-
related goal. The formation of teams and buddy networks are dynamic and emergent from
various social contexts.
There are two major differences between online health buddy networks and teams in
virtual worlds. First, teams in virtual worlds tend to have well-defined boundaries whereas online
health buddy networks tend to be more diffused. Second, teams in virtual worlds are more task-
oriented and transient whereas online health buddy networks tend to last for a longer period. For
example, a combat team in gaming virtual worlds usually assembles and lasts for 30-40 minutes
to complete one combat session together. After finishing the combat, teams are dissolved and
individuals do not have to maintain relations with previous teammates (Huang, Ye, Bennett &
Contractor, 2013). At least for online games which do not support a permanent and immersive
gaming environment, research has shown that there is minimal actual communication and social
interaction taking place in those short-term, task-orientated and transient teams (Huang et al.,
2013; McEwan, Gutwin, Mandryk & Nacke, 2012).
In contrast, online health buddy networks are more enduring. They are built on mutual
agreement on the reciprocal relationship, just like friendship ties in Facebook. After the buddy
relationship is confirmed from both relational parties, they will get news feeds and status updates
from each other, and they can communicate with one another by writing on walls and/or sending
25

private messages. Research on OHSNs has revealed that the number of health buddies/friends
positively predicted one’s perceived emotional support gained from the network (Chung, 2013).
Combined with findings in virtual gaming teams, the findings on health buddy networks and
emotional support suggests that online health buddy networks may present more enduring and
meaningful social relationships than in teams in gaming virtual worlds.
Given the above discussion on the similarities and differences between online health
buddy networks and teams in virtual worlds, it is worth studying the mechanisms that drive the
formation of online health buddy networks. The duality of goal orientation and relatively more
enduring social relationships of online health buddy networks makes it theoretically interesting
to further the research in the assembly of virtual social networks. As pointed out in Chapter 2,
with respect to social media and health as a research topic, little is known about how individuals
select their health buddies in virtual environments. Therefore, one focus of this study is to
explore the formation mechanisms of online health buddy networks as a self-organizing process.
The findings will be important not only to scholars in social media and health, but also to
communication technology designers. Online health buddy networks may be engineered to better
facilitate healthy outcomes by following the natural motivations of individuals as discovered in
this study.
Formation Mechanisms of Online Health Buddy Networks
The MTML framework proposes nine families of theories to explain the motivations of
forming communication and social networks. Literature on motivations and benefits to
participation in online health support groups and health-related social media is useful in this case
to suggest which theoretical mechanisms are important in the context of online health buddy
networks.
26

Motivations for participating in online health networks.
Using Uses and Gratifications theory, a number of studies have investigated individuals’
motivations to participate in online health social networks. One major motivation for
participating in health-related social media is to find people who share similar health concerns
(Wright & Bell, 2003). Web technologies enable people to overcome geographic barriers to
aggregate and interact together based on their common interests. Similarity in online health
social networks is especially important because it is the foundation for meaningful information
sharing and social support exchange. With similarity in health concerns and health experiences,
members in online health social networks can provide great empathy and social support.
Similarity between network members also leads to other important motivations identified in the
literature. For example, similarity may foster identification with other network members. This
contributes to the motivation of seeking a sense of belonging to the community (Tang & Lee,
2006). Moreover, similarity can facilitate the motivation of seeking emotional support in that
similarity helps to build swift trust and mutual understanding (Meyerson, Weick & Kramer,
1996).
Another major motivation for participating in health-related social media is to gain
diverse informational sources. Online health social networks help to pool distributed health
information, practice and experience from individuals around the world (Wright & Bell, 2003).
While one’s offline social networks, such as family, friends and physicians, are critical channels
of health information seeking, they have limitations. The first is that one’s offline social network
may be restricted with respect to the scope of relevant health information and experience. It is
possible that one’s family and friends simply do not concern themselves with the relevant health
issues, and thus they lack the appropriate information and experience to support the focal person
27

(Wright & Bell, 2003; Wright, 2009). Another limitation of one’s offline social network as an
effective information source is that, at times, the person may not want to seek health information
from family, friends or physicians. It is possible that people may feel embarrassed or stressed
disclosing personal health information to their offline social networks, because they may be
judgmental, overprotective or too emotionally fragile to cope with the situation (Tanis, 2008;
Thoits, 1995).
Fortunately, with the affordances of web technologies, people are able to reach out a
diverse sources of health information who share similar health concerns but come from different
social, cultural and/or geographical backgrounds (Wright & Bell, 2003). Those people who are
electronically connected are valuable sources of health information, practice and experience
(Idriss, Kvedar & Watson, 2009). They can be considered as “experiential experts”, whose
information and first-hand experiences are less accessible offline (Tanis, 2008). By pooling
distributed “experiential experts,” OHSNs act as a system in which individuals with shared
health goals are able to observe and learn from one another, and glean information about how to
behave or to cope with the situation they are facing. In addition, OHSNs are mostly weak tie
relationships (Walther & boyd, 2002). Therefore, people may have less concern about
embarrassment or stress in disclosing and seeking health information with one another (Idriss,
Kvedar & Watson, 2009). The diverse and rich information sources in the online environment
can be very effective, if people choose to activate the latent ties. More recently, Chung (2013)
studied motivations of online social networking for diabetics. The author found that information
seeking was the strongest motivation, followed by motivation to help similar others, and
motivation to meet others in similar conditions.
28

The above literature on OHSNs informs us that two theories out of the nine families of
theories may be particularly important in explaining the formation of online health buddy
networks. They are homophily theory and social exchange theory. The following sections will
provide detailed discussions of each theory and its potential application in the context of OHSNs.
In addition to the exogenous factors explained by homophily and social exchange theories, self-
interest theory, balance theory and preferential attachment will be used to derive endogenous
factors as controls in the analysis of network formation.
Multidimensionality in homophily.
Homophily is a simple yet basic organizing principle in a variety of communication and
social networks (McPherson, Smith-Lovin & Cook, 2001). The homophily mechanism refers to
the tendency for individuals to be attracted to those who have similar attributes (Monge &
Contractor, 2003). Seminal works on homophily and communication relations suggest that
homophily breeds interpersonal attraction, empathy and effective communication (McCroskey,
1974; Rogers & Bhowmik, 1970). Two primary scholarly works have shed light on the
multidimensionality of homophily. The concept of homophily went back to Lazarsfeld and
Merton (1954), who distinguished status homophily from value homophily. Status homophily
includes major sociodemographic characteristics such as age, sex, ethnicity, religion and
education. Value homophily includes internal states, such as beliefs and attitudes that orient
future behaviors. The other important work, conducted by McPherson and Smith-Lovin (1987),
distinguished baseline homophily from inbreeding homophily. The work was situated in dyadic
relationships in face-to-face groups. Baseline homophily refers to similarities in demographic
characteristics, while inbreeding homophily refers to homophily induced by “social structures
below the population level” (McPherson et al., 2001, p. 149). The critical distinction between
29

baseline homophily and inbreeding homophily is that baseline homophily emphasizes within-
group variance, whereas inbreeding homophily emphasizes between-group variance (McPherson
and Smith-Lovin, 1987). To put in another way, baseline homophily is about individual
preferences and choices of relational partners, whereas inbreeding homophily is about relational
ties induced by being in the same organized foci. Building on the existing literature on
multidimensional homophily, the current study will focus on demographic homophily,
inbreeding homophily. More importantly, the current study will propose a health-related
dimension of homophily, informed by research on the role of health characteristics in network
selection, and the Stages of Change Theory (Prochaska & DiClemente, 1983).
Demographic homophily. Demographic homophily is the earliest and most examined
dimension of homophily in the formation of communication and social networks in various
social settings. Extensive research has found evidence of age and gender homophily in large
populations (McPherson et al., 2001), as well as in face-to-face work organizations and small
groups (Hinds, Carley, Krackhardt & Wholey, 2000). Recent studies have started examining the
hypotheses of demographic homophily in virtual worlds, yielding somewhat mixed empirical
findings. For example, Yuan and Gay (2006) studied distributed virtual teams in an online
learning community. The authors found that neither gender nor ethnic homophily was a
significant predictor of the communication network. Similarly, researchers found that gender
homophily was neither a significant predictor of forming teammate relationships in MMORPGs
(Huang, et al., 2013; Zhu et al., 2013), nor a significant predictor of tie decay in collaboration
networks in MMOGs (Shen, 2010). In contrast, age similarity was found to be a significant
predictor of the formation of teams in the online gaming environment (Huang, et al., 2013; Zhu
et al., 2013).
30

The explanation behind significant and non-significant findings centers on the visibility
of demographic attributes in the virtual environment and the extent to which online social
networks mirror one’s offline social networks. In both online learning communities and
MMOGs, which are virtual worlds, demographic attributes are not immediately visible.
Individuals may have a clue about each other’s age and gender after a certain period of online
interaction. Yet, existing empirical findings seem to suggest that the lack of non-verbal cues in
computer-mediated environment indeed undermines the effects of demographic homophily to
some extent. This helps to explain the non-significant findings of demographic homophily in
virtual learning teams (Yuan & Gay, 2006). Moreover, researchers argued that both the
significant and non-significant findings on demographic homophily may result from the fact that
individuals migrate their offline social relationships to the virtual worlds (Huang et al., 2013;
Shen, 2010). For example, people would be very likely to form a team with others they have
already known in the offline world (Williams, Consalovo, Caplan & Yee, 2009). Age similarity
is a consistently observed variable in the organization of offline relationships (McPherson, et al.,
2001). Also, female players tend to play with their romantic partners in virtual games. Those
facts may explain the significant findings for age similarity and non-significant finding for
gender similarity in virtual gaming teams.
With a built-in feature of social networking in OHSNs, demographic information such as
age and gender are immediately visible to other site members on one’s personal profile. People
do not need to rely on a certain period of social interaction to have basic clues about each other’s
background. Instead, they can acquire such information in a second by browsing others’ profiles
on the site. In social network sites, information presented on personal profiles serves as an
important cue in signaling individual identity (Ellison & Hancock, 2013), and plays a critical role
31

in self-presentation and impression management (Walther, Van Der Heide, Hamel & Shulman,
2009). A series of studies conducted by Walther and his research team manipulated Facebook
profile cues that conveyed personal information in controlled experimental settings. What they
consistently found was that profile cues indeed influence others’ perceptions and evaluations
about the focal person (e.g., Walther et al., 2009; Walther, Van Der Heide, Kim, Westerman &
Tong, 2008). Hancock and his research group have focused on online dating social network sites,
with similar findings on the importance of profile cues in interpersonal perception and
relationships (e.g., Ellison & Hancock, 2013; Toma & Hancock, 2012).
Similar to virtual gaming communities, people may bring their offline relationships to
online health social networks. The online health community may act as an additional channel for
offline social contacts to provide social support and help to monitor health conditions (Hossain,
2014). Given the importance of demographic homophily in the offline communication and social
interactions, and the visibility of demographic cues in OHSNs, the first two hypotheses are
proposed below:
Hypothesis 1a (H1a): Individuals are more likely to become online health buddies with
others of the same gender.
Hypothesis 1b (H1b): Individuals are more likely to become online health buddies with
others of similar age.
Inbreeding homophily. McPherson et al. (2001) theorized that inbreeding homophily that
is induced by being affiliated to the same groups and organizations is another important
dimension of homophily. Group membership creates an important context for social connections
because it serves as a focus for shared activities and experience (Feld, 1981; McPherson et al.,
2001). The Internet and online health communities provide basic electronic infrastructure to form
32

social ties. The shared activities and experience may serve to foster the transformation from
latent ties to weak ties (Haythornthwaite, 2002).
The importance of group affiliation in the formation of virtual social networks has
received some empirical support. In studying a distance learning community, Haythornthwaite
(2001) found that the social network structure of a class converged on group assignment. In other
words, communication of collaboration and socialization was much more frequent among people
of the same group than that of different groups. Similarly, Yuan and Gay (2006) found that
members of the same group in the context of online learning communities were more likely to
communicate task-related content and seek advice from one another. Moreover, studies on team
assembly in online gaming environments consistently found that group affiliation increased the
likelihood of players’ teaming up together (Zhu et al., 2013).
In online health social network sites, people are able to create their own interest groups
and to join any groups that they are passionate about. The groups may be organized around
health-related categories (e.g., specific diseases and health conditions), and/or individual
attributes (e.g., personal identity and personal interests). For example, in an online weight
management social network, voluntary groups created by health-related categories can be
“diabetics on a diet”, in which diabetics try to manage weight together, and “Atkins diet group”,
in which those who use that particular diet regime gather. Voluntary groups organized around
individual attributes can be very diverse. For instance, there are groups like “nursing moms,”
“military spouses,” “vegans on a mission,” “night owls.” etc.
The voluntary groups may create a basis for trust between members in an online health
social network site. Shared interests, experiences and personal identities provide important
common grounds for mutual understanding. Valuable information exchanged facilitates the
33

development of expectations of each other and fosters a sense of accessibility of interpersonal
relations. In addition, as explained by Monge and Contractor (2003), self-categorization is an
explanatory process that underlies the formation of homophilous ties. People tend to categorize
themselves and others in terms of social characteristics. Communication and relationships are
more likely to occur between two persons if they think they belong to the same social category.
Belonging to the same voluntary group may act as a cue to activate self-categorization.
Therefore, the following hypothesis is proposed:
Hypothesis 2 (H2): Individuals are more likely to become online health buddies with
others of the same interest group.
Health-related homophily. As discussed earlier in this chapter, looking for others who
have similar health concerns and medical problems is one of the most important motivations to
participate in online health social networks (Wright & Bell, 2003). There is a dearth of research
about the roles of health-related similarities in facilitating social ties among members in an
OHSN. This is an important topic that deserves more attention because only by knowing which
specific health-related similarity motivates people to establish social ties can researchers and
practitioners better evaluate the effects of social connections and engineer OHSNs to promote
health. Smith and Christakis (2008) called for more research into how health-related
characteristics affect the creation and structure of networks. Understanding the relationship
between health-related similarity and network formation is critical to distinguishing the effects of
health on network structure from the effects of network structure on health.
In responding to the call by Smith and Christakis (2008), a series of empirical studies has
shown that similarity in health characteristics indeed influences the selection of relational
partners in one’s social networks (Veenstra, Dijkstra & Steglich, 2013). Most of these works
34

have been conducted with an adolescent population. Findings indicate that adolescents seek out
friends who have similar patterns of alcohol use, smoking behavior and substance use (Pearson,
Steglich & Snijders, 2006; de la Haye, Green, Kennedy et al., 2013). Not only do similar health
behaviors breed social connections, but also a similar health status may facilitate formation of
social ties. Take the health status of obesity as an example. Valente et al (2009) found that
overweight adolescents were more likely to have overweight friends than their normal-weight
peers. De la Haye et al. (2011b) reported that the phenomenon of weight-based clusters among
adolescent friends was a pure selection process, rather than a social contagion process. The
authors also warned that the selection process indicated that overweight adolescents were
socially marginalized from those who had normal weight. This unpleasant fact would result in
overweight adolescents’ staying together and reinforcing each other’s unhealthy behaviors (de la
Haye, Robins, Mohr & Wilson, 2010, 2011a).
OHSNs are different from adolescent peer networks in that the purpose of online health
networks is to work together to achieve a common health goal. Unlike naturally occurring offline
social networks such as friends, coworker and neighborhood networks, OHSNs are created with
an inherent motivation to look for others who are similar on health-related characteristics
(Walther, Pingree, Hawkins & Buller, 2005). Research has also shown that homophily in health
status enhances the perception of health information credibility (Wang, Walther, Pingree &
Hawkins, 2008). By becoming health buddies, network members are able to get access to each
other’s health information. Therefore, it is reasonable to expect a strong effect of health-related
homophily in creating online social ties in the specific context.Below I consider three aspects of
health-related characteristics: initial health status, health goal and health progress.
35

Initial health status refers to the health condition that an individual has when he or she
joins an OHSN. This variable could be operationalized in various ways in different health
contexts. For example, it could be the degree of one’s tobacco addiction when the person joins an
online smoking cessation community; it could be the severity of one’s mental health problem
when the person joins an online mood management community; it could be one’s actual weight
when the person joins an online weight management community. Health goal refers to the
specific level of health that a person hopes to reach by participating an OHSN. Similar to initial
health status, health goal could be to quit smoking in three months, to manage one’s levels of
distress within a low to moderate degree, or to lose 50 lbs by working together with health
buddies connected in OHSNs.
In OHSNs, individuals are able to indicate their initial health statuses and health goals on
personal profiles by sharing personal health data (Swan, 2009). Given that members of online
health social networks look for others similar in health conditions, it is reasonable to speculate
that initial health status would play a role in the generation of health buddy ties. In the offline
world, friendships are more likely to exist between peers with the same health status (Valente, et
al., 2009; de la Haye et al., 2011b). In the online world, similarity in initial health status is the
most direct and clear cue to legitimate the initiation of a relationship for the purpose of
collectively coping with a common health concern. Health goal maybe another important health-
related form of homophily in driving the formation of health buddy ties. Dyadic similarity in
health goal suggests that two persons may need similar sets of informational resources and
practices in the process of behavioral change and/or personal health management. In other
words, if holding different health goals, two persons may not be motivated to initiate an online
relationship because each other’s health information and experiences are not relevant or useful.
36

For example, a person who wants to lose 100 lbs may not see the point of becoming a health
buddy with someone who wants to lose only 10 lbs. Instead, the person may well choose to be a
health buddy with someone who wants to lose 105 lbs. The following hypotheses are proposed
with respect to health-related homophily:
Hypothesis 3a (H3a): Individuals are more likely to become online health buddies with
others having similar initial health status.
Hypothesis 3b (H3b): Individuals are more likely to become online health buddies with
others having similar health goals.
Moreover, The Transtheoretical Model (TTM) informs us the importance of similarities
in stages of illness recovery and behavioral change in facilitating the creation of online health
buddy ties. TTM focuses on health behavioral change. It claims that behavioral change unfolds
through a series of stages (Prochaska & DiClemente, 1983). Initial TTM studies investigated
smoking cessation behavior, and later on TTM was applied to a variety of health behaviors
including cancer screening, physical activity, obesity, alcohol and substance abuse, etc.
(Prochasks, Redding & Evers, 2008). Given the different stages of change, TTM emphasizes that
health intervention strategies need to match characteristics of stages to be successful.
TTM tells us that any behavioral change is not a discrete event, but a continuum
unfolding over time. Individuals in OHSNs may find others who are at similar stage of illness or
recovery to be more relevant sources of information and support. The match between two
persons on stages of illness or recovery may be important in providing and acquiring expertise
just as it is needed (Walther, et al., 2005). Since TTM is a theory of behavioral change, this study
will use a different term progress other than stage. Many OHSNs aim to promote actual health
outcomes, during which process a variety of behavioral changes can be involved. Specifically,
37

health progress refers to the extent to which an individual has moved from his or her initial
health condition towards his or her health goal. During the process of behavioral change and
disease recovery, individuals may be at different stages of progress with respect to their goals.
For example, in the process of weight loss, some people may have achieved 70% of their goals
while some people may have made negative process with respect to their health goals by gaining
weight. The following hypothesis is proposed with respect to homophily in health progress:
Hypothesis 3c (H3c): Individuals are more likely to become online health buddies with
others with similar health progress.
The paradox of social exchange in online health networks.
Looking for diverse information sources and experiential experts is another major
motivation to participate in OHSNs (Wright & Bell, 2003; Walther, et al., 2005). As discussed
earlier, an OHSN is able to pool distributed patient expertise and create a space wherein
members can collectively cope and manage their health conditions. The strength of weak ties is
to provide non-redundant and useful information (Granovetter, 1973). Individuals should be
motivated to establish connections with those who possess rich health information. One
prominent indicator of one’s informational resource is personal health data posted on one’s
profile. By reading those data, a person may gain information about health conditions, treatments
and medical regimes that other people have. One study examined a subscription network on
PatientsLikeMe, wherein patients can follow each other to get updates on treatment and drug
information (Yan et al, 2010). It revealed that PatientsLikeMe users were more likely to follow
other users who present high-quality personal health data on their profiles, controlling for various
homophily factors.
38

An online health buddy network is different from a subscription network in that a health
buddy network resembles a friendship network, where mutual confirmation about the
relationship is required. Social exchange theory (Blau, 1964) views the creation and maintenance
of social relations from a resource perspective. It posits that individuals form interpersonal
relations to exchange the valuable resources that each needs. Despite having developed in
diverse research disciplines, social exchange theorists agree that reciprocity is one essential
characteristic in exchange relationships (Blau, 1964; Emerson, 1976). Given the importance of
reciprocity in the relationship, it is highly possible that individuals who possess rich health
information would not like to form a health buddy tie with someone who has little information.
The situation in OHSNs is a bit more unclear than that. The first factor that complicates
the speculation is types of reciprocity. Blau (1964) drew a qualitative difference between direct
reciprocity and indirect reciprocity. Direct reciprocity is characterized by an immediate return of
resources, while indirect reciprocity is characterized by the lack of one-to-one correspondence
between what two relational parties give and take from each other (Takahashi, 2000; Molm et al.,
2007). In the context of exchange activities in larger groups or networks, individuals may not
expect immediate direct reciprocity from other members. Instead, they may expect resources are
passed along and they will be rewarded ultimately from someone else in the group or network
(Molm et al., 2007). Indirect reciprocity has been shown to be prevalent in online exchange
relationships. Flynn (2005) theorized that people with collective identity orientation express a
preference for indirect reciprocity in the exchange relationship. The reason is that resources are
distributed and emergent among different members in an online community. In addition, indirect
reciprocity has consistently been linked to norms of sharing and helping (Ekeh, 1974). Norms of
sharing and helping are widely accepted in online health communities (Katz & Rice, 2002). Oh
39

(2012) found that altruism is the most influential while personal gain was the least influential
motivation across different groups of answerers in the health knowledge sharing site. Following
the logic of indirect reciprocity, individuals who possess rich health information would like to
form a health buddy tie with those who have limited information. Given the fact that most
community members would want to get access to good information sources, it is reasonable to
speculate that individuals who present rich personal health information tend to have more online
health buddies than those who present little information.
The second factor – privacy of personal health information – may turn the argument in a
different direction. Studies that support indirect reciprocity in online networks have been mainly
conducted in the context of message boards wherein people post and answer questions. OHSNs
differ from message boards in that networks are organized around each individual person, rather
than threads and messages. Although norms of sharing and helping still exist, concerns of
privacy of personal information are more salient across several major social network sites
(Lewis, Kaufman & Christakis, 2008; Fogel & Nehmad, 2009). Personal health information can
be even more sensitive to privacy concerns (Eysenbach, 2008; Li, 2013; Wicks et al., 2010).
Therefore, people may be cautious about becoming health buddies with others, leaving
information sharing activities to message boards. A recent work provides somewhat indirect
support to this speculation, finding that individuals who were motivated to seek information
would use message board more frequently rather than have more friends in online health social
networks (Chung, 2013). However, the online health social networks studied did not incorporate
the variable of personal health information sharing on profiles.
Personal health information is a valuable information resource to members of online
health social networks. On the one hand, norms of sharing and helping may facilitate indirect
40

reciprocity, and thus make individuals who present rich health information more accessible to
other members in the community. On the other hand, concerns about privacy of personal health
information may make individuals cautious in making online health buddies. This will bring the
decision about forming social ties down to interpersonal trust. Individuals who present rich
health information do not necessarily have more health buddies because they want to avoid
potentially risky relationships. Given the plausibility of both arguments, the following research
question is proposed:
RQ1: Will individuals who present richer health information have more health buddies
than those who present less health information?
Endogenous Factors
The formation of social networks greatly depends on the other network relationships each
person has. That is, the generation of social networks is inherently interdependent among
network ties. Social network analysis allows researchers to simultaneously examine endogenous
factors, while testing exogenous factors (i.e., the individual and dyadic attributes mentioned
previously) in studying the generating mechanisms of social networks. The failure to include
endogenous factors will lead to biased and overestimated effects of exogenous factors (Monge
and Contractor, 2003; Robins & Pattison, 2005).
Research on relational networks of offline friendship and online interaction indicates that
endogenous structural tendencies such as sparsity, popularity and transitivity influence the
formation of virtual social networks (Burt, 1980; Feld, 1981; Huang et al., 2013). With respect to
sparsity in the network structure, self-interest theory postulates that individuals are rational in
making decisions about establishing social ties with others (Monge and Contractor, 2003). Given
that individuals have limited time, energy and resources at their disposal to initiate, develop and
41

maintain social relations, they would not randomly create ties with others. Put in another way, in
online health social networks, individuals tend not to become health buddies with other members
at random. Instead, the choice to make a health buddies seems to be driven by certain laws. With
respect to popularity, the power law distribution is a fundamental and common property found in
many large networks (Barabási & Albert, 1999; Adamic & Huberman, 2000). The power law
distribution is that a few individuals have many ties while most individuals only have a few ties
in the complex network system. The factor that generates the power law distribution is
preferential attachment, such that an actor that has more connections than another actor will
increase its connectivity at a higher rate (Barabási & Albert, 1999; Newman, 2001). Therefore,
popular individuals will get even more social ties in online health social networks. This study
will include this structural property as a control variable, such that an online health buddy
network will exhibit a structural tendency toward preferential attachment.
In addition to preferential attachment to actors who are already popular in the network,
most relational networks are characterized by a greater number of triangles. Balance theory
(Heider, 1958) posits that if two individuals are friends, they tend to have similar attitudes
towards an object or another person. While the original balance theory emphasizes the cognitive
consistency in directed networks, later researchers extended the theory to various types of triads
of three connected nodes (Harary, Norman & Cartwright, 1965; Holland & Leinhardt, 1976).
Transitivity describes the phenomenon that two individuals who are both tied to a third person
are also tied to each other. That is, friends of friends tend to become friends. Recent empirical
studies provide evidence of the network tendency of transitivity. For example, Huang et al.
(2013) found transitivity as an important network structural configuration in social interactions in
an online gaming world. By conducting an online experiment, Golder and Yardi (2010) found
42

that transitivity played a key role in establishing new ties on Twitter. Particularly in online health
social networks, it is not only the balanced and symmetric relationships that matter, but also that
one’s friend might be an important person to introduce to health buddies. It could be the simple
fact that friends of friends are more accessible and easier to get connected to than the rest of
members hidden somewhere in the online network. Therefore, this study will include this
structural property as a control variable, such that an online health buddy network will exhibit a
structural tendency toward transitivity.
Social Influence of Online Health Buddy Networks
Online health buddy networks are peer networks in which people share information and
provide support to one another. To health communication researchers, it is critical but not
enough to know the generating mechanisms of peer networks. The other central question is if
peer networks exert social influence on people’s health behaviors and/or health outcomes. There
has been an upsurge in studies examining the effects of social networks on health behavior and
outcomes. For decades, researchers have framed studies of peer influence as a competition
between influence and selection as explanations of similarity among peers (Veenstra et al.,
2013). Thanks to advanced social network analysis, studies of peer influence have started
examining the joint dynamics of network selection and influence (Pearson et al., 2006), rather
than treating them as competing alternatives. Following this research paradigm, this study will
empirically test if social influence on health outcome exists in online health peer networks, after
controlling for selection effects based on health-related characteristics.
Christakis and Fowler’s (2007; 2008) seminal works on the spread of obesity and
smoking behavior through offline social networks provide strong empirical support to peer
influence on health outcomes. In their 32-year Framingham Heart Study, with a sample of adults
43

aged from18 to 70, Christakis and Fowler (2007) reported that a person’s chance of becoming
obese increases by 57% if a friend becomes obese. Peer influence can also be positive on health
behaviors. In a later study (Christakis & Folwer, 2008), the researchers found that a person’s
chance of quitting smoking increases if a friend stops smoking. More recent studies demonstrate
social influence in adolescent and college peer networks while controlling for network selection
effect (Lewis, 2012; Veentra et al., 2013). For example, Simpkins and Schaefer et al. (2012)
studied a sample of 1,896 adolescents for eight months. The authors found that participants
decreased their levels of physical activity if the average level of physical activity of their friends
decreased.
A few other studies have shown the evidence of peer influence in web-based health
interventions and online health social networks. Leahey and colleagues (2012) conducted a 12-
week team-based online intervention program for weight management. Over 5,000 overweight
individuals formed teams of 5-11 members by self-selection and recruitment. During the
intervention program, team members were provided an online system to log their weight and
physical activity information and to track progress towards their weight goals. They also had
access to information about their teammates’ performance. Team members were encouraged to
support one another by sharing weight loss information and exercising together. With 987 teams,
the study found that a greater percentage of teammates in the weight loss division was associated
with a greater percentage of weight loss. Similarly, achieving a clinically significant weight loss
tended to cluster within teams. What is interesting is that individual weight outcome was
significantly influenced by perceived team-level social influence.
Moreover, Centola (2010) conducted an experiment by creating an online health
community to compare which network structure (i.e., random vs. clustered ties) was a better
44

conduit for the spread of health behavior. The author found that healthy behaviors indeed spread
efficiently in online health social networks, on the condition that the network is built upon some
homophilous traits among network members. By studying an existing online health social
network, Ma, Chen and Xiao (2011) added more interesting findings in this line of research. The
authors compared health attributes of community users who were friends with each other in the
community, and found that users’ changes in health conditions had rippling effects through
friends’ network. Specifically, the authors found that the users’ weight changes correlated
positively with their friends’ weight-change performance. Yet, the study did a simple correlation
analysis, which was not sufficient to rule out alternative interpretations.
Peer influence has a profound theoretical foundation. Hoffman et al. (2006) reviewed
intrapersonal, interpersonal, group and network theories that address the processes of peer
influence. Among those, theoretical perspective of social learning (Bandura, 1986) and social
norms (Cialdini, 1990; Fishbein & Ajzen, 1973) are thought to partially account for the social
contagion effects of social networks on health outcomes (Christakis & Fowler, 2007; de la Haye
et al., 2013). Studies provide preliminary support for both theoretical perspectives. Leahey et al.
(2011) examined the mechanism underlying the obesity clusters in a young adult population.
They found that having more social contacts trying to lose weight is associated with greater
intention to lose weight, and that weight control social norms fully mediate this effect. Without
discovering any significant effect of social norms, de lay Haye et al. (2013) found that social
influence happened through simply imitating eating habits of others in one’s social network. The
aim of this study is not to untangle which theoretical mechanism is best to explain social
contagion effects. Therefore, arguments with respect to theoretical comparison and critique go
beyond its scope. Nonetheless, this study is interested in rigorously investigating if social
45

influence would happen in naturally existing online health social networks. Given the previously
mentioned theoretical and empirical support, the following hypothesis is proposed:
Hypothesis 4 (H4): In online health social networks, an individual’s health outcome will
become similar to the average health outcome of his or her health buddies in the network.
46

CHAPTER 4
Study II: The Effects of Ego Network Structure and Social Support on Self-Tracking
Behavior and Health Progress
As discussed in the Introduction, recent technological innovations give rise to the
incorporation of self-tracking tools in online health social networks. Self-tracking is an important
skill in self-management of chronic disease (McCorkle et al., 2011). Extensive studies have
provided evidence of the effectiveness of self-tracking in promoting physical activities and
maintaining healthy diet for diabetes patients and obese people (e.g., Olander, Fletcher, Williams
et al., 2013). Yet, researchers have paid relatively little attention to the antecedents to self-
tracking behaviors. Literature on social support and disease self-management suggests that social
support provided by one’s primary social relationships such as family and friends helps to
encourage and sustain self-management behaviors (see review in Gallant, 2003). This study is
motivated by curiosity about the role of online health social networks in enhancing self-tracking
behavior through provision of social support. This chapter draws from literature on social
networks, social support and self-regulation. It aims to propose and test a theoretical model of
the relationships among ego network structures, perceived social support and self-tracking
behavior.
Social Support and Social Networks
In Chapter 3, I discussed the effect of social networks on individual weight outcomes,
under the theoretical assumption of social influence exerted through the networks. While social
influence is an important pathway through which social networks affect health, there are several
alternative theoretical mechanisms (Smith & Christakis, 2008). For example, social capital is a
theoretical perspective that explains the contextual influence of networks and groups on
47

individual health outcomes (Kawachi, 2010). Social capital is defined as “the aggregate of actual
or potential resources linked to possession of a durable network” (Bourdieu, 1986, p. 248).
Theorists such as Lin (1999; 2001) also view the concept as relational resources embedded in the
connections that one has with others to mobilize to achieve personal goals. By conceptualizing
social capital from a network perspective, the approach explicitly recognizes the inequality in the
access to social capital between individuals, as individual networks are not all the same
(Kawachi, 2010). Some networks are more powerful than others in terms of the stock of
resources.
Another closely related theoretical perspective is social support (Berkman & Glass,
2000). Similar to social capital perspective, social support theorists consider social relationships
as important channels through which various types of support may flow (Albrecht & Adelman,
1987a). Yet, social support research specifies a more functional content of social relationships
that can be categorized into four types of supportive behaviors (Heaney & Israel, 2008):
emotional support, informational support, appraisal support and instrumental support. More
specifically, emotional support involves the provision of empathy, love, trust, and caring.
Informational support involves the provision of advice, suggestions and information that a
person can use to address problems. Appraisal support involves the provision of information that
is useful for self-evaluation purposes, such as constructive feedback and affirmation.
Instrumental support involves the provision of tangible aid and services that directly assist a
person in need (Heaney & Israel, 2008).
Social support is distinguished from other mechanisms of network effects on health, such
as social influence and social capital, in the following ways. First, the supporter usually
consciously offers support, instead of exerting social influence through the support receiver’s
48

observation and imitation (Bandura, 1977), or social comparison processes (Festinger, 1954).
The purpose of social support is not to exert influence, but to create a supportive context and
respect each person’s autonomy of making his or her own decision (Albrecht & Adelman,
1987b; Heaney & Israel, 2008). Even though social influence may happen during the process of
supportive communication, the intention of social support is to provide four types of support so
that support receivers can enhance their senses of self-control in a stressful event (Albrecht &
Adelman, 1987b). Second, social support is always intended to be helpful by the provider of the
support. The essence of social support is to foster an interpersonal context of caring, trust and
respect. In contrast to social capital, social support has a much less sense of competition for
resources, or exploiting and withholding desired resources to exert control over others.
In sum, while social capital refers to relational resources embedded in one’s social
networks, social support emphasizes the functional content of social relationships. Social
network refers to linkages or connections between people, and both social capital and social
support are explanatory mechanisms of the effects of social networks on individual health
behaviors and outcomes. In studies of social networks and health, researchers often conflate the
three related yet distinct concepts by operationalizing social networks and/or social capital as the
actual resources that one’s social contacts can provide (e.g., Berkman, et al., 2000; Wills &
Ainette, 2012).
Although the functional view of social support dominates the literature in the past decade,
the structural view of social support appears in earlier sociological works on social support. The
structural social support refers to the availability of potential support-givers irrespective of the
actual exchange of support (Barrera, 1986). In offline settings, potential support-givers often
refer to significant others in one’s social environment, such as families, friends, close co-workers
49

and religious groups. Even if connections with those potential support-givers do not necessarily
indicate the actual exchange of information, emotional, appraisal and tangible social support, the
existence of those social ties could potentially serve as “social support resources” (Sandler, 1980.
P. 43). The structural social support typically uses social network measures to capture the size of
primary social relationships and the frequency of communication between the focal subject and
his or her primary social contacts (Wills & Fegan, 2001). The structural view of social support is
a network perspective of social support. In nature, it resembles one’s overall social network.
Therefore, the measures have been criticized to erroneously assume that all of person’s
connections involve the provision of social support (Wellman, 1990).
Then, researchers embraced the functional view of social support by explicating the
content of support provided. Social psychologists further developed sophisticated scales to
measure one’s perceived social support in the form of information, emotional, appraisal and
tangible support (e.g., Sherbourn & Stewart, 1991). Perceived social support is a psychological
measure that considers the useful features of one’s structural social support, or one’s overall
social networks. Sociologists have also incorporated the functional view of social support to the
network measures of social support. The measures usually identify just those network members
who provide social support exchange by specifying the content of social support provided by
each linkage (Wills & Fegan, 2001). For example, in a typical survey, respondents were asked to
nominate names who provided social support, such as personal advice, information and material
assistance, etc. (Wills & Ainette, 2012). Social capital researchers have also used similar
measures to identify the network of potential and actual resource providers. Lin’s Position
Generator (2001) and van der Gaag and Snijders’ Resource Generator (2005) are often used to
operationalize resources embedded in one’s social networks (Kawachi, 2010). Taking Resource
50

Generator as an example, the survey instrument typically provides a checklist of different kinds
of social resources that respondents can potentially access through their networks. Respondents
may be asked if they know anyone who can provide instrumental support such as giving a ride or
informational support such as knowledge about financial matters. As is evident from the phrasing
of the survey instrument, many network-based social capital researches also focus on social
support that a person could glean and gather from his or her social network.
While the concept of structural social support has received considerable criticism and has
almost been abandoned in recent work on social support, its conceptual meaning is very
intriguing. It focuses on potential support givers rather than the actual support givers that a
person has already recognized or anticipated with a certain degree of certainty. Although
structural social support is not quite welcomed by social support researchers, it is an important
component that health interventions usually adopt and make use of for health promotion
(Verheijden et al., 2005). Therefore, to develop effective intervention programs, it is essential to
understand whether increasing structural social support would lead to increased functional social
support. Moreover, as structural social support resembles one’s overall social network, it needs
more scholarly research to investigate the mechanism through which structural social support
(i.e., social networks) would influence health-related behaviors and outcomes (Zhu, Woo, Porter
& Brzezinski, 2013). This study argues that the psychological aspect of functional social support
(i.e., perceived social support) could be such a mechanism.
A few recent studies started to examine the relationship between structural social support
and perceived social support in both offline and web-based health interventions. Kroenke, Kwan,
Neugut et al. (2013) studied the influence of structural social support and perceived social
support on individuals’ quality of life among breast cancer survivors. The authors collected data
51

with respect to the presence of a spouse, the number of close friends and relatives, religious ties
and community ties, as well as perceived social support manifested in the availability of the four
types of assistance. They found that large social network size and greater perceived social
support predicted better quality of life. Among different components of one’s social network, the
number of close friends and relatives had the most weight in predicting higher quality of life.
More interestingly, perceived social support partially mediated the positive impact of the size of
the social network on participants’ quality of life. Another study examined the relationship
between structural social support and perceived social support in an online weight loss
community (Hwang, Etchegaray, Sciamanna et al., 2012). The authors operationalized structural
social support as the usage of social media tools such as discussion forums and blogs in the
community. They found that community members who used forums and blogs on average at
least once a week were almost five times more likely to receive actual social support for their
weight loss effort. In other words, the availability of potential support givers, enabled by the
usage of social media tools, positively predicted perceived social support.
It is encouraging to learn the positive and significant relationship between structural
social support and perceived social support. As mentioned earlier, structural social support is a
vital component that can be and has frequently been manipulated in weight management
intervention programs. That said, structural social support has great value in that health
professionals and researchers can work on, improve or engineer it in order to achieve better
weight loss outcomes (see review in Verheijden et al., 2005).
The current applications of structural social support in health have focused on the size of
social network and the frequency of communicating with social contacts in the network. Those
measures may constrain the ability of social network analysis to help understand the relationship
52

between structural and perceived social support. Social network privileges the web of social
relations around an individual. Social network studies actually map out individuals’ networks
and probe the impact of particular network components, rather than simply counting the number
of people who provide social support in an individual’s social network (Berkman, 1984; Smith &
Christakis, 2008; Wellman, 1990). The reason that social networks are worth studying is that
they are characterized by properties that are emergent from the whole, which cannot be
explained by the constituent parts. Put in a different way, the structural form of a network may
influence the flow of resources and have different implications than the size of the network in
perceived social support. Given that I have collected the complete network data of a large
component in an online health community, I have full information about to whom an individual
actor was connected. The sociometric data were not limited to a simple count of health buddies
that an individual actor has, but also contain complete information with respect to the
interconnections between the individual actor’s alters. This gives me an opportunity to derive the
structure of one’s ego network, and probe its relations to social support as actually perceived by
individual actors. This research effort contributes to the understanding of relationships between
one’s structural support network and perceived social support. More importantly, this study is
able to juxtapose structural and perceived social support in one theoretical model to examine
their effects on health behaviors and outcomes.
Egocentric Social Network and Perceived Social Support
Ego social network consists of “a focal actor, termed ego, a set of alters who have ties to
ego, and measurements on the ties among these alters” (Wasserman & Faust, 1994, p .42). Ego
network is also referred as personal network, which describes a network of a focal actor and
those who are directly connected (i.e., alters) to the focal actor. The size of one’s ego network is
53

the number of alters or connections that the focal actor has in his or her personal network. It has
been an important predictor included in all studies on the effects of one’s structural social
support on perceived social support and health-related outcomes (e.g., Kroenke et al., 2013; Zhu,
et al., 2013).
Size of egocentric networks.
The size of one’s ego network may influence the quantity as well as the variety of
resources and assistance that one could potentially glean from his or her networks. In other
words, the odds of receiving social support should be higher for a person who has more social
contacts than a person who has very few social contacts or is isolated. Individual ties and
connections are the conduits to transmit informational and emotional support. A number of early
studies in the 1980s and 1990s have examined the size of one’s offline ego network and
perceived social support. For example, Lin (1999) studied the size of one’s ego network in
community organizations, such as clubs and churches. He found that the number of a person’s
contacts in the community was positively associated with the person’s perceived instrumental
support and perceived expressive support. In addition, both the size of one’s ego network and
perceived expressive support was associated with a lower level of depressed mood. A more
recent study examined the relationships among network size, perceived social support and
subjective well-being in the population of first-year students in a public university (Zhu, et al.,
2013). The authors found that the number of contacts nominated as potential social support
providers was positively associated with the student’s perceived social support. The size of the
ego network had a positive effect on the student’s subjective well-being, mediated through the
student’s perceived social support.
54

In the context of online health communities, connections between community members
are usually created with a common interest in a health-related issue (Demiris, 2006). As
computer and web technologies provide communication infrastructure to community members,
each pair of them has a latent tie (Haythornthwaite, 2002). To choose someone as a health buddy
or health partner converts a latent tie into an actual and weak-tie relationship. The realization of
weak-tie relationships has an important implication to perceived social support. Weak ties refers
to a wide range of potential supporters who lie beyond the primary network of family and friends
(Adelman, Parks & Albrecht, 1987). As in Granovetter (1973)’s seminal work on the strength of
weak ties, weak-tie relationships could bring the focal actor non-redundant and diverse
information that is usually not available from one’s primary social relationships. In online health
communities, weak-tie relationships are great sources of health-related information as
community members are experiential experts. Given the shared health-related experiences, those
weak-tie relationships are also good at providing sympathy and encouragement because they
“have been there” (Adelman et al., 1987) and could relate to a person even better than his or her
primary social relationships. Computer-mediated social support groups have been demonstrated
to provide a variety of social support resources. It is reasonable to speculate that the more
relational partners a user has in an online health community, the more likely that the user would
perceive the availability of social support in that community. Therefore, the following hypothesis
is proposed:
H5: The size of one’s ego network in an online health community positively predicts his
or her perceived social support in that community.

55

Triadic closure in egocentric networks.
Beyond the size of the network, triadic closure is another important structural property to
describe one’s ego network. In the context of non-directed friendship networks, triadic closure
refers to the structure that A and B are friends, B and C are friends, and A and C are friends with
each other as well. Triadic closure can be a structural property of a sociometric network (i.e., the
whole, global network). For example, Figure 3 presents two networks in comparison. The
network on the left side has a lower degree of triadic closure than the network on the right side.
By forming the linkage between B and C, G and F, G and D in the network, there are more
closed triangles in the network on the right side.
Figure 3. Triadic Closure in Sociometric Networks (Easley & Kleinberg, 2010, Chapter 3, p. 47)

In an egocentric social network, triadic closure occurs when the focal actor is connected
to individuals who are also linked with one another. In other words, a person’s ego network
would present a high degree of triadic closure if the person’s alters are all interconnected with
one another. Figure 4 illustrates two ego networks in comparison. The ego network on the left
side has a lower degree of triadic closure than the ego network on the right side.
56

Figure 4. Triadic Closure in Egocentric Networks (actor A is the ego)

Triadic closure helps to create tightly-knit small groups, in which group members have a
high degree of mutual trust and share a strong group identity. In tightly-knit small groups, group
members tend to communicate more frequently with one another, do activities together, and their
beliefs and attitudes may reinforce those of each other. Thus, group members may become more
homogenous. Even though this type of small group is not good at diffusing innovative ideas or
disseminating diverse information due to its high homogeneity, it may be good at providing
social support due to its solidarity and trust among group members (Granovetter, 1973).
Literature on social support has emphasized the importance of strong ties in relation to perceived
social support. Lin (1999) theorized three layers of social relationships as belongingness,
bonding and binding relationships. Belongingness relationship is at the outer layer and it refers to
a broad range of community engagement, such participation in voluntary community
organizations. This layer of social relations does not require the actual person-to-person
interactions, but it provides a general sense of belongingness to the community. A sub-layer
within the outer layer is bonding relations referring to interpersonal interactions. These relations
57

tend to be stronger in that they require that egos and alters must invest in maintaining the
relations with certain amount of commitment to interactions. The inner-most layer only contains
a small number of alters and the ego through intimate and intense interactions. They constitute
binding relations because relational partners have strong norms of sharing and obligations to
reciprocate resources. The belongingness-bonding-binding formulation of social relations
presents a spectrum of the strength of social ties, with the binding relations as the strongest and
belongingness relations as the weakest. Lin (1999) found that in offline communities, bonding
and binding relations better predicted one’s perceived social support as well as the actual
received social support. This finding informed that stronger social relations may carry more
weight in providing and predicting perceived social support.
Moreover, triadic closure is an important network structure in theorizing social capital in
the form of network closure and brokerage (Burt, 2005; 2011). The networked approach of
theorizing social capital may also shed some light on the relationship between triadic closure in
egocentric networks, tie strength and perceived social support. Putnam (2001) proposed two
types of social capital, defined in relation to the strength of social relationships in which social
capital is embedded and accumulated. Bonding social capital is often linked to strong-tie
relationships with whom the focal person has more frequent interaction, as well as higher
emotional intensity and intimacy (Granovetter, 1973; Putnam, 2001). In contrast, bridging social
capital is often linked to weak-tie relationships with whom the focal person has less frequent
interaction, as well as lower emotional intensity and intimacy (Granovetter, 1973; Putnam,
2001). As weak ties may facilitate information diffusion and acquisition of new opportunities
because of the availability of non-redundant and diverse information (Granovetter, 1973), strong
ties are better suited for the provision of social support because of the degree of emotional
58

intimacy involved in the relationships (Granovetter, 1973). In Granovetter (1973)’s study, weak
ties were acquaintances while strong ties were family and friends. In his study, tie strength
served as a proxy measure of the proportion of shared contacts among a focal actor’s alters. That
is very similar to how I define triadic closure in one’s egocentric network. As simply explained
by Shen, Monge and Williams (2012), “weak ties are the people with whom one has few
common contacts, but strong ties are the people with whom one shares many friends” (p. 4).
Following the logic such that strong ties would be more effective in providing social support, it
is natural to speculate that a higher degree of shared contacts among alters in one’s ego network
would facilitate the provision of social support to the ego.
A few recent works examined the relationship between network closure and bonding
social capital in virtual worlds. Burt (2011) studied a large online social network in a virtual
world, Second Life, and demonstrated the social implications that different network structures
would have. This study confirmed that people who were embedded in more closed networks
tended to have more trust in one another. Similarly, Shen et al. (2012) studied the degree of
network closure of players in a virtual gaming community. She found that closure in players’
social networks was positively associated with their trust towards co-players in guilds. Trust is
an important indicator of bonding social capital (Coleman, 1988). These empirical studies
provide evidence that closure in virtual social networks indicates relatively strong social
relationships, as in the offline world.
In an online health social network, as discussed earlier in hypothesis 5, weak ties can be
more useful than weak ties in one’s life in the physical world because of the shared interests and
experiences in health issues. Yet, stronger ties in online health social networks may play an even
more important role in providing social support. If a person has a closely-knit network of health
59

buddies or health partners in an online health community, it is more likely that those health
buddies do activities together and care about each other enough to provide more and sustained
support. Take a weight management online community as an example: a tightly-knit group of
health buddies are more likely to discuss and check on each other’s updates on diet, exercise and
weight. Therefore, compared with a group of health buddies who have a very low degree of
shared contacts, health buddies in a tightly-knit group are more like to provide appraisal on a
group member’s persistence when that member is consistently tracking his or her weight
management activities. Similarly, health buddies in a tightly-knit group are more like to catch the
moment to provide encouragement to a group member when that member is working hard to lose
weight, yet is not seeing any improvement or progress. The mutual trust between members in the
closely-knit group also makes them believe that they would have someone to count on for
support and the availability of assistance is realistic. Therefore, the following hypothesis is
proposed:
H6: Triadic closure in one’s ego network in an online community positively predicts his
or her perceived social support in that community.
Following the argument of triadic closure, strong tie relations and closely-knit small
groups, triadic closure in one’s ego network may also influence one’s health behavior – self-
tracking behavior in this case. In Centola (2010)’s experiment of testing the effectiveness of
network structure in spreading health behavior (i.e., joining an online health community), he
manipulated network structure in two distinct forms. One was referred to as a clustered-lattice-
network and the other was a random network. The clustered-lattice-network was characterized
by a high level of clustering. The clustering coefficient refers to the probability that two
randomly selected friends of an ego are friends with each other. It is the same as the fraction of
60

pairs of an ego’s friends that are connected to each other by edges (Easley & Kleinberg, 2010).
In other words, the clustered-lattice-network had more triadic closure at the global level of the
network than the random network. Centola (2010) discovered that the clustered-lattice-network
was a better network for facilitating the diffusion of healthy behavior. He argued that a person’s
health behavior needs repeated reinforcement from his or her health buddies. Redundancy of
information about healthy behavior and the reinforcing norm of performing such healthy
behavior were critical in influencing the ego’s adoption of the healthy behavior. Although the
experiment was conducted at the network level, the finding and argument are consistent with the
idea that closely-knit ego networks helps to create homogenous health behaviors. As closely-knit
networks may reinforce both healthy and unhealthy behavior, it is not clear if closure in one’s
ego network would increase one’s self-tracking behavior. But it is interesting and important to
know if triadic closure tends to be a desired ego network structure in online health communities.
Therefore, this study attempts to explore the following research question:
RQ2: Will triadic closure in one’s ego network at an earlier time point predict one’s
number of self-tracking entries at a later time point?
Perceived Social Support and Self-tracking Behavior
Perceived social support has been shown to have a positive effect on human well-being
and health-related behavior. The power of social support comes from its ability to mitigate stress
when people face stressful life events (Cobb, 1976; Cohen & Wills, 1975; Uchino, Holt-Lunstad,
Smith, & Bloor, 2004). The two dominate theoretical perspectives that explain the relationship
between social support and perceived stress are the buffering hypothesis (Cohen & Wills, 1975)
and the direct effect hypothesis (Berkman & Breslow, 1983). The buffering hypothesis states that
the presence of adequate social support enhances an individual’s ability to cope with stressful
61

events, buffering the amount of psychosocial distress experienced (Bambina, 2007; Wallace,
2005). Therefore, social support exerts a positive influence on health by diminishing the negative
effects of stressful life events. The direct effect hypothesis states that rather than acting as a
buffer to stress, high-quality social support protects people who are experiencing stressful events
so that they do not suffer to the same extent as those who have very little social support
(Berkman & Breslow, 1983). Therefore, social support exerts a positive influence health by
directly ameliorating the stress a person is facing. Even though the two theoretical perspectives
propose alternative explanations, they both support the negative relationship between social
support and perceived stress.
Given the role of social support in mitigating perceived stress, how would it influence
self-tracking behaviors? Literature on self-regulation and resource depletion may provide a
useful lens to understand the relationship between received social support and self-tracking
behavior. In self-regulation, individuals make a conscious effort to align their behaviors with
established standards or preferred actions (Muraven, Tice & Baumeister, 1998). Self-regulation
is involved in a variety of tasks, from getting up in the early morning, to refusing to eat a
delicious dessert after dinner. Those effortful activities to achieve certain goals requires
expenditure of regulation resources. The regulatory depletion theory posits that each individual
has limited resources for self-regulation activities, and one self-regulatory activity may take up
the limited resources and cause the subsequent self-regulatory behavior to fail (Muraven et al.,
1998). For example, researchers have shown that people who initially exerted self-regulation by
suppressing their thoughts (Vohs & Faber, 2007), their feelings of frustration (Muraven et al.,
1998) and their behaviors by performing faking gestures (Baumeister et al., 2005), consistently
62

performed worse on the following self-regulatory behaviors such as refusing to purchase and
resistance to unhealthy food.
In the process of weight management, individuals are subject to several self-regulation
activities. For example, they may need to follow and adhere to a certain diet regimen, and
increase physical activities even if they are short of time and energy to so do. At the same time,
they may want to keep records of their diet, exercise, and weight to see if they are making
progress. Maintenance of self-tracking behavior can be stressful. A study interviewed both health
professionals and laypeople about their attitudes towards longitudinal health monitoring behavior
(Beaudin, Intille & Morris, 2006). While health professionals and laypeople agreed that self-
tracking behavior has some merits, they all mentioned that it could bring more stress and
frustration. The reasons are that self-tracking records may demonstrate a decline in one’s health
condition, which would be depressing or disempowering, and the actual behavioral records may
reveal how irresponsible they are for their own health (Beaudin et al., 2006). In other words, the
information being reported from self-tracking data are not only about one’s health condition, but
also about what kind of a person one is. In many cases, self-tracking behavior can become a
threat to self-image (Beaudin et al., 2006). In addition, health-related self-tracking behavior
seems to naturally incur or involve other self-regulatory activities. As in the case of weight
management, people often try to make an effort to diet and to exercise while they track their
weight. Thus, the stress incurred in the self-regulatory process may discourage self-tracking
behavior.
This is where social support may play a vital role in encouraging and sustaining self-
tracking behavior. People with great social support may perceive less stress when they are
practicing self-tracking behavior, and following and maintaining diet and exercise regimens. It is
63

possible that a person’s support providers may express understanding and sympathy about less
desirable results that the self-tracking records show. Then, the person may feel less frustrated or
threatened, negative feelings incurred by the self-tracking experience. It is also possible that a
person’s support providers may act as a companion to do weight loss together. Then, the
regulatory resources may not be consumed as much and fast as those regulatory activities alone.
The companionship and comradeship may turn self-regulatory activities into enjoyable and
pleasant experience, and thus people do not need to make a great effort to perform those
activities. In either scenario, social support should help to encourage self-tracking behavior.
Although previous studies did not test the effect of social support on self-tracking
behavior, they consistently showed that the availability of social support from one’s offline
social relationships positively predicts the maintenance of recommended health behavior. A
meta-analysis (DiMatteo, 2006) reported that emotional and practical social support from one’s
family members have the highest correlation with adherence to medical regimens. Even though
self-tracking behavior is somewhat different from medication adherence, they are closely related
behaviors under the umbrella of self-management, especially in the context of chronic disease
(McCorkle, Ercolano, Lazenby et al., 2011). Self-management is the ability to manage the
symptoms of living with a chronic condition; it is a lifelong dynamic process of self-monitoring
and self-evaluation (McCorkle et al., 2011). Important self-management skills include self-
tracking of one’s health condition, adherence to medical regimens, and forming partnerships with
health care providers (Lorig & Holman, 2003). Family members are able to monitor patients’
health behaviors as to whether they persistently follow recommended health practice. The
support they provide can help to remove the practical barriers and alleviate psychological stress
that may hinder the maintenance of self-management behaviors. A more recent study showed
64

that family and friends’ social support had a strong positive influence on self-management health
practices, including both care adherence and self-tracking of diet and exercise in managing
diabetes (Vaccaro, Exebio, Zarini & Huffman, 2014).
Even though those studies have primarily focused on offline social support, I believe
online social support can play a similar role in encouraging self-tracking behavior. Extensive
research on virtual health communities has shown that social support can happen through weak-
tie relationships (see Cummings et al., 2002; Walther & boyd, 2002; Wright, 2009). As discussed
in Chapter 3, in particular with respect to health-related purposes, computer-mediated weak-tie
relationships could outperform one’s offline support networks in that support providers in virtual
health communities are more experienced and knowledgeable about the health issue, and they
can provide more sympathy and understanding because of shared health experiences (see Wright,
2009; Wright, Rains & Banas 2010 for review). Based on strong theoretical and empirical
support, the following hypothesis is proposed:
H7: Perceived social support positively predicts the number of weight self-tracking
entries at a later time point.
Functional social support has been found to be related to psychological well-being, such
as depression (Wright, Rosenberg, Egbert et al., 2013), as well as physical health variables such
as recovery from disease, illness onset and mortality (Ikeda, Kawachi et al., 2013). Specifically
related to weight management, a systematic review of social-support-based weight management
interventions showed that functional support had a strong correlation with weight loss outcomes
(Verheijden, et al., 2005). More recent intervention programs also confirm the positive impact of
functional social support on physical activities. For example, Zook, Saksvig, Wu & Young
65

(2014) reported that family and friends’ social support increased physical activities among
female adults. Therefore, the following hypothesis is proposed:
H8: Perceived social support positively predicts health progress (i.e., the percentage of
weight goal achievement) at a later time point.
Self-tracking Behavior and Health Progress
Self-tracking behavior refers to actions that log one’s health-related information, such as
food intake and exercise activities. Self-tracking is an important component in many Web-based
self-management health intervention programs in physical activities (see review in Arem &
Irwin, 2011). In a meta-analysis of the effectiveness of a variety of intervention techniques in
changing adults’ physical activities, self-tracking was found to be one of the few techniques that
had consistent efficacy in increasing physical activities (Olander, 2013). The reason behind the
effectiveness of self-tracking is that it enhances an individual’ sense of control over their own
health behaviors. Such enhanced self-efficacy in taking certain actions to improve health is the
pathway through which self-tracking may positively influence health outcomes.
One recent study (Hwang, Ning, Trickey & Sciamanna, 2012) examined a commercial
online weight management community SparkPeople, which is a rather similar website to
FatSecret. The study used a retrospective cohort analysis of a systematic random sample of
members who joined the program during the first four months in 2008, and included follow-up
data through May, 2010. The main outcome was net weight change based on self-tracked weight.
The main finding was that active users had better outcomes, such that users who made weekly
weight entries had an additional 5 kg weight loss. More interestingly, after controlling for weight
entry days, the other website usage variables were not associated with weight loss. Those website
66

usage variables included log-in days, number of forum posts, number of days that a member
made a food entry, and number of friends that a user made in the community.
Randomized trial studies also yielded similar findings with respect to the effectiveness of
self-tracking. For example, Brindal, Freyne et al. (2012) designed 7 websites representing 3
functional groups characterized by different features. Self-tracking was one feature that was
included across all conditions of websites. The results found that consistent throughout all
randomized conditions, the level of use of the self-tracking feature predicted actual weight loss,
after controlling for age and initial BMI. Funk, Stevens et al. (2010) designed a web-based
program to address the issue of weight regain after successful weight loss. Similarly, the
researchers found that self-tracking of weight was associated with less weight regain. Given the
support of these empirical findings, the following hypothesis is proposed:
H9: The number of self-tracking at an earlier time point positively predicts health
progress (i.e., the percentage of weight goal achievement) at a later time point.
In summary, I hypothesize a structural model that contains five endogenous variables,
including perceived social support, two characteristics of egocentric networks, self-tracking
behavior and health progress at a later time point. In this model, ego network characteristics at an
earlier time point predict perceived social support, perceived social support predicts self-tracking
behavior and health progress at a later time point, self-tracking behavior at an earlier time point
predicts health progress at a later time point. In addition, to control autoregressive components,
variables at an earlier time point directly predict the same variables at a later time point (Frees,
2004).
Figure 5 presents the theoretical model hypothesized in Study II. Table 1 presents a
summary of all hypotheses and research questions proposed in both studies.
67

68

Figure 5. Hypothesized Model in Study II

H9
RQ2
H6
H5
H8
H7
Perceived
Social Support
Size
Self-Tracking
Health
Progress
Triadic
Closure
Size
Triadic
Closure
Self-Tracking
Health
Progress
An earlier time point
A later time point

69

Table 1. A Summary of Hypotheses and Research Questions
Sociometric Network Formation and Influence
H1a
Individuals are more likely to become online health buddies with others of the
same gender.
H1b
Individuals are more likely to become online health buddies with others of
similar age.
H2
Individuals are more likely to become online health buddies with others of the
same group.
H3a
Individuals are more likely to become online health buddies with others having
similar initial health status.
H3b
Individuals are more likely to become online health buddies with others having
similar health goal.
H3c
Individuals are more likely to become online health buddies with others of
similar health progress.
RQ1
Will individuals who present richer health information have more health
buddies than those who present less health information?
H4
In online health social networks, an individual’s health outcome will become
similar to the average health outcome of his or her health buddies in the
network.
Ego Network Effects and Social Support
H5
The size of one’s ego network in an online health community positively
predicts the one’s perceived social support in that community.
H6
The triadic closure of one’s ego network in an online health community
positively predicts the one’s perceived social support in that community.
RQ2
Will triadic closure of one’s ego network at an earlier time point predict the
one’s number of self-tracking at a later time point?
H7
Perceived social support positively predicts the number of self-tracking at a
later time point.
H8
Perceived social support positively predicts health progress at a later time
point.
H9
The number of self-tracking at an earlier time point positively predicts health
progress at a later time point.

70

CHAPTER 5 Method
Overview
The current dissertation employs mixed methods to collect data. It includes a longitudinal
data collection by extracting unobtrusive behavioral data from an existing OHSN, a one-time
online survey, and content analysis of profiles of users in that OHSN. To date, studies on social
media and health have employed diverse methodologies, including survey methodology (e.g.,
Chung, 2013; Lapinski, Anderson, Shugart & Todd, 2013), controlled laboratory experiments
(e.g., Smith, Hitt, Nazione, Russel, Silk & Atkin, 2013), randomized trials (e.g., Joseph et al.,
2012) and content analysis (e.g., Shim et al., 2011). It was in recent years that unobtrusive
behavioral data
2
have received more scholarly attention as an additional source of data to study
human behaviors in various virtual worlds (e.g., Contractor, 2013; Han, Wise, Kim et al., 2011;
Shen & Williams, 2011).
In the current dissertation, one advantage of extracted web data is that it provides
unobtrusive and direct assessment of individual networking and self-tracking behaviors. In
comparison with survey methodology, extracted web data do not rely on respondents’ subjective
recall of past behaviors, and thus can provide a more accurate and objective report on their past
activities. In addition, it provides better external validity to employ the combination of the three
methods in a naturalistic setting (Berkowitz & Donnerstein, 1982) than a controlled experiment.
The naturalistic setting is an advantage in the study design in that it provides individuals with an
authentic lived experience. Individuals are more genuinely motivated to self-track their health
statuses and make health buddies. Sense of networked community and perceptions of social
2
One may access unobtrusive behavioral data by requesting log data from the site administrator
or using computer programming techniques to extract data from the Web. The dissertation
adopted the latter approach. The site administrator advised that university policy (IRB) regarding
data collection of publicly available data be followed.

71

support often grow by participating over time. These factors are all important to test the
hypotheses and answer research questions proposed in the previous two chapters.
Test Bed: FatSecret
There are a number of sites that meet the definition of OHSNs in the present dissertation.
They provide services of information and emotional support, social networking features to help
create health buddy or peer networks, and quantified self-tracking tools for self-monitoring and
retrospection. Compared with specific diseases such as Parkinson’s disease and cancer, sites on
weight management usually allow for more direct and explicit health outcomes – the actual
weight – through the self-tracking system. As the goal of the dissertation involves an
investigation of network effects on individual health outcomes, I selected OHSNs designed for
the purpose of weight management.
FatSecret is one of OHSNs that aims to provide a platform for quantified self-tracking
and getting its users connected for the purpose of weight management. Similar sites include
specialized sites such as SparkPeople, Traineo, WeightLossBuddy, extraPounds, and general
sites with specialized communities devoted to weight management such as Wellsphere. There
was some exploratory work that had been done on FatSecret (Ma et al., 2011), which gave us
more basic information about the site. FatSecret was founded in 2007. Although it was originally
an Australia-based social network, FatSecret attained its current popularity in the U.S.
(Arrington, M., 2007; Kincaid, J., 2009). After becoming popular in the U.S., FatSecret
expanded its market across six continents and thirty-five countries, including U.S., Canada,
Mexico, Brazil, Chile, Argentina, China, India, Japan, Australia, New Zealand, South Africa and
a number of countries in Europe (www.fatsecret.com, 2014). As of November, 2013, FatSecret
had over 25 million registered members across the world (blog.fatsecret.com, 2013).

72

On FatSecret website, users are able to create personal profiles with photos, bios, displays
of health buddies. They can use self-tracking tools to keep a life-time record of weight, and a
food and exercise diary. All the self-tracking information is displayed on users’ profile walls
(Figure 6). Users are encouraged to post personal blogs and customized recipes as both personal
records and information that they want to share with other community members. In addition to
profile-related features and activities, interest or voluntary groups are organized as separate
message boards. These groups are self-organized and community members are allowed to join
any groups that they are interested in. In each group, discussions are carried out in threads.
FatSecret users can manually input their weight, diet and exercise information onto the website,
or they can use a mobile app to keep track of themselves anywhere and anytime. FatSecret has its
own Android and iOS app, and they both have good ratings from users.
The present dissertation only focuses on the FatSecret site in the U.S. According to Ma et
al. (2011), there were in total 107,907 users on the site in 2011, based on more than 6 months’ of
data collection. The authors claimed that they had extracted the majority and particularly the
complete largest connected component on the site. Among those users, 83% were public users
and the rest kept their profiles at different levels of privacy. Among the public users, about 50%
had less than two self-tracking updates. It meant that those users probably logged in once when
they registered on the site and came back to the site only once afterwards. The information is not
encouraging, as the number of registered users seems to be big but in reality around half of
registered users would not come back to the site. Yet, the information describes a normal and

73

Figure 6 An Example of Weight Tracking on FatSecret

74

consistent picture as in the majority of online communities (Courtois, Mechant, De Marez, &
Verleye, 2009). These basic statistics can provide some reference to the sample in the current
dissertation.
Data Collection
After getting the approval from the University of Southern California Institutional
Review Board (IRB), data collection started in April, 2013. There were three stages in data
collection. Stage one involved extraction of self-tracking behavioral data, health buddy network
data and weight outcome data from the website. A snowball sampling strategy was employed to
identify an initial sample in the online health social network. Snowball sampling is a widely-used
sampling strategy to sample networks in large-scale online social networking services (Ahn,
Han, Kwak, Eom, Moon & Jeong, 2007). Given the large-scale of network data on the web, node
and link random sampling strategies may generate a more biased sample than snowball sampling.
The reason is that it is very possible to randomly select not well-connected nodes or links that
give very little information except for only one neighbor (Ahn et al., 2007). In that case, the
network crawl algorithm is interrupted very quickly. Without a substantially long period of data
collection, the sample tends to underestimate node degree and not to find hubs in the network, if
using node or link random sampling (Ahn et al., 2007). In contrast, snowball sampling is a much
more feasible and efficient sampling method in the context of online social networking services.
Power-law nature in the degree distribution is well conserved under snowball sampling since the
method easily picks up hubs (Lee, Kim, & Jeong, 2006). According to Ma et al. (2011), among
the over 10,000 total users on FatSecret, the maximum number of friends a user had was 1002,
and he or she was the only one who had friends over 1000. The second largest number of friends
was 471, and only 24 users had more than 100 friends. From 2011 to 2013, these numbers must

75

have changed, but they provided some clues for picking up the initial seeds. By extensively
browsing the site, two users were found who had more than 1000 friends on FatSecret, and one
of them was randomly picked as the initial seed to start the snowball sampling process.
The project used a python programming technique to crawl the health buddy network on
FatSecret. After a few pilot tests of crawling on the site, on April 16, 2013, a python program
was run by crawling friends, friends’ of friends, and friends’ of friends’ of friends, etc. of the
initial seed. The crawling process was iterative until no new single user could be found.
Therefore, the health buddy network generated using the current snowball sampling process was
a large social network without isolates. If someone was an isolate, he or she would not be able to
be reached using the current sampling method. This sample network was a big component that
represented a large group of people who were connected to one another in the online health
buddy network of FatSecret. It could be a large fraction of the entire network of FatSecret, and it
typically contained most of the highly active and gregarious individuals (Kumar, Novak &
Tomkins, 2006). The crawling process took about one day, and this process gave me 20,667
users in total. Each user was given a unique ID number, and network data were stored in 20,667
separated text files. Each text file was a user’s ego network information in the form of an
edgelist.
After having the 20,667 users as an initial sample, the study used a python programming
technique to collect their self-tracking and weight outcome data. By tracing the usernames of the
20,667 users, their personal profiles were located. For each person, the program automatically
collected time-stamped weight and tracks, voluntary groups joined, start weight, goal weight and
registration data. To collect data on changes in social networks, self-tracking behaviors and
weight outcomes, the initial sample of 20,667 users were followed in the next three months.

76

Specifically, I repeated the above-described process to collect data on users’ health buddy
networks, time-stamped weight and tracks and voluntary groups joined on May 16, June 16 and
July 16 respectively. During May, June and July, there were more new users added on top of the
initial sample because during that time period, users made new friends beyond the initial sample.
New users were also assigned unique ID number continuous from 20,667, and their ego network
information was stored in additional text files. In total the four months data collection yielded
21,492 users. The data of these 21,492 users were subjected to the following data cleaning
process discussed in the next section.
Stage two of data collection involved an administration of an online survey to the 21,492
users on FatSecret. An internal message enclosed with the link to an online survey was sent to
the users through the FatSecret message system. The online survey was administrated on June 1,
2013, and was open to participants for two weeks. The survey was short and lasted for five
minutes. The online survey consisted of three sets of questions. The first set of questions asked
about perceived social support received in the FatSecret community; the second set of questions
asked about whether users had any pre-existing offline friends and families as their health
buddies in the FatSecret community; and the last set of questions asked for socio-demographic
information. Survey participants were asked to provide their usernames in the FatSecret
community to receive an incentive for participating in the survey. Participants were told that the
first two hundreds survey respondents would be rewarded with a $10 e-gift card from several
major retailers of sporting and exercise goods, such as Big 5 and Sports authority.
Stage three of data collection involved content coding to gather data on age, gender,
ethnicity and geographic location for a subsample of the 21,492 users. This stage of data
collection was conducted after the data cleaning process (described in next section). Two

77

undergraduate coders were recruited to do the content coding task to earn credit for independent
study. The coders were trained on basic procedures of content coding. FatSecret users may
include personal information about age, gender, ethnicity and geographic location in their brief
bios posted on personal profiles. In many cases, information about gender could also be inferred
from users’ profile picture. Given the clarity of the variables to be coded, the two coders quickly
reached a favorable internal agreement in the training session. Cohen’s Kappa was used to
calculate inter-coder reliability. The two coders had an average score of .96 for the actual coding
of the subsample.
Table 2 presents the timeline and methods of data collection.

78

Table 2. Timeline of Data Collection and Variables Summary
April May June July August-
September
Study I Study II
Web Data Extraction
Health buddy networks X X X X
Self-tracking behaviors X X X X
Weight tracking records
(Current weight)
X X X X
Start weight X
Goal weight X
Registration date X
Voluntary groups X X X X
Content Coding
Age X
Gender X
Ethnicity X
Geographic Location X
Online Survey
Perceived Social Support X
Migration of offline relationships X
Usage of Internet and FatSecret X
Socio-demographics X

79

Data Cleaning
Data cleaning involved of a series of steps. First, among the 21,492 users, 80.87% were
user who set their profiles as fully public. The rest of the users set their profiles as either
completely private, or partially private. As agreed with IRB, the current dissertation only studied
users who hold fully public profiles on FatSecret community. This resulted in a usable dataset of
17,380 users in total. Second, to reduce the heterogeneity in purposes of using FatSecret
community, the study only selected users whose goal was to lose weight and excluded users
whose goal was to gain weight or maintain weight. This step involved a subtraction of one’s goal
weight from one’s start weight. If the result was above zero (i.e., start weight – goal weight > 0),
it meant that goal was to lose weight; otherwise, the goal was to gain or maintain weight (i.e.,
start weight – goal weight <= 0). This step left 16, 295 users in total. The third step aimed to
identify “relatively meaningful” users who actually participated in FatSecret Community. Data
on weight self-tracking records were used to help with the identification process. When
registering with the community, everyone was asked to report their initial weight as a starting
point of weight loss. That report was considered as one self-tracking record. Very likely, people
may not even provide that start weight when they joined the community. It’s possible that some
people were just trying out to see what the site was about, or they may have considered losing
weight but were not determined enough to come back to the community, or they did not intend to
lose weight at all. Therefore, users who had no or only one self-tracking record since they joined
FatSecret were excluded from the sample. This step left13, 787 users, or 79% of public users.
The fourth step moved onto identification of active users during the data collection
period. Among the 13,787 users, only 7,445 users updated their weight three months before the
first data collection date. In other words, the rest of the users had not been publicly active in

80

weight loss activities since three months before April, 2013. And 3008 users had at least one
weight update during the data collection period. Given that the purpose of the first study was to
examine the joint dynamics of network selection and network effects, users who had a weight
self-tracking record at each time point of the data collection were selected because weight
outcome as a dependent variable was needed at each time point for data analysis. This procedure
left1023 users. Moreover, given its large scale, online relational networks are usually sparse
(Kumar, Novak, & Tomkins, 2006). The rate of tie change per month within a large component
of the entire network can be extremely low due to the inactivity of tie establishment and the base
number of edges in the large component. Given that the purpose of the first study focused on the
changes in networks, users who had at least one change in their network ties at any time point of
the data collection were selected for further data analysis. Thus, in total there were709 users left
in the final subsample. In sum, the final subsample contained users who had weight updates at
every time point of the data collection, and had at least one change in their ties during the entire
period of data collection.
Measurement
Measures from Web Crawling Data
Online health buddy network.
As noted earlier in the data collection section, each user’s ego network information was
stored in a separate text file, in the form of an edgelist. From these text files, a global network of
the subsample of 709 users was derived for each time point of data collection. The global
network was generated in both forms of the edgelist and adjacency matrix. Therefore, I had four
709 × 709 adjacency matrices for April, May, June and July respectively. Data transformation
was performed using MatLab software.

81

Ego network measures.
In addition to the global network, several ego network measures were calculated based on
information about connections between all public users. I used an R package called egonet
(Sciandra, Goiachin & Finos, 2013) to generate ego network measures. The egonet package
provided a convenient tool dealing with ego-centric network measures, including size,
constraints, transitivity etc. It also allowed for an import code suitable for a large number of
adjacency matrices (Scinandra et al., 2013). Specifically, degree and transitivity were calculated
using package ‘egonet’. Degree was the number of health buddies that a user had in his or her
ego network. It served as a measure of the size of one’s ego network. Density was measured by
the number of ties divided by the number of pairs in his or her ego network (Hanneman &
Riddle, 2005). It served as a basic network measure for description of the networks.
To measure the interconnectedness of alters in one’s ego network, the network measure
transitivity was adopted (O’Malley, Arbesman, Steiger, Fowler & Christakis, 2012). An ego’s
transitivity is the average value of the relationship between all pairs of alters, herein assumed to
be mutual (i.e., the relationship from j to k is the same as that from k to j). When relationships
are binary-valued (e.g., corresponding to the presence of any relationship at all), transitivity is
given by
𝑡𝑡 𝑡𝑡 𝑡𝑡𝑡𝑡𝑡𝑡
𝑖𝑖 = 2 𝑡𝑡 𝑖𝑖 − 1
( 𝑡𝑡 𝑖𝑖 −1)
− 1
� 𝐼𝐼 ( 𝑍𝑍 𝑗𝑗𝑗𝑗
>0)
𝑗𝑗 < 𝑗𝑗
in which the degree of ego i by n
i, the strength of the relationship between alter j and alter k by
Z
jk, and I is contingent on the value of the relationship. If Z jk is present, then I = 1; if Z jk is non-
present, then I = 0. Then, transitivity is interpreted as the proportion of pairs of alters for which
some form of relationship exists.

82

Frequency of self-tracking behavior.
Two types of measure for frequency of self-tracking behavior were generated. The first
measure was the average number of self-tracking records per week since a user joined FatSecret.
It was calculated by using the total number of self-tracking records divided by the total number
of weeks of the individual’s tenure on FatSecret. The second measure was the average number of
self-track records per week in the past month. It was calculated by using the total number of
self-tracking records divided by four weeks.
Group affiliation network.
The web crawling program collects names of voluntary groups that each user had joined
on FatSecret. Each voluntary group was then assigned a unique ID. Based on each user’s lists of
groups, a group affiliation network was constructed. The network had users as nodes and edges
representing co-affiliation to the same voluntary group. Specifically, the value of 0 indicating an
absence of the edge, such that two users did not belong to any of the same groups; the value of 1
indicating an existence of the edge, such that two users had at least one joint group membership
in common. The group affiliation network was created for each time point in the data collection.
Thus, there were four 709 × 709 adjacency matrices for April, May, June and July respectively.
Data transformation was performed using MatLab software.
Health-related homophily measures.
Initial health condition was operationalized as start weight and health goal was
operationalized as how much weight a person aimed to lose. The raw data of start weight
collected from each person’s profile were used as the exact measures of initial health condition.
Then, I used the difference between one’s start weight and goal weight (i.e., start weight – goal
weight) as a proxy of health goal. Given that everyone had a somewhat different initial weight to

83

start the weight loss process, it was more reasonable to consider two persons shared a similar
health goal if they both wanted to lose 50 lb. The absolute number of the goal weight may not
accurately reflect one’s goal as well as the difference between start and goal weight would. Thus,
each individual user in the final subsample had one score for initial health condition and one
score for health goal.
Health progress was operationalized as the percentage of goal achievement or attainment.
To calculate the score for health progress, each user’s current weight was extracted from self-
tracked weight records. In the final subsample, users may have from 1 to 29 new weight records
at each time point of data collection. The latest one before the date of data collection was used as
a proxy of “current weight” for each month. Scores for health progress were calculated using the
following formula: (Start weight – Current weight) / (Start weight – Goal weight). The
calculation was repeated for at each month, and it gave each user scores on health progress at the
four waves of data collection.
Weight outcome.
The dependent variable – weight outcome – was a binary measure of the success of
weight loss. It was calculated by using the difference between one’s start weight and the current
weight at each time point of data collection. Specifically, it was “start weight – current weight”.
If the score was above zero, then a user’s weight outcome was coded as 1, indicating that the
user was on the right track by losing weight. If the score was equal to or below zero, then a
user’s weight outcome was coded as 0, indicating that the user was not on the right track because
he or she was not losing weight. The calculation was repeated for each month. This process
produced weight outcome scores for 709 users at the four waves of data collection.

84

Measures in Online Survey
Weight management social support.
To best fit the current study context, I chose to use a validated scale of social support
specifically for weight management (Rieder & Ruderman, 2007). The original weight
management social support scale contains 26 items on the frequency and helpfulness of receiving
four types of social support for weight management. The four types of social support are
emotional, informational, instrumental and appraisal social support. The scale was re-validated in
an online weight loss community, SparkPeople, which was a very similar online health social
network site to FatSecret (Hwang, Ottenbacher, Lucke, Etchegaray, Graham & Thomas, 2011).
The researchers found that the scale was appropriate for measuring social support among
members of Internet weight loss communities, with the exception of items of instrumental social
support. It turned out that instrumental social support was not relevant to participants in virtual
weight loss communities. Therefore, the current online survey dropped scale items for
instrumental support, and used modified items of the other three types of social support (Hwang,
et al., 2011).
The instruction statement asked respondents to rate on a 5-point scale how frequently
they experienced a certain interaction with others who were members on FatSecret in the past 4
weeks, ranging from 1– never to 5 – daily. Example items are “Other FatSecret members tell me
about the calorie or fat content of foods” (informational support), “Other FatSecret members tell
me about the things that they have done to lose weight” (informational support), “Other
FatSecret members tell me that they are confident that I can lose weight” (emotional support),
“Other FatSecret members listen to my concerns about the difficulty of dieting”, and “Other

85

FatSecret compliment me on sticking to an exercise routine” (appraisal support). The scale had a
good reliability with a Cronbach’s alpha at 0.94, M = 3.25 (SD = 1.02).
Migration of offline social networks.
Survey participants were asked if they had pre-existing offline friends and family
members as their health buddies in FatSecret. The answer options were Yes, No (Cummings et
al., 2002).
Internet use.
The survey asked participants about their general Internet use and different activities in
using FatSecret. With respect to general Internet use, participants were asked how frequently
they used the Internet for anything in the past 4 weeks. With respect to different activities in
using FatSecret, participants were asked how frequently they (a) “posted messages on forums”,
(b) “wrote journals”, (c) “posted recipes”, (d) “looked for information from fitness, diet and
nutrition database”, and sent “private messages” in the past 4 weeks. Response options were
daily (=6), two to three times a week (=5), once a week (=4), two to three times in last month
(=3), Once a month (=2), never (=1). These measures were used in Chung’s (2013) study of
usage of health-related social network sites.
Measures in Content Coding
Age, gender, ethnicity and geographic location were coded from users’ profiles. For age,
the actual numerical age indicated in each user’s bio was used as the measure of age. Even
though the real age of a user may differ from the number provided in the bio, the number
displayed was the one visible to other community members. Therefore, that number served the
cue of age similarity or dissimilarity when community members made decisions about creating
the tie or not. Similarly, gender was coded from the user’s bio. Sometimes, coders were able to

86

refer to the profile picture if gender information was missing in the bio. Male was coded as 0 and
female was coded as 1. Information on age and gender could be identified for over 90% of the
final subsample of users. In addition, both ethnicity and geographic location were also coded
from the bio. However, there was a substantial amount of missing information with respect to
ethnicity and geographic location. Only 23% and 53% of the final subsample, respectively,
included ethnicity and geographic information in their bios. Due to the percentage of missing
data on these two variables, they were not included in the social network analysis.
Analysis
To test hypotheses and answer the research question proposed in the first study,
Stochastic Actor-Based (SAB) models, implemented in RSIENA computer program (Macy &
Willer, 2002; Ripley, Snijders & Preciado, 2013; Snijders, van de Bunt, & Steglich, 2010) were
used. SAB models are one social network analytical method that captures network selection and
network influence processes simultaneously. They allow for the estimation of parameters for
factors that influence network and behavioral change over time. RSIENA is a computer package
that supports the implementation of the SIENA computer program in R environment. SIENA
uses network-behavior panel data and model fitting procedures to estimate parameters for
making inferences about the mechanisms driving the network and behavioral change process.
There are three major advantages of SIENA in testing hypotheses with respect to
longitudinal network and health outcome dynamics. First, SIENA allows for testing multiple
micro-mechanisms that drive the formation of social ties (Snijders et al., 2010). Each micro-
mechanism, such as homophily and social exchange theories, can be tested while controlling for
the simultaneous operation of each other. Second, network dependency inherent in modeling
changes of network ties violates the assumption of independence in standard statistics. SIENA is

87

able to take into account the interdependence of individual decisions of creating social ties within
a network. In other words, SIENA allows us to represent dependencies between network ties as
the result of processes where one tie is formed as a reaction to the existence of other ties.
Moreover, it enables the inclusion of network endogenous properties in the model so that the
model will not misdiagnose or overestimate the effect of hypothesized micro-mechanisms in
influencing the global network dynamics.
Third, a SIENA model enables us to disentangle selection effect from social influence
(Steglich, Snijders & Pearson, 2010). The underlying change processes operates in a continuous-
time manner, while traditional statistical analysis of panel data depends on incomplete and
discrete observation, which may easily lead to misinterpretation of the change mechanisms
(Steglich et al., 2010). The actor-based stochastic model considers the compound change that is
observed between two observations as resulting from many small, unobserved changes that
occurred between the observation moments. In other words, it decomposes the change process
into its smallest steps to obtain a relatively parsimonious model. Those small, unobserved
changes are modeled using a simulation-based process.
A SIENA model follows the objective function that is defined by researchers with respect
to the generation mechanisms of a network. The objective function represents the kinds of ties
that actors value in a network, such as ties to actors who share similar health status, or actors
who have already had many ties in the network. The program uses Markov Chain Monte Carlo
(MCMC) simulation to produce a distribution of networks that would be likely to occur based on
the generation mechanisms defined in the objective function (Snijders et al., 2010). The goal of
the simulations is to obtain a converged and well-fitting model. A model is well fitting if the
converged objective function values generate the simulated networks that closely approximate

88

the observed networks (Snijders et al., 2010). The SIENA model is particularly well-suited for
estimating gradual change in networks (Snijders et al., 2010; Steglich et al., 2010). The number
of changes between any two consecutive waves of networks should not be too high. Snijders et
al. (2010) suggested using the Jaccard index to assess the rate of change in networks. It is
suggested that values higher than 0.6 are preferable for SIENA models as no dramatic changes
occurring in networks and the objective function are more likely to be constant across different
time points of observations.
To test hypotheses proposed in the second study, the dissertation uses structural equation
modeling techniques (Bollen, 1989; Byrne, 1998) with PC LISREL 8.0. To evaluate the
proposed structural equation model, three methods were used. First, we examined chi-square,
degrees of freedom, and the ratio of chi-square to degrees of freedom, and as a rule of thumb, a
ratio that is less than 5 is considered as an indicator of good fit. The second method uses
comparative fit index (CFI), and adjusted goodness-of-fit indices (Joreskog & Sorborn, 1989).
The adjusted goodness-of-fit (AGFI) is the goodness-of-fit index (GFI) adjusted for degrees of
freedom and it penalizes for additional parameter inclusion and less parsimonious models
(Mulaik et al., 1989). Third, we examined the root mean squared residual (RMSEA), and a value
less than .08 justifies the adequacy of the model fit. In addition, a ratio of estimated coefficient
to standard error was used to test the significance of individual paths. A value greater than 2 was
considered as significant at the .05 level.

89

CHAPTER 6 RESULTS
Study I
Preliminary Analysis
For the final subsample of 709 users, 90% and 93% had valid gender and age information
indicated in bios on their personal profiles. There were 72% female users and 18% male users.
The average age of users was 39.08 (SD = 11.12, with a minimum of 22 and maximum of 60).
On average, users’ start weight was 210.24 lb. (SD = 54.75, with a minimum of 104.00 lb. and
maximum of 451 lb.), and users’ goal weight was 157.87 lb. (SD = 31.62, with a minimum of 98
lb. and maximum of 300 lb.). By subtracting a user’s goal weight from his or her start weight, it
showed that on average, users aimed to lose 52.27 lb. (SD = 39.41, with a minimum of 5 lb. and
a maximum of 180 lb.). These numbers seemed to fall within a fairly normal range for weight
loss.
Accumulatively since each user joined FatSecret up to July 16, 2013, the average
frequency of self-tracking was 1.74 (SD = 1.65) records per week, the average number of
voluntary groups in which users participated was 3.48 (SD = 3.71), and the average health
progress was 0.36 (SD = .42) which meant that average users had achieved 36% of their health
goal. Somewhat encouraging, 64% of users were on the right track by losing weight, whereas
36% users either gained weight or kept steady at their start weight, at the last time point of data
collection in July. In addition, among the 709 users, the average tenure for membership was
17.41 months (SD = 12.13, with a minimum of half a month and a maximum of 71 months).
Specifically, 50% joined FatSecret 12 months before the first data collection, and 79.5% joined
the FatSecret 24 months before the first data collection.

90

Descriptive statistics comparing across four time points of data collection are presented in
Table 3. In April, 2013 (T1), users had 2.26 (SD = 3.13) weight self-track records per week in
the past month; in May, 2013 (T2), users had 2.43 (SD = 3.02) weight self-track records per
week in the past month; in June (T3) and July (T4), users had 1.72 (SD = 2.11) and 1.75 (SD =
1.93) self-track records per week in the past month respectively.
Table 3
Descriptive Statistics of Self-Tracking, Groups, Health Progress and Weight Outcome across
Four Time Points of Data Collection (N = 709)
T1 T2 T3 T4
M (SD) M (SD) M (SD) M (SD)
Frequency of self-tracking 2.26 (3.13) 2.43 (3.02) 1.72 (2.11) 1.75 (1.93)
Number of groups 3.02 (3.25) 3.22 (3.54) 3.28 (3.65) 3.48 (3.71)
Health progress 0.28 (0.53) 0.29 (0.44) 0.34 (0.37) 0.36 (0.42)
Weight outcome 0.62 (0.48) 0.62 (0.49) 0.63 (0.47) 0.64 (0.47)
Note. T in the first row represents time point of data collection. M represents Mean and SD
represents Standard Deviation.

Moreover, users had an average of 9.66 health buddies (SD = 12.21, with a minimum
number at 0 and a maximum number of 132) within the network of subsample (N = 709) in July,
2013. Figure 7 presents the degree distribution of the online health buddy network for the final
subsample.

91

Figure 7. Degree Distribution for the Final Subsample (N = 709)

To compare across four time points of data collection, Table 4 presents information on
degrees, density and the rate of tie change. The online health buddy network had a very low
density around 0.02 to 0.03, which is consistent to the density found in other kinds of online
networks, such as collaboration networks in open source communities (Shen & Monge, 2010)
and trading networks in online gaming communities (Shen, 2011).
Figure 8 visualizes the network of the final subsample. Blue dots represents community
members whose weight outcome were 1 (i.e., losing weight) and red dots represents community
members whose weight outcomes were 0 (i.e., gaining weight).

0
20
40
60
80
100
120
140
160
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Frequency
Degree

92

Figure 8. Visualization of the Network Using Gelphi (Blue dots represent members who
lost weight while red dots represent members who either gained weight or kept initial weight)

93

As discussed earlier in the Methods chapter, one assumption in SAB models is that the
network change process needs to be gradual. The number of tie changes between any pair of
consecutive waves should not be too high. Snijders et al. (2010) suggested using the Jaccard
index to calculate the amount of change between two waves by

N
11
N
11
+ N
01
+ N
10

where N11 is the number of ties present at both waves, N01 is the number of ties newly created,
and N10 is the number of ties terminated. Snijders et al. (2010) recommended that the Jaccard
value between two consecutive waves should preferably be higher than 0.30, unless the first
wave has a much lower density than the second. The densities of the online health buddy
network were rather similar across waves, and the Jaccard index between each two consecutive
waves was higher than 0.30. It indicatess that change in the network was not dramatic, but
gradual.
Table 4.
Descriptive Statistics of Degree, Density and the Network Change (N = 709)
T1 T2 T3 T4
Mean of Degree 7.08 8.11 8.87 9.66
Median of Degree 3.00 3.00 4.00 4.00
SD of Degree 9.55 10.56 11.52 12.21
Density 0.02 0.02 0.03 0.03
Jaccard index -- 0.86 0.90 0.89

94

Hypothesis Testing
To test H1a to H4, and answer RQ1, a few steps of statistical analysis were conducted.
Table 5 presents the results of each step in the SIENA analysis. First, network endogenous
variables were added into the model. The results showed that density had a significantly negative
coefficient at -8.76, p < .05, indicating that individual actors would not like to make health
buddies at random. Instead, they followed certain mechanisms to decide with whom they would
connect. As in other relational social networks, network structures transitivity (PE = 1.21, p <
.05) and preferential attachment (PE = 0.76, p < .05) were significant in online health buddy
networks. This meant that individual actors were more likely to become health buddies if they
shared a common health buddy, and individuals who were already popular in the network would
have more health buddies. These results were consistent with the classic phenomena of “friends
of friends become friends”, and “the rich get richer”. In addition to density, transitivity and
preferential attachment, another network structure “IsolateNet” was significant, PE = 3.77, p <
.05, in predicting the formation of health buddy ties. The purpose of adding this network
structure was to get a converged model in SIENA. It turned out that there was a tendency to
being an isolate in the current subsample of the network. The aforementioned results were
presented in Model 1.
In the second step, to test H1a to H3c and answer RQ1, I added homophily-related
variables including similarities in age, gender, group membership, start weight, health goal and
health progress into Model 2. One control variable,
3
similarity of tenure in the FatSecret
3
More than one control variable was tested initially, including individual attributes such as
tenure in FatSecret, age, gender, start weight, health goal and health progress. Except for tenure
similarity in FatSecret, the other control variables were not significant. Age and gender did not
have a significant effect on whether an individual actor would have more health buddies in
FatSecret. In addition, one’s start weight, health goal and health progress did not make him or

95

community, was also included in the analysis. The results indicated that tenure similarity had a
significant effect on tie formation, PE = 0.95, p < .05, indicating that users were more likely to
become health buddies with those who joined the community around the same time.
With respect to demographic homophily, H1a and H1b posited that individual actors
would be more likely to seek health buddy ties with those who were of similar age and gender.
The results showed that if two individual users were of similar age, the chance of their becoming
health buddies was 2.53 times (PE = 0.93, exp (0.93) = 2.53, p < .05) greater than those with
different ages. Similarly, the chance of two individuals with the same gender becoming health
buddies was 0.99 times (PE = 0.69, exp (0.69) = 1.99, p < .05) higher than those with different
gender. Therefore, H1a and H1b were supported.
With respect to inbreeding homophily, H2b posited that individual users were more likely
to form ties with those who share the same group membership in the FatSecret community. The
results showed that similarity in group membership indeed had a significant impact on the
formation of health buddy relationships, PE = 1.98, SE = 0.20, p < .05. Specifically, users’ being
in the same interest group had a 7.24 (i.e., e
1.98
) times greater odds of having a tie than users not
sharing any group membership. Thus, H2 was supported as well.
With respect to health-related homophily, H3a to H3c posited that similarity in initial
health status, health goal and health progress should play significant roles in predicting the
formation of ties. The results showed that initial health status did have a significant effect, PE =
3.65, SE = 1.50, p < .05, such that individual users who shared similarity in initial health status
her more likely to have more or fewer health buddies. To keep a parsimonious model in the
SIENA analysis, I only included tenure in FatSecret as a control variable in the following
analysis. Too many parameters in the SIENA model may lead to the problem of getting a
converged model. Parsimony is an advantage of SIENA models (Snijders et al., 2010).

96

had a 38.47 times greater chance to have a tie. It indicated that similarity in start weight was an
important factor that users considered in choosing health buddies. However, it turned out that
health goal was not a significant predictor, PE = -1.11, SE = 0.94, p > .05. Individual users did
not necessarily tend to become health buddies with those who had similar health goals to
achieve. Health progress was also a strong and positive predictor of tie formation, PE = 5.37, SE
= 2.70, p < .05, indicating that individuals who were at the same stage of achieving their health
goals were more likely to become health buddies. Therefore, H3a and H3c were supported while
H3b was not supported.
RQ1 asked if individuals who had relatively richer health information posted on their
profiles would have more health buddies. Results indicated that users who frequently update
their personal health information had 17% (PE = 0.16, exp (0.16) =1.17, p < .05) more health
buddies than users with less frequent health information tracking. Yet, the parameter become
marginally significant after adding the variable of social influence in Model 3.
In the third step, Model 3 added the weight outcome variable to test the social influence
hypothesis, while controlling for the network selection factors that had been included in Model 2.
The result showed that the parameter for average similarity on weight outcome was positive and
significant, PE = 0.32, SE = 0.15, p < .05. It meant that an individual increased the odds of losing
weight by 38% (exp (0.32) = 1.38) if his/her health buddies were losing weight on average. The
model was converged and well-fitting, with all convergence t-ratio less than 0.1.

97

Table 5
Results of SIENA Analysis for Study I
Model 1 Model 2 Model 3
Convergence
t-ratio
Hypotheses Parameter Ets. SE Ets. SE Ets. SE
Network Formation
Rate of change period 1 1.12 * 0.23 0.98 * 0.23 1.28 * 0.17 0.01
Rate of change period 2 0.97 * 0.22 0.78 * 0.21 1.05 * 0.17 0.02
Rate of change period 3 0.86 * 0.13 0.67 * 0.13 0.57 * 0.09 0.00
Density -8.76 * 0.28 -7.81 * 0.33 -6.99 * 0.06 0.03
Transitivity 1.21 * 0.25 0.62 * 0.27 0.48 * 0.14 0.05
Preferential attachment 0.76 * 0.07 0.30 * 0.07 0.26 * 0.07 0.06
IsolateNet 3.77 * 0.12 3.75 * 0.96 2.37 * 0.80 0.01
Control: Tenure similarity 0.95 * 0.32 0.61 * 0.20 0.07
Homophily
H1a Age 0.93 * 0.29 0.91 * 0.23 0.07
H1b Gender 0.69 * 0.24 0.60 * 0.20 0.07
H2 Group membership 1.98 * 0.20 1.97 * 0.20 0.04
H3a Initial health condition 3.65 * 1.50 2.65 * 0.88 0.00
H3b Health goal -1.11 0.94 -1.66 0.87 0.02
H3c Health progress 8.94 * 3.82 5.37 * 2.70 0.01
Information Resource
RQ1 Frequency of Self-tracking 0.19 * 0.08 0.15 + 0.08 0.01
Weight Outcome
Rate of change period 1 0.99 * 0.18 0.03
Rate of change period 2 0.74 * 0.12 0.09
Rate of change period 3 0.69 * 0.16 0.03
Weight outcome linear shape 0.10 0.14 0.00
Weight outcome quadratic shape -2.33 * 0.20 0.00
H4 Weight outcome average similarity 0.32 * 0.15 0.04
Note. * p < .05, + p < .10. Ets. represent parameter estimations.

98

Study II
Preliminary Analysis
In the second study, I received 357 completed surveys. Table 6 presents descriptive
statistics about the sample. The average age of respondents was 36.3, with 92% being female and
93% being White. The average BMI was 30.6 (SD = 7.4), indicating that the respondents in this
sample were overweight. The characteristics of the current sample in age, gender and BMI were
very similar to self-selected survey respondents in recent published works (e.g., Chung, 2013;
Hwang et al., 2013; Hwang et al., 2011).
This sample generally used Internet frequently, ranging from two to three times a week to
daily on average. With respect to the usage of specific technological features on FatSecret, users
posted messages on forums about once a month, wrote journals about their progress on weight
loss around two to three times in the past month and searched information from nutrition, diet
and fitness databases a bit more than once a month. In comparison, users were not very engaged
with activities such as posting recipes and sending private messages to one another. In this
sample, 32% of respondents said that they had offline family members and/or friends who were
also in their health buddy network on FatSecret. Table 6 presents zero-order correlations between
variables in Study II.
Figure 9 presents the statistics of start weight and goal weight of the survey sample.

99

Table 6
Descriptive Statistics of Survey Sample
Mean SD %
Age 36.3 9.7
Gender (= Female) 92%
Ethnicity (= White) 93%
BMI 29.6 7.4
Internet use 5.2 0.88
FatSecret Use
Post messages on forums 2.04 1.81
Write journals 2.97 1.06
Post recipes 1.33 1.12
Look for information from database 2.26 0.79
Send private messages 1.54 0.88
Perceived social support 3.53 1.27
Migration of offline social networks (Yes = 1, No = 0) 32%

Figure 9 A Description of Difference between Start Weight and Goal Weight (N = 305)

0
50
100
150
200
250
300
350
400
450
1
11
21
31
41
51
61
71
81
91
101
111
121
131
141
151
161
171
181
191
201
211
221
231
241
251
261
271
281
291
301
311
321
331
341
351
Start weight Goal weight

100

Table 7
Zero-Correlations of Variables
1 2 3 4 5 6 7 8 Mean SD
1 T2 degree 1

17.7 27.68
2 T2 transitivity -.47** 1.00

0.18 0.24
3 T3 degree .99** -.47** 1

19.96 22.49
4 T3 transitivity -.46** .95** -.47** 1

0.14 0.21
5 T2 health progress 0.09 -0.05 0.1 -0.07 1

0.39 0.49
6 T3 health progress 0.08 -0.50 0.09 -0.07 .90* 1

0.45 0.46
7 T2 self-tracking .09+ -0.10+ 0.13* -.14** .20** .26** 1 6.77 7.52
8 T3 self-tracking 0.09 -0.04 0.09 -0.03 .05 .10+ 0.74** 1 6.88 8.23
Note. + p < .10; * p < .05; ** p < .01.

101

Hypothesis Testing
At the global level, the hypothesized model did not have an acceptable fit: χ
2
(19) =
39.53, p < .001, with other fit indices CFI at 0.99, GFI at 0.97, AGFI at 0.90, and RMSEA at
0.09. The chi-square statistic was large enough that the p value indicated a significant difference
between the hypothesized model and the relationships in the observed data. The following
sections examine tests of the individual hypotheses at the local level in order to determine how
the structural model might be improved. The results were reported with standardized path
coefficients. Figure 10 presents the results of parameter estimations in the hypothesized model.

102

Figure 10 Parameter Estimation for the Hypothesized Model

0.88
1.00
0.95
0.09
0.76
Triadic
Closure
0.10
0.22
0.08
0.05 ns.
0.17
Perceived
Social Support
R1
R2
R3
R4
R5
Size
Triadic
Closure
Self-
Tracking
Health
Progress
Size
Self-
Tracking
Health
Progress

103

H5 stated that the size of one’s ego network in an online health community positively
predicts one’s perceived social support in that community. The result showed that the path from
degree of ego network to social support was not significant, β = 0.08, p = 0.06. It indicated that
the number of health buddies that an individual had on FatSecret did not significantly contribute
to increased perceived social support. Therefore, H5 was not supported.
H6 stated that triadic closure in one’s ego network in an online health community
positively predicts one’s perceived social support in that community. The result showed that the
path from transitivity of ego network to perceived social support was positive and significant, β
= 0.22, p < .05. The results meant that the number of connections between a person’s alters in the
person’ ego network would increase the person’s perceived social support. Therefore, H6 was
supported.
RQ2 asked if triadic closure in one’s ego network at an earlier time point would predict
one’s number of self-tracking at a later time point. The result showed that triadic closure T2 had
a very small yet significant effect on individuals’ self-tracking behavior at T3, b = 2.70, β = 0.09,
SE = 1. 13, p < .05. It indicated that the higher degree of interconnectedness of one’s health
buddies on FatSecret, the more frequently that the individual would self-track his or her weight.
H7 and H8 stated that perceived social support for weight management would have a
direct effect on individual self-tracking behavior and weight loss progress at a later time point.
The results showed that the path from social support to self-tracking was positive and significant,
β = 0.17, SE = 0.03, p < .05, while the path from social support to weight loss progress was not
significant, β = 0.05, SE = 0.04, p > .05. The statistics meant that individuals who perceived
more social support for weight loss from the FatSecret community tended to engage more
actively in self-tracking behaviors. However, there was no evidence that perceived social support

104

would enhance one’s weight loss progress. Therefore, H7 was supported while H8 was not
supported.
H9 stated that one’s self-tracking behavior at an earlier time point would have a direct
effect on weight loss progress at a later time point. The result indicated that self-tracking
behavior indeed had a positive and significant effect on one’s percentage of goal achievement, β
= 0.10, SE = 0.01, p < .01. Therefore, H9 was supported.
Moreover, the coefficient of path from degree at T2 to degree at T3 was 1.00, p < .05; the
coefficient of path from density at T2 to degree at T3 was 0.95, p < .05; the coefficient of path
from self-tracking at T2 to self-tracking at T3 was 0.76, p < .05; the coefficient of path from
health progress at T2 to health progress at T3 was 0.88, p < .05. The great values in parameter
estimation demonstrated the important to control autoregressive components in the model.
Overall, the hypothesized model explained 99.60% variance in degree, 90% variance in density,
58% variance in self-tracking behavior and 83% variance in health progress.
In general, the results of the hypothesized model showed that perceived social support
mediated the positive relationship between the transitivity of an individual’s ego network and
self-tracking behavior. That said, triadic closure in one’s ego network exerted a positive impact
on self-tracking behavior through a higher level of perceived social support from other
community members.
Post-hoc Analysis: Model Revision
As mentioned earlier, although the hypothesized model had a decent fit, the Chi-Square
statistics showed a value that was large enough for further examination of the theoretical model.
I conducted model revision in the following steps. First, I deleted insignificant paths from the
hypothesized model. This step gave me a model (Model 2) with 17 degrees of freedom and a

105

Chi-Square value at 36.71. More degrees of freedom and a smaller Chi-Square value indicated
that the deletion of nonsignificant paths yielded a better model than the hypothesized model. In
addition, model 2 outperformed the hypothesized model in a couple of fit indices. It received a
better AGFI at 0.93 and a lower RMSEA at 0.07.
The second step of model revision involved an addition of a path from health progress at
T2 to self-tracking behavior at T3. Although the addition of this path was suggested by
modification indices, it made a lot of sense. If a person made positive health progress at an
earlier point of time, it is possible that the person would reduce the tracking behavior because he
or she felt less urgency or necessity to do so anymore. This step gave me a model (Model 3) that
fit the data better than Model 2, but not good enough though, χ
2
(18) = 27.66, p < .01, CFI =
0.99, GFI = 0.98, AGFI = 0.94, and RMSEA = 0.06. Model 3 had one more in degree of freedom
and 9.45 less in Chi-Square value. It indicated that Model 3 was a significantly better model than
Model 2. The parameter estimate showed that health progress at T2 had a significant negative
effect on self-tracking behavior at T3, β = 0.12, p < .05. It indicated that the higher percentage of
goal achieved at T2, the less frequently a person would engage in self-tracking behavior on
FatSecret.
The last step of model revision involved an addition of a path from self-tracking behavior
at T2 to degree at T3. This path was actually related to H2 proposed in Study I, such that
individuals who frequently updated personal health information would have more health buddies
in the community. While in the SIENA analysis the result was marginally significant, in the
SEM analysis, modification indices suggested the addition of this path. The rationale behind the
addition of this path could be the one presented in Study I. Specifically, individuals who
presented more personal health information were more attractive to other community members

106

as they present more health information resources that others may need and learn from.
Therefore, individuals who frequently updated personal health information might receive more
health buddy requests from other members. Given the generalized exchange and norms of
sharing and helping in online health communities, it was very likely that active self-trackers
would accept those health buddy requests and ended up with more health buddies. This step give
me a model that fit the data very well, χ
2
(19) = 14.33, p = 0.57, CFI = 1.00, GFI = 0.99, AGFI =
0.98, and RMSEA = 0.01. The parameter estimate showed that self-tracking at T2 had a very
small yet significant effect on one’s degree in the ego network at T3, β = .02. SE = .01, p < .05.
The final revised model explained 59.3% of the variance in self-tracking behavior and 83% of
the variance in health progress.
Table 8 presents a summary of model comparison with all the fit statistics.
Figure 11 depicts the final revised model with parameter estimations.
Table 9 presents a summary of all the results in both studies.

107

Table 8
A Summary of SEM Model Comparisons
Chi-Square Degree of Freedom Probability level CFI GFI AGFI RMSEA
Model 1
Hypothesized Model 39.53 19 0.00 0.99 0.97 0.90 0.09
Model 2
Delete insignificant paths 36.71 17 0.00 0.99 0.97 0.93 0.07
Model 3
Add a link from T2 Health
Progress to T3 Self-Tracking 27.66 18 0.07 0.99 0.98 0.94 0.06
Model 4 (final model)
Add a link from T2 Self-
tracking to T3 Degree 14.33 19 0.57 0.99 0.99 0.97 0.03

108

Figure 11 Parameter Estimation for Final Modified Model

0.02
Triadic
Closure
0.17
0.95
0.76
0.88
1.00
.11
-0.12
0.09
0.22
Perceived
Social Support
R1
R2
R3
R4
R5
Size
Triadic
Closure
Self-
Tracking
Health
Progress
Size
Self-
Tracking
Health
Progress

109

Table 9
A Summary of Results of Hypotheses and Research Questions
Sociometric Network Formation and Influence Results
H1a
Individuals are more likely to become online health buddies with others
of the same gender. Supported
H1b
Individuals are more likely to become online health buddies with others
of similar age. Supported
H2
Individuals are more likely to become online health buddies with others
of the same group. Supported
H3a
Individuals are more likely to become online health buddies with others
having similar initial health status. Supported
H3b
Individuals are more likely to become online health buddies with others
having a similar health goal. Not Supported
H3c
Individuals are more likely to become online health buddies with others
with similar health progress. Supported
RQ1
Will individuals who present richer health information have more health
buddies than those who present less health information? Partially Yes
H4
In online health social networks, an individual’s health outcome will
become similar to the average health outcome of his or her health
buddies in the network. Supported

Ego Network Effects and Social Support
H5
The size of one’s ego network in an online health community positively
predicts one’s perceived social support in that community. Not Supported
H6
The triadic closure of one’s ego network in an online health community
positively predicts one’s perceived social support in that community. Supported
RQ2
Will triadic closure in one’s ego network at an earlier time point predict
one’s number of self-tracking at a later time point? Yes
H7
Perceived social support positively predicts the number of self-tracking
at a later time point. Supported
H8
Perceived social support positively predicts health progress at a later
time point. Not Supported
H9
The number of self-tracking at an earlier time point positively predicts
health progress at a later time point. Supported
Post Hoc Analysis
P1
The number of self-tracking at an earlier time point positively predicts
the size of one's ego network at a later time point. Yes
P2
Health progress at an earlier time point negatively predicts the number
of self-tracks at a later time point. Yes

110

CHAPTER 7: DISUCSSION AND CONCLUSION
Discussion
The current dissertation is inspired by Christakis and Fowler’s (2007, 2009) findings on
the power of offline social networks in influencing people’s health, and Centola’s (2010, 2011)
work on the role of network structure in influencing people’s health behavior. Given the rise of
the application of social media in health promotion and education, the current dissertation is a
timely and valuable piece of work to further the understanding of online health social networks
established on intentionally designed health-related social media. This dissertation provides
insights into the mechanisms that underly individual’s decisions to form online social networks
with a shared health interest and goal. It offers an empirical test of social influence flowing
through online health social networks. It sheds light on the role of an individual’s ego online
social network in influencing health behaviors through the mechanism of perceived social
support. These aspects of online health social networks are somewhat neglected in previous
literature, yet they are critical to scholars, health professionals and designers of health-related
social media to develop effective health interventions by better organizing and engineering social
networks.
In addition to its general scope and application value, the current dissertation contributes
to the literature in social media and health by situating its inquiries in entrepreneurial online
health communities. Compared with professional and theory-driven designs of health-related
social media for health promotion, fewer studies have paid attention to entrepreneurial health-
related social media (Hwang, et al., 2011; 2013). Most of those media platforms are products of
the bandwagon effects of social networking technologies and a paradigm change to personalized
health care. With much less guidance of health professionals and sophisticated regulations, it is

111

imperative to study the population who are actively using and participating in those social media
platforms. This dissertation is valuable in that it addresses itself to those entrepreneurial websites
and their users. It also facilitates the comparison with formally designed web-based health
interventions with respect to the effectiveness of various components and technological features
included.
Moreover, the third contribution of the dissertation lies in its longitudinal study design.
Without an experimental design, it is often hard to argue the causality of the relationships
between variables of interest. In studies of online health communities, researchers are often
trapped in the question of whether it is the participation (e.g., in the form of making health
buddies and conducting self-tracking) in an community that helps to enhance one’s health, or it is
someone’s better health status that induces one to participate more in a community. Although a
longitudinal panel study design is not able to solve all the arguments with respect to causality, it
offers major advantages over cross-sectional research design in the analysis of causal
interrelationships among variables.
While the three contributions discussed above are important, the greatest value of the
dissertation resides in its findings.
Study I
The first study in the current dissertation examined the underlying theoretical factors that
drive the formation of online health social networks and the joint dynamics of network selection
and network influence. Built on the MTML framework, the study examined different dimensions
of homophily and a resource-based perspective in guiding the selection of health buddies in an
online weight management community. The study found that homophily in the demographic
variables of age and gender both served as significant predictors of network selection.

112

Individuals tended to choose community members who were of a similar age and of the same
gender to be health buddies. This finding was not consistent with the mechanism found in virtual
gaming communities in choosing team members (Huang, et al., 2012; Zhu et al., 2013). In their
studies of team assembly in online games, homophily in age had a mixed effect while homophily
in gender had no significant effect. There are two possible explanations to the inconsistency. The
first explanation could be that social networking features are more sophisticated in the online
weight management community than in virtual games. In FatSecret, community members are
able to create personal profiles just like in Facebook. Cues about their age and gender are more
visible to others through individual presentations in bios and pictures. In contrast, in gaming
communities, the primary zone of activities is in the 3D fantasy world where individual players
are represented by characters and avatars. Cues about age and gender are not directly visible to
other players. Therefore, the visibility of demographic cues in online weight management
communities help participants to identify similar others.
The other reason that demographic homophily is important is that similarity may play a
more important role in talking about and working together to achieve a health-related goal.
While a person may win a game if the person’s teammates are competent, a patient may achieve
his or her health goal if the patient’s buddies can provide relevant and trustworthy support. The
relevancy and trust in health information primarily derives from similarity (Zhang, 2013). People
usually consider similar others as more credible and trustworthy than experts –someone who
maybe more competent – in the provision of health information and social support (Wang et al.,
2008). Therefore, demographic homophily serves as an important clue in forming social ties in
online health social networks.

113

The study also found that homophily in group affiliation positively predicted tie
formation, such that an individual was more likely to become health buddies with someone who
belonged to the same interest group in the community. This finding is consistent to research in
online communities of distance learning (Haythornthwite, 2005), virtual teams (Yuan & Gay,
2006), and virtual games (Zhu et al., 2013). It is interesting to see that groups serve as a
significant organizing mechanism in virtual worlds across different subject areas. The power of
groups reside in their ability to provide a focus wherein individuals collectively participate
because of shared identity, characteristics and/or membership (Feld, 1981). The collective
participation would naturally create shared activities and increase the odds of social interactions.
Therefore, individuals have a greater chance of choosing someone from the same group to
become health buddies.
Interestingly, the study showed that homophily in initial health status and health progress
predicted the selection of health buddies, but homophily in health goals did not. In the online
weight management community, friend selection mirrors the rule in real life. In the real world,
normal-weight adolescents tend to nominate those as friends who are also normal-weight, while
over-weight adolescents tend to be nominated as friends by those who are also over-weight (Ali,
Amialchuk, & Pentina, 2013). In online weight management communities, there may be less an
issue of marginalization because of being overweight as in the real world. People choose those
who had similar initial weight probably because they may be more realistic health buddies who
can work together to lose weight. However, homophily in health goal was not a significant
predictor of health buddy ties. Since health goal was operationalized as the actual weight that a
person aimed to lose, it is possible that people starting with similar weights would naturally have
similar health goals anyway. An examination of the relationship between initial weight and the

114

weight that a person aims to lose showed a significant positive correlation, r = 0.68, p < .05. This
helps to solve the puzzle of the nonsignificant result. It tells us that in the context of weight
management, people starting with similar weight usually have similar goals in terms of how
much weight they want to lose. The effect of homophily in initial weight thus shrinks the
importance of homophily in goal weight. However, the concepts of homophily in initial health
status and health goals need more examinations in other health conditions and behaviors when
the two are not necessarily strongly correlated with one another.
The result with respect to health progress suggested that individuals tended to choose
those as health buddies who were at a similar stage of losing weight, in terms of the percentage
of goal achievement. As indicated in one of the control variables in the model, health progress as
an individual attribute was not a significant predictor of having more health buddies. It indicated
that people who were at a more advanced stage of losing weight did not have more health
buddies than those who were at a less advanced stage. While a person may learn more from those
who were at a more advanced stage of losing weight, it seems that it is not necessarily that
people would choose successful weight losers as health buddies. This dataset represents an
undirected social network, which limits the access to information about tie initiations in the form
of health buddy requests. If that information were available, it would be possible disentangle the
whether it is that successful people refuse to accept health buddy requests or that people do not
initiate requests to become health buddies with successful people. However, even if successful
people might decline health buddy requests, there should have been greater odds for them to
have more health buddies if they were receiving more requests. The finding that homophily in
health progress was a significant factor may suggest that people might choose health buddies for
the purpose of companionship and comradeship. During the journey of weight loss, they were

115

looking for health buddies who could perform activities (e.g., dieting, exercising) together. In
other words, although successful people were competent models to learn from, when it came to
choosing health buddies, people preferred someone who could experience and do things together
with them in the process of weight loss.
The study found that an individual’s frequency of presenting personal health information
(i.e., weight entries) was a marginally significant predictor of tie formation. As discussed in the
above paragraph, the examination of the richness of personal health information in influencing
tie formation was plagued by the undirected network data. But the marginal significance of the
variable may result from the fact that people were indeed looking for health buddies with richer
information resources. People who constantly self-track their own health data might have
received more health buddy requests and thus tended to have more buddies in general. If this is
the case, it is encouraging to learn that people indeed actually pay attention to personal health
data, which is a great advantage and valuable resource in this kind of online health social
network. However, it is also possible that people who were active in self-tracking behavior were
simply more engaged in general participation in the community. That means they tended to
update more health information, post more messages on the discussion boards and make more
health buddies. To scrutinize the alternative explanation, I am collecting additional data on those
users’ other community activities, including posting messages in forums. The data will serve to
control for general participation in the community.
Finally, the study provided empirical support to social contagion in online health social
networks, such that one’s weight loss outcome was positively related to the average outcome of
one’s health buddies. By controlling for the tendency that individuals might select those who
were at a similar stage of weight loss, the analysis showed that an individual tended to lose

116

weight if the average of his or her health buddies represented a weight loss. Although the study
did not specifically analyze the degree of social influence, it showed that online health social
networks could influence individual weight loss outcomes. This is consistent with what Ma et al.
(2011) found in a commercial weight loss social networking site, and what Leahey et al. (2012)
discovered in their team-based weight loss intervention.
Overall, the findings have implications for designing online peer to peer network-based
weight loss programs. First, social media designers must be able to create a health buddy
recommendation system that facilitates the formation of health buddy networks. Both
demographic homophily and health-related homophily play important roles in connecting health
buddies. It is possible to match and team up social media users who share similar demographic
traits and initial weight, and those who are at a similar stage of achieving their goal weights. In
this way, social media users may feel naturally connected to the recommended health buddies.
More importantly, as groups serve as another important network organizing factor, network
interventions are possible by creating desirable groups and sending invitations to users to join
those groups. For example, if we want to connect newbies with active users who diligently
perform exercises, we could create a group for sharing tips on exercise persistence and invite
those active users and newbies to join. It would probably boost the chance of active users to
become health buddies with newbies and newbies may benefit from participation in the
community.
There is not a clear conclusion about the extent to which personal health information
determines the selection of health buddies. But social media designers need to pay attention to
potential barriers that may prevent users from benefiting from shared personal health
information. Both health professionals and laypeople have pointed out that disadvantages of self-

117

tracking are information overload and creating a burden with respect to reviewing such
information (Beaudin, 2006; Li, 2011). This disadvantage may also prevent users from seeking
health buddies with rich personal health information. Too much personal information is
overwhelming and becomes irrelevant. Therefore, it would better to have some built-in
computing algorithm to summarize personal health data on each user’s profile. For example, on
the current profiles in FatSecret, there are various pieces of information about weight, different
types of durations of exercises, diet regimen, protein and calorie consumption and so on. It is not
easy to gain any insight from those data in a quick way. It may make a big difference to provide
a few takeaway points by summarizing the data. With some concise summary, individual users
may perceive personal health information as more valuable and relevant. Then, personal health
information may appear to be a significant factor that people choose their health buddies.
The second implication in designing online peer-to-peer network-based weight loss
programs is that interventions can try to insert successful users into the network consisting of
unsuccessful users. The social influence finding revealed that if the average of one’s health
buddies were gaining weight or keeping steady as their initial weight, the ego person was not
going to lose weight either. The design of social media needs to better tackle this issue;
otherwise, online social networks are simply duplicates of offline social networks, even if those
online social networks aim to promote health. Centola (2011) and DiMaggio (2011) found an
interaction between homophily and status heterogeneity in influencing the spread of healthy and
pro-social behaviors. Speaking of the spread of healthy behaviors, Centola (2011) pointed out
that individuals with higher health status were more influential in social networks than those with
lower health status. The most influential individual was the one who had a high health status and
shared some trait similarities with other network members. Therefore, to solve the dark side of

118

social influence in online health social networks, interventions could try to get connected users
who are not performing in weight loss and users who are performing well and share similarity in
age, gender, initial weight and groups. Participants in intentionally designed health-related social
media may benefit more if we could better organize and engineer their social networks.
Study II
The second study in the current dissertation proposed and tested a model of the effects of
ego network structure and perceived social support on individuals’ self-tracking behavior and
weight loss progress. It draws on literature on structural and functional social support to examine
perceived social support as a mechanism through which a person’s ego network structure may
influence the person’s self-tracking behavior. By considering self-tracking activity as self-
regulation behavior, the study borrowed the resource depletion perspective to examine the role of
perceived social support in enhancing the practice of self-tracking. Based on the panel data of
individuals’ ego network structure, self-tracking behavior and health progress at two different
waves, the study yielded a few interesting findings.
The study found that both size and triadic closure in a person’s network positively
predicted the person’s perceived social support. It contributes to the current literature on the
distinction of structural and perceived social support by showing that some characteristics of
structural social support could be antecedents of perceived social support. While health
interventions often affect structural support, such as forming peer-to-peer groups, functional
social support such as perceived social support shows a stronger correlation with less stress and
better health (Verheijden et al., 2005). The findings are valuable in providing insights into the
best way of organizing social networks in enhancing perceived social support. It was not
surprising to find that the number of health buddies was positively associated with perceived

119

social support. The more buddies that a person had, the more likely that the person would be to
find sources of support. This result is consistent with studies conducted in the offline world
finding that the size of one’s social network predicted one’s psychologically perceived social
support (e.g., Zhu, Woo et al., 2013). In addition to the size of the ego network, the proportion of
connections between alters of an ego was another important predictor of perceived social
support. This is consistent with literature on strength of ties and social capital (Granovetter,
1973; Burt, 2005). With a higher degree of interconnectedness between one’s alters, the health
buddy relationships in such an online community may become more stable and “sticky”
(Conwell, 2009. P 93). The tightly-knit networks may create a relatively closed small group
structure, wherein strong ties may appear and last longer. Such characteristics of social
relationships may facilitate the provision of social support and the perception of psychological
closeness.
Moreover, the study found that triadic closure in ego networks had a direct effect on self-
tracking behavior. As discussed above, more closed and dense networks may offer increased
monitoring, stronger shared identity and greater pressure toward conformity (Conwell, 2009). It
is possible that an individual may feel a stronger norm to be disciplined and self-track personal
health conditions. It is also possible that health buddies in a tightly-knit network agree on a
mutual monitoring. In either explanation, a more closely-knit ego network of health buddies may
be more desired in encouraging and maintaining self-tracking behavior. It is interesting to note
that additional post hoc analysis showed that the size of one’s ego network did not have a
significant effect on self-tracking behavior. In other words, having more health buddies in the
community did not help with self-management behavior.

120

Perceived social support also had a direct positive effect on self-tracking behavior, but no
significant effect on one’s weight loss progress. The result of the positive relationship between
perceived social support and self-tracking is consistent with research on perceived social support
and self-management for patients with chronic diseases (see review in Gallant, 2003). Gallant
has pointed out that in offline social networks, disease- or regimen-specific social support could
better predict self-management behaviors than general social support. This study followed the
insight by operationalizing perceived social support as specific support for weight loss in an
online community. The positive effect may be due to the mechanism that was discussed in the
literature review, such that greater perceived social support would reduce a person’s levels of
stress, and then the person would have more resources and energy to perform self-management
behavior. It is also possible that perceived social support may increase a person’s self-efficacy in
performing self-management (Maeda, Shen, Schwarz, Farrell, & Mallon, 2013), so that the
person is able to self-track more frequently and constantly.
Even though the study did not find a significant effect of perceived social support on
health progress, it did discover that self-tracking behavior positive predicted health progress. The
finding tells us that it is not enough to feel psychologically supported to lose weight, but one
needs to take actual actions to lose weight. It is possible that self-tracking behavior mediates the
positive relationship between perceived social support and health progress. Perceived social
support helps to enhance one’s self-efficacy and alleviate one’s stress so that the person is more
prepared to take the subsequent actions for self-management. By actively engaging in self-
tracking activities, individuals are able to better monitor their diet, exercise and other important
habits that may influence weight loss.

121

Finally, the post-hoc analysis revealed one interesting and important finding: that a
person’s health progress at T2 had a negative effect on the person’s self-tracking behavior at T3.
It is very likely that once a person found that he or she was progressing towards the goal weight,
the person would become relaxed or not motivated to do self-tracking any more. This could be
the moment when intervention messages are sent out to remind and encourage users to continue
self-tracking or at least coming back after certain while. Research has shown that weight
management often fails because of how easy it is to regain weight after achieving one’s goal
weight (Burke, Conroy, Sereika et al., 2012). Moreover, weight management is related to other
chronic diseases such as diabetes, which requires long-term and constant self-management. The
major advantage of self-tracking behavior is that it increases individuals’ awareness of their
energy intake and expenditure, and other physical activities. Therefore, it is important to sustain
the behavior of self-monitoring for successful weight management. The finding may warn health
professionals and intervention programmers to pay attention to the moment when individuals
have achieved a certain percent of their goals. That moment may be when individuals
discontinue self-tracking behaviors. It might be useful to send reminders or persuasive messages
at that moment to emphasize the importance of self-monitoring. The purpose of self-monitoring
is not solely to change health-related behaviors, but to maintain the awareness of individuals’
behavior and the circumstances that surrounds the behavior (Wadden, Crerand & Brock, 2005).
In other words, self-monitoring serves an important function of prevention.
Limitations and Future Research
The current dissertation has a number of limitations. The first limitation is that only one
type of social network – a health buddy network – is included in the studies. In an online weight
loss community, social interactions and communications are not restricted within health buddy

122

networks. Community users may interact with one another by commenting on each other’s
journals or exchange social support messages in discussion forums. As information and
emotional social support is one primary characteristic of online health social networks, message
exchange is an important communication activity in the form of posting and replying in threads.
The examination of the network of support message exchange may provide different insights into
social interactions in online health social networks (Bambina, 2007). Facebook researchers have
revealed that friendship ties on Facebook do not manifest the actual social interaction. It could be
true in online health social networks as well. After becoming health buddies, relational partners
may not communicate with each other at all. The network of support message exchange may
more accurately represent communication networks in online health communities. However, I
would also argue that it does not mean that the network of support message exchange is more
important than the health buddy network. Facebook researchers have pointed out that the value
of social network sites is that people can maintain social relationships with a low effort by
getting news feeds and catching up through updates from their friends (Ellison et al., 2011). In
the context of online health social networks, instead of actually asking and answering questions,
many informational needs can be more efficiently and easily met by receiving news updates. To
become health buddies is the way to have direct access to each other’s weight loss activities and
progress.
In future research, it would be interesting to compare the underlying mechanisms that
drive the formation of message exchange networks and health buddy networks. From the
perspective of individual motivation (Chung, 2013), seeking individuals with richer and more
credible information may be an important factor in forming ties of message exchanges. It is also
possible that the message exchange network may predict the health buddy network. As theorized

123

in Lin (1999), message exchange ties could be the outer layer of one’s social relationships in a
community, as those ties can be just one-shot interactions and transient. In comparison, health
buddy ties could be the inner layer of one’s social relations in the community, as those ties are
established on explicit mutual confirmation and tend to last longer. According to Lin (1999),
inner layers of social relations are embedded within outer layer of social relations. Additional
analysis of the relationship between message exchange networks and health buddy networks will
shed light on the extent to which message exchange in discussion forums serves as a pathway of
socialization in an online health community. Moreover, future research can examine network
multiplexity in online health communities and if relationship multiplexity would have any
implications for the study of social influence on health behaviors and outcomes.
The second limitation of the dissertation is related to the network sample in Study I.
Although I constructed a complete network for the final subsample, the data cleaning process
excluded a proportion of network members who were health buddies of the actors included in the
final subsample. While it would be better to include every actor in the whole social network, I
would argue that the exclusion of those inactive actors should not have biased the finding with
the respect to the formation of online health buddy networks. By employing actor-based models,
it was the connections that are added, deleted and maintained at different time points that were
modeled to make inferences about rules of network selection. However, the subsample may bias
the result on social influence on health outcomes through health buddy networks. The reason is
that the most significant or close health buddies might have been excluded during the data
cleaning process. If that is the case, I would argue that the current analysis of social influence
might be a more stringent test of peer influence given that health buddies included were not the
most influential ones. On the other hand, it is possible that the active health buddies during data

124

collection were the important sources of influence. Active health buddies were the ones who
were engaged in self-tracking behaviors and practicing weight loss regimens. News feeds would
update information about their weight loss activities and therefore, their activities were the most
visible to each other. Future research can make use of other communication activities in the
community to extract information about the strength of health buddy relations. By DOIng so, it
will be possible to assign a value to each health buddy tie and weigh the social influence based
on tie strength.
The third limitation of the dissertation is related to the sample in Study II. The survey
respondents in the second study were not randomly selected, and thus it was a self-selected
sample. The results may subject to self-selection bias. For example, survey respondents could be
a group of active users in FatSecret given the time window for survey completion. It is also
possible that the sample consisted of those who were more inclined to interact with other
FatSecret members. The reasons is that the link to the survey was sent through FatSecret internal
message system, through which users’ private messages were usually sent. A user would be very
likely to discard the message if he or she does not interact with other FatSecret users at all.
The next limitation is related to the generalizability. All the analyses and results in the
current dissertation were exclusively based on one online weight management community,
FatSecret. Therefore, caution is in order when generalizing the results to other online weight
management communities and online health social networks intentionally designed for other
diseases, such as diabetes and cancers. It is also true that some sites present rather different
features from FatSecret. For example, some social network sites for weight loss have a relatively
closed system in selection of health buddies. Instead of having the freedom to send buddy
requests to anybody in FatSecret, users are encouraged to bring their offline contacts (e.g.,

125

families and friends) on those sites to form teams and networks for weight loss. In some cases,
users’ profiles are not public at all and only visible to the members in their personal networks.
Under these circumstances, the findings will be not generalizable. However, I would argue that
the findings in this dissertation may apply to those online weight management social networks
characterized by social networking and self-tracking features. The features and formats of
FatSecret are very representative of online social networking sites for weight management. A
comparison also shows that descriptive statistics of survey participants in the second study were
very similar to that of survey participants in other published studies using different test beds,
such as SparkPeople (Hwang, et la., 2011; 2012). Respondents were comparable in their age,
gender, BMI and general Internet use. Thus, it is reasonable to expect that the FatSecret
population may not be necessarily different from community members of other online weight
management social networking sites.
Future research should replicate the current studies with other similar online weight loss
sites to establish the robustness of the findings. More importantly, more research is needed to
examine the same research problems discussed here on other online weight loss sites with
different features and mechanisms. A comparison of results from websites built on different
mechanisms may yield fruitful insights into the designs of online health communities and
interventions to better organize social networks for health promotion.
Finally, there are several potential variables that may confound the findings. As
mentioned earlier in the discussion of the first study, a user’s levels of participation in FatSecret,
may confound the finding on the positive effect of self-tracking of personal health information
on the number of health buddies. An on-going project is collecting data on users’ participation

126

behaviors in the form of commenting on others’ journals on profiles and posting messages in
discussion forums.
Conclusion
The current dissertation provides an examination of online health social networks in
intentionally designed health-related social media. It investigated the theoretical mechanisms that
drive the formation of online health social networks, examined the joint dynamics of network
selection and network influence on individual health outcomes, and tested a model of ego
network structure, social support and self-tracking behavior. The dissertation is situated in an
entrepreneurial online weight management social networking site, FatSecret. It employed a four-
month longitudinal study design, and collected system data extracted from the site and self-
reported data from an online survey with the users of FatSecret.
Drawing on the MTML framework, the first study found that demographic homophily
including age and gender similarity, inbreeding homophily in the form of group affiliation, and
health-related homophily including initial health status and health progress significantly
predicted the selection of online health buddies. Similarity in health goal was not a significant
predictor due to its strong correlation with initial health status in the specific health issue of
weight loss. A person’s frequency of updating personal health information was marginally
significant in predicting tie formation. By conducting a SIENA analysis, the study also found
empirical support to a social influence effect among health buddies, such that an individual’s
weight outcome tended to become similar to the average of the individual’s health buddies’
weight outcomes. The second study drew from literature on structural and functional social
support, the buffering model of social support and self-regulation depletion studies. By
conducting SEM analyses, the study found that both the size and triadic closure of an

127

individual’s ego network predicted perceived social support for weight loss in FatSecret. Then,
perceived social support predicted more active self-tracking behavior, and self-tracking behavior
predicted improved health progress. Post hoc analysis showed that there was a negative and
significant effect of health progress on self-tracking behavior, such that improved health progress
at an earlier time point would reduce the amount of self-tracking at a later time point.

128

References
Adamic, L.A., & Huberman, B.A. (2000). Power-Law Distribution of the World Wide Web.
Science, 287(5461), 2115.
Adelman, M. B, Malcolm, R.P. & Albrecht, T.L. (1987). Beyond close relationships: support in
weak ties. In T. L. Albrecht & M. B. Adelman (Eds.), Communicating Social support (pp.
126–147). Newbury Park, CA: Sage.
Ahn, Y-Y-., Han, S., Kwak, H., Moon, S., & Jeong, H. (2007). Analysis of topological
characteristics of huge online social networking services. In Proceedings of ACM
WWW, May, Banff, Alberta, Canada.
Ali, M.M., Amialchuk, A., & Pentina, I. (2012). The effects of drinking and sports on popularity
of overweight adolescents: implications for public policy. Journal of Public Health,
21(3), 221-230.
Arem, H., & Irwin, M. (2011). A review of web-based weight loss interventions in adults.
Obesity Review, 12(5), e236-e243. DOI: 10.1111/j.1467-789X.2010.00787.x
Arrington, M. (2007). Fatsecret: For fat people who want to be less so. Retrieved from
http://techcrunch.com/2007/04/29/fatsecret-for-fat-people-who-want-to-be-less-so/
Atouba, Y. & Shumate, M. (2010). Interorganizational networking patterns among development
organizations. Journal of Communication, 60, 293-317.
Albrecht, T. L., & Adelman, M. B. (1987a). Communicating social support: A theoretical
perspective. In T. L. Albrecht & M. B. Adelman (Eds.), Communicating Social Support
(pp. 126–147). Newbury Park, CA: Sage.
Albrecht, T. L., & Adelman, M. B. (1987b). Communication networks as structure of social
support. In T. L. Albrecht & M. B. Adelman (Eds.), Communicating Social support (pp.
126–147). Newbury Park, CA: Sage.
Bambina, A. (2007). Online social support: The interplay of social networks and compute-
mediated communication. New York: Cambria Press.
Bandura, A. (1977). Social Learning Theory. Englewood Cliffs, NJ: Prentice-Hall.
Barak, A., Boniel-Nissim, M., & Suler, J. (2008). Fostering empowerment in online support
groups. Computers in Human Behavior, 24(5), 1867-1883.
Barrera, M. (1986). Distinctions between social support concepts, measures, and models.
American Journal of Community Psychology, 14(4), 413-445.

129

Bartlett, Y. K., & Coulson, N. S. (2011). An investigation into the empowerment effects of using
online support groups and how this affects health professional/patient communication.
Patient Education and Counseling, 83, 113–119.
Barabási, A-L., & Albert, R. (1999). Emergence of scaling in random networks. Science,
286(5439): 509-512. DOI: 10.1126/science.286.5439.509
Baumeister, C. DeWall, N., Ciarocco, N. J., Twenge, J. M. (2005). Social exclusion impairs self-
regulation. Journal of Personality and Social Psychology, 88(4), 589-604.
Beaudin, J., Stephen, I., Margaret, M. (2006). To track or not to track: User reactions to concepts
in longitudinal health monitoring. Journal of Medical Internet Research, 8(4), 5
Berkman, L.F. & Glass, T. (2000). Social integration, social networks, social support and health.
In L.F. Berkam and I. Kawachi (Eds). Social Epidemiology, pp. 137-173. New York:
Oxford University Press.
Berkman, L. F., & Breslow, L. (1983). Health and ways of living: the Alameda county study.
New York: Oxford University Press.
Berkman, L.A., Kawachi, I. (2000). A historical framework for social epidemiology. In: L
Berkman and I Kawachi (Eds). Social Epidemiology, pp.3-12. New York: Oxford
University Press.
Bourdieu, P. (1986). The forms of capital. In J.G. Richardson (Eds). The Handbook of Theory:
Research for the Sociology of Education, pp. 241-258. New York: Greenwood Press.
Blau, P. (1964). Exchange and power in social life. New York, NY: Wiley.
Bollen, K. A. (1989). Structural equations with latent variables. New York: John Wiley.
Boulos, M.N.K., & Wheelert, S. (2007). The emerging Web 2.0 social software: An enabling
suite of sociable technologies in health and health care education. Health Information and
libraries Journal, 24, 2-23.
Börner, K., Contractor, N., Falk-Lrzesinski, H.J., Fiore, S.M., Hall, K.L., Keyton, J., Spring, B.,
Stokols, S., Trochim, W., & Uzzi, B. (2010). A multi-level systems perspective for the
science of team science. Science Translational Medicine, 2, 49cm24.
DOI:10.1126/scitranslmed.3001399
boyd, d.m., & Ellison, N.B. (2007). Social network sites: Definition, history and scholarship.
Journal of Computer-Mediated Communication, 13, article 11.

130

Boyd, D.M., & Ellison, N.B. (2012). Social network sites. In H. Donelan, K. Kear & M. Ramage
(Eds.), Online Communication and Collaboration: A Reader. pp. 261-283. NY. New
York: Routledge.
Brindal, E., Freyne, J., Saudners, I., Berkovsky, S., Smith, G., Noakes, M. (2012). Features
predicting weight loss in overweight or obese participants in a web-based intervention:
randomized trial. Journal of Medical Internet Research. 14(6), e173.
Burk, L.E., Conroy, M.B., Sereika, S. M., Acharya, S. D., Sevick, M.A., Ewing, L. J., & Glanz,
K. (2012). The effect of electronic self-monitoring on weight loss and dietary intake: A
randomized behavioral weight loss trial. Obesity, 19(2), 338-344.
Burt, R.S. (1980). Models of network structure. Annual Review of Sociology, 6(1), 79-141.
Burt, R.S. (2005). Brokerage and closure: An introduction to social capital. Oxford; New York:
Oxford University Press.
Burt, R.S. (2011). Structural holes in virtual worlds. University of Chicago Press.
Byrne, B. M. (1998). Structural equation modeling with LISREL, PRELIS, and SIMPLIS.
Mahwah, NJ: Lawrence Erlbaum Associates.
Centola, D. (2013). Social media and the science of health behavior. Circulation, 127: 2135-
2144. DOI: 10.1161/CIRCULATIONAHA.112.101816
Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329,
1194-1197.
Centola, D. (2011). An experimental study of homophily in the adoption of health behavior.
Science, 334(2), 1269-1272.
Christakis, N. A., & Fowler, J. H. (2007). The spread of obesity in a large social network over 32
years. The New England Journal of Medicine, 357(4), 370-379.
Christakis, N. A., & Fowler, J. H. (2008). The collective dynamics of smoking in a large social
network. The New England Journal of Medicine, 358, 2249-2258.
Christakis, N. A., & Fowler, J. H. (2009). Connected: The surprising power of our social
networks and how they shape our lives. New York, NY: HachetteBookGroup.
Chung, J. E. (2013). Social networking in online support groups for health: How online social
networking benefits patients. Journal of Health Communication: International
Perspective, DOI: 10.1080/10810730.2012.757396

131

Chuang, K.Y., & Yang, C.C. (2012). Interaction patters of nurturant support exchanged in online
health social networking. Journal of Medical Internet Research, 14(3): e54.
DOI:10.2196/jmir.1824.
Cobb, N.K., Graham, A.L., Abrams, D.B. (2010). Social network structure of a large online
community for smoking cessation. American Journal of Public Health, 100(7), 1282-
1289. DOI: 10.2105/AJPH.2009.165449
Cornwell, B. (2009). Good health and the bridging of structural holes. Social Networks, 31(1),
92-103.
Coursaris, C., & Liu, M. (2009). An analysis of social support exchanges in online HIV/AIDS
self-help groups. Computers in Human Behavior, 25(4), 911-918.
Cialdini, R. B. (2001). Influence: Science and practice (4th ed.). New York: HarperCollins.
Cialdini, R. B., Reno, R. R., & Kallgren, C. A. (1990). A focus theory of normative conduct:
Recycling the concept of norms to reduce littering in public places. Journal of
Personality & Social Psychology, 58, 1015-1026.
Cobb, S. (1976). Social support as a moderator of life stress. Psychosom Medicine, 38(5): 300-
314.
Cohen, S., & Wills, T. A. (1985). Stress, socila suport, and the buffering hypothesis. Psychology
Bulliten, 98, 310-357.
Coleman, J.S. (1988). Social capital in the creation of human capital. American Journal of
Sociology, 94, S95-S120.
Contractor (2013). Some assembly required: leveraging Web science to understand and enable
team assembly. Philosophical Transactions of the Royal Society A, 371(1987), DOI:
10.1098/rsta.2012.0385
Contractor N, & Su C. (2011). Understanding groups from a network perspective. In Research
Methods for Studying Groups and Teams: A Guide to Approaches, Tools, and
Technologies (eds A Hollingshead, MS Poole), pp. 284–310. New York, NY: Routledge.
Contractor, N. S., Wasserman, S. & Faust, K. (2006). Testing multitheoretical, multilevel
hypotheses about organizational networks: An analytic framework and empirical
example. Academy of Management Review, 31(3), 681-703.
Courtois, C., Mechant, P., De Marez, L., & Verleye, G. (2009). Gratifications and seeking
behavior of online adolescents. Journal of Computer-Mediated Communication, 15, 109-
137.

132

de la Haye, K., Green, H.D., Kennendy, D.P., Pollard, M.S., & Tucker, J.S. (2013). Selection and
influence mechanisms associated with marijuana initiation and use in adolescent
friendship networks. Journal of Research on Adolescence, 23(3), 474-486.
de la Haye, K., Robins, G., Mohr, P., & Wilson, C. (2010). Obesity-related behaviors in
adolescent friendship networks. Social Networks, 32, 161-167.
de la Haye, K., Robins, G., Mohr, P., & Wilson, C. (2011a). How physical activity shapes, and is
shaped by, adolescent friendships. Social Science and Medicine, 73, 719-718.
de la Haye, K., Robins, G., Mohr, P., & Wilson, C. (2011b). Homophily and contagion as
explanations for weight similarities among adolescent friends. Journal of Adolescent
Health, 49, 421-427.
De la Torre-Díez, I., Díaz-Pernas, F.J., & Antón-Rodríguez, M. (2012). A content analysis of
chronic diseases social groups on Facebook and Twitter. Telemedicine and e-Health,
18(6), 404-408. DOI: http://dx.DOI.org/10.1089/tmj.2011.0227
DiMatteo, M. R. (2006). Social support and patient adherence to medical treatment: A meta-
analysis. Health Psychology, 23(2), 207-218.
Easley, D., & Kleinberg, J. (2010). Networks, crowds, and markets: Reasoning about a highly
connected world. Cambridge University Press.
Ekeh, P. P. (1974). Social exchange theory: The two traditions. Cambridge, Mass: Harvard
University Press.
Ellison, N.B., & Hancock. J. T. (2013). Profile as promise: Honest and deceptive signals in
online dating. Security & Privacy, IEEE, 11(5), 84-88.
Emerson, R. M. (1976). Social exchange theory. Annual Review of Sociology, 2, 335-362.
Eysenbach, G. (2008). Medicine 2.0: Social networking, collaboration, participation,
apomediation and openness. Journal of Medical Internet Research, 10(3), e22.
DOI:10.2196/jmir.1030
Faraj, S., & Johnson, S. L. Network exchange patterns in online communities. Organization
Science, 2011,1-17.
Feld, S.L. (1981). The focused organization of social ties. The American Journal of Sociology,
86(5), 1015-1035.
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7: 117-140.
Flynn, F. J. (2005). Identity orientations and forms of social exchange in organizations. Academy
of Management Review, 30(4), 737-750.

133

Fishbein, M. & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to
theory and research. Reading, MA: Addison-Wesley
Fogel, J., & Nehmad, E. (2009). Internet social network communities: Risk taking, trust, and
privacy concerns. Computers in Human Behavior, 25, 153-160.
Fox, S., & Duggan, M. (2013). Health online 2013. Pew Research Internet Project, retrieved
from http://www.pewinternet.org/2013/01/15/health-online-2013/
Frees, E. (2004). Longitudinal and Panel Data: Analysis and Applications in the Social Sciences.
New York: Cambridge University Press
Frohlich, D.O., & Zmyslinski-Seelig, A. (2012). The presence of social support messages on
YouTube videos about inflammatory bowel disease and ostomies. Health
Communication, 27(5), 421-428.
Frost, J.H., Massagli, M.P. (2008). Social uses of personal health information within
PatientsLikeMe, an online patient community: What can happen when patients have
access to one another’s data. Journal of Medical Internet Research, 10(3): e15.
Funk KL, Stevens VJ, Appel LJ, Bauck A, Brantley PJ, Champagne CM, Coughlin J, Dalcin AT,
Harvey-Berino J, Hollis JF, Jerome GJ, Kennedy BM, Lien LF, Myers VH, Samuel-
Hodge C, Svetkey LP, Vollmer WM (2010). Associations of internet website use with
weight change in a long-term weight loss maintenance program. Journal of Medical
Internet Research,12(3):e29. DOI: 10.2196/jmir.1504
Gallant, M. P. (2003). The influence of social support on chronic illness self-management: A
review and directions for research. Health Education and Behavior, 30, 170- 197.
Golder, S.A. & Yardi, S. (2010). Structural predictors of tie formation in Twitter: Transitivity
and mutuality. In Proceeding of IEEE Social Computing Second International
Conference, Minneapolis, MN.
Gotsis, M., Wang, J., Spruijt-Metz, D., Jordan-Marsh, M., & Valente, T. (2013). Wellness
partners: Design and evaluation of a Web-based physicial activity diary with social
gaming features for adults. Journal of Medical Internet Research Protocol, 2(1), e10,
DOI:10.2196/resprot.2132
Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360-
1380.
Greene, J., Sacks, R., Piniewski, B., Kil, D., Hahn, J.S. (2013). The impact of an online social
network with wireless monitoring devices on physical activity and weight loss. Journal of
Primary Care & Community Health, 4(3), 189-194. DOI: 10.1177/2150131912469546

134

Hanneman, R. A., & Riddle, M. (2005). Introduction to social network methods. Riverside, CA:
University of California, Riverside (published in digital form at
http://faculty.ucr.edu/~hanneman/ )
Harary, F., Norman, R. Z., & Cartwright, D. (1965). Structural models: An introduction to the
theory of directed graphs. New York: Wiley.
Hawn, C. (2009). Take two aspirin and tweet me in the morning: How Twitter, Facebook, and
other social media are reshaping health care. Health Affairs, 28(2), 361-368. DOI:
10.1377/hlthaff.28.2.361
Haythornthwaite, C. (2002). Strong, weak and latent ties and the impact of new media. The
Information Society, 18(5), 385-401.
Heaney, C.A. & Israel, B.A. (2008). Social networks and social support. In K.Glanz, B.K.
Rimer, and K. Viswanath. (Eds). Health Behavior and Health Education, pp.189-210.
San Francisco, CA: John Wiley & Sons.
Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley.
Hoffman, B. R., Sussman, S., Unger, J.B., Valente, T.W. (2006). Peer influence on adolescent
cigarette smoking: A theoretical review of the literature. Substance Use & Misuse, 41,
103-155.
Holland, P.W., & Leinhardt, S. (1976). Local structure in social networks. In D. R. Heise (Ed.),
Sociological Methodology (pp. 1–45). San Francisco: Jossey-Bass.
Hood, L., & Auffray, C. (2013). Participatory medicine: A driving force for revolutionizing
healthcare. Genome Medicine, 5: 110.
Hossain, M.A. (2014). Perspectives of human factors in designing elderly monitoring system.
Computers in Human Behavior, 33, 63-68.
Hu, X., Bell, B.A., Kravitz, R.L., & Orrange, S. (2012). The prepared patient: Information
seeking of online support group members before their medical appointments. Journal of
Health Information: International Perspectives, 17: 960-978.
Huang, Y., Shen, C., & Contractor, N. S. (2013). Distance matters: Exploring proximity and
homophily in virtual world network. Decision Support Systems, 55, 969-977.
Huang, Y., Ye, W., Bennett, N., & Contractor, N. S. (2013). Functional or social? Exploring
teams in online games. In Proceeding CSCW. San Antonio, TX, USA.

135

Hwang, K. O., Etchegaray, J. M., Sciamanna, C. N., Bernstam. E. V., Thomas, E. J. (2012).
Structural social support predicts functional social support in an online weight loss
program. Health Expectations, DOI: 10.1111/j.1369-7625.2011.00759.x
Hwang, K. O., Ning, J., Trickey, A. W., Sciamanna, C. N. (2013). Website usage and weight loss
in a free commercial online weight loss program: Retrospective cohort study. Journal of
Medical Internet Research, 15(1), e11. DOI: 10.2196/jmir.2195.
Hwang, K. O., Ottenbacher, A. J., Lucke, J. F., Etchegaray , J. M., Graham, A. L., & Thomas, E.
J. (2011). Measuring social support for weight loss in an Internet weight loss community.
Journal of Health Communication, 16: 198-211.
Idriss, S.Z., Kvedar, J.C., & Watson, A.J. (2009). The role of online support communities:
Benefits of expanded social networks to patients with psoriasis. Archives of Dermatology,
145(1), 46-51.
Ikeda, A., Kawachi, I., Iso, H., Iwasaki, M., Inoue, M., & Tsugane, S. (2013). Social support and
cancer incidence and mortality: the JPHC study cohort II. Cancer Causes & Control,
24(5): 847-860. DOI: 10.1007/s10552-013-0147-7
Jones, M., Luce, K.H., Osborne, M.I., Taylor, K., Cunning, D., Doyle, A.C., Wilfley, D.E.,
Taylor, B. (2008). Randomized, controlled trial of an Internet-facilitate intervention for
reducing binge eating and overweight in adolescents. Pediatrics, 121(3): 453-462.
Joreskog, K.G., & Sorbom, D. (1996). LISREL 8: User's Reference Guide. Chicago, IL:
Scientific Software International.
Joseph, R.P., Durant, N.H., Benitez, T.J., & Pekmezi, D.W. (2013). Internet-based physical
activity interventions. American Journal of Lifestyle Medicine, DOI:
10.1177/1559827613498059
Katz, J. E., & Rice, R. E. (2002). Social consequences of Internet use: Access, Involvement and
Interaction. Cambridge, MA: MIT Press.
Kawachi, I. (2010). Social capital and health. In C.E. Bird, P. Conrad, A.M. Fremont, S.
Timmermans (Eds). Handbook of Medical Sociology, Chapter 2, pp. 18-32, Nashville,
Tennessee: Vanderbilt University Press.
Kawachi, I., & Berkman, L.A. (2003). Neighborhoods and health. New York: Oxford University
Press.
Kaplan, A.M., & Haemlein, M. (2010). Users of the world, unite! The challenges and
opportunities of social media. Business Horizons, 53: 59-68.

136

Kiesler, C.A. (1985). Policy implications of research on social support and health. In S. Cohen
and S.L. Syme (Eds.), Social Support and Health, pp. 347-364. San Diego, CA:
Academic Press.
Kincaid, J. (2009). FatSecret looks to become a central hub for nutrition data with new API.
Retrieved from http://techcrunch.com/2009/08/14/fatsecret-looks-to-become-a-central-
hub-for-nutrition-data-with-new-api/
Kraut, R.E., Resnick, P., Kiesler, S., Burke, M., Chen, Y., Kittur, N., Konstan, J., Ren, Y., &
Riedl, J. (2012). Building successful online communities: Evidence-based social design.
Boston, MA: MIT Press
Kroenke, C.H., Kwan, M.L., Neugut, A.I., Ergas, I.J., Wright, J.D., Caan, B.J., Hershman, D.,
Kushi, L.H. (2013). Social networks, social support mechanisms, and quality of life after
breast cancer diagnosis. Breast Cancer Research and Treatment, 139(2), 515-527. DOI:
10.1007/s10549-013-2477-2
Korda, H., & Itani, Z. (2011). Harnessing social media for health promotion and behavior
change. Health Promotion Practice, 14, 15-22.
Kumar, R., Novak, J., & Tomkins, A. (2006). Structure and evolution of online social networks.
In Proceedings of KDD, August, Philadelphia, Pennsylvania.
Love, B., Himelboim, I., Holton, A., & Stewart, K. (2013). Twitter as a source of vaccination
information: Content drivers and what they are saying. American Journal of Infection
Control, 41, 568-570.
Lasasfeld, P.F., & Merton, R.K. (1954). Friendship as a social process: A substantive and
methodological analysis. IN M. Berger, T., Abel & C.H.Page (Eds.), Freedom and
control in modern society. New York: Van Nostrand.
Leahey, T. M., LaRose, J. G., Fava, J. L., Wing, R. R. (2011). Social influences are associated
with BMI and weight loss intentions in young adults. Obesity, 19, 1157-1162. DOI:
10.1038/oby.2010.301
Leahey, T.M., Kumar, R., Weinberg, B.M., Wing, R. R. (2012). Teammates and social influence
affect weight loss outcomes in a team-based weight loss competition. Obesity, 20(7),
1413-1418. DOI: 10.1038/oby.2012.18
Lejbkowicz, I., Caspi, O., & Miller, A. (2012). Participatory medicine andpatient empowerment
towards personalized healthcare in multiple sclerosis. Expert Review of
Neurotherapeutics, 12(3), 343-352.

137

Lewis, K., Gonzalez, M., & Kaufman, J. (2012). Social selection and peer influence in an online
social network. In Proceedings of the National Academy of Sciences and the United
States of America. DOI: 10.1073/pnas.1109739109
Lewis, K., Kaufman, J., Christakis, N. (2008). The taste for privacy: An analysis of college
student privacy settings in an online social network. Journal of Computer-Mediated
Communication, 14(1), 79-100. DOI: 10.1111/j.1083-6101.2008.01432.x
Lorig, K.R., & Holman, H. R. (2003). Self-management education: History, definition,
outcomes, and mechanisms. Annals of Behavioral Medicine, 26 (1), 1-7.
Li, J. (2013). Privacy policies for health social networking sites. Journal of American Medical
Informatics Association, 20, 704-707. DOI:10.1136/amiajnl-2012-001500
Lin, N. (1999). Social networks and status attainment. Annual Review of Sociology, 25, 467-487.
Lin, N. (2001). Social capital: Theory and Research. New York: Aldine de Gruyter.
Lin, N., Ye, X., & Ensel, W. M. (1999). Social support and depressed mood: A structural
analysis. Journal of Health and Social Behavior, 40(4), 344-359.
Lungeanu, A., Huang, Y., & Contractor, N. S. (2014). Understanding the assembly of
interdisciplinary teams and its impact on performance. Journal of Informetrics, 8, 59-70.
Ma, X., Chen, G., Xiao, J. (2010). Analysis of an online health social network. In Proceedings of
IHI, Arlington, Virginia, USA.
Macy, M., Willer, R. (2002). From factors to actors: Computational sociology and agent-based
modeling. Annual Review of Sociology, 28: 143-166.
McCorkle, R., Ercolano, E., Lazenby, M., Schulman-Green, D., Schiling, L. S., Lorig, K., &
Wagner, E.H. (2011). Self-management: Enabling and empowering patients living with
cancer as a chronic illness. CA: A Cancer Journal of Clinicians, 61(1), 50-62.
McCroskey, J.C. (1974). The effect of interaction behavior on source credibility, homophily, and
interpersonal attraction. Human Communication Research, 1(1), 42.
Maeda, U., Shen, B-J., Schwarz, E.R., Farrell, K. A. & Mallon, S. (2011). Self-efficacy mediates
the association of social support and depression with treatment adherence in heart failure
patients. International Journal of Behavioral Medicine, 20(1), 88-96.
McEwan, G., Gutwin, C., Mandryk, R. L., Nacke, L. (2012). “I'm just here to play games”:
Social dynamics and sociality in an online game site, In Proc. CSCW. Seattle, WA, USA.

138

McLaughlin, M., Nam, Y., Gould, J., Pade, C., Meeske, K .A., Ruccione, K. S. & Fulk, J.
(2012). A videosharing social networking intervention for young adult cancer survivors.
Computers in Human Behavior, 28, 631-641.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social
networks. Annual Review of Sociology, 27, 415-444.
Meyerson, D.K., Weick, K.E., Kramer, R.M. (1996). Swift trust and temporary groups. In R.M.
Kramer & T.R. Tyler (Eds.), Trust in Organizations: Frontiers of Theory and Research,
Pp. 166-195, CA, Thousand Oaks: Sage Publications.
Mo, P. K. H., Coulson, N. S. (2010). Empowering processes in online support groups among
people living with HIV/AIDS: A comparative analysis of ‘lurkers’ and ‘posters’.
Computers in Human Behaviors, 26, 1183-1193.
Molm, L. D., Collett, J. L., & Schaefer, D. R. (2007). Building solidarity through generalized
exchange: A theory of reciprocity. American Journal of Sociology, 113(1), 205-242.
Monge, P., & Contractor, N. (2003). Theories of Communication Networks. Oxford University
Press, Oxford, UK.
Mulaik, S.A., James, L.R., Van Alstine, J., Bennett, N., Lind, S., & Stilweel, C.D. (1989).
Evaluation of goodness-of-fit indices for structural equation models. Quantitative
Methods in Psychology, 105(3), 430-445.
Muraven, M., Tice, D. M., & Baumeister, R.F. (1998). Self-control as limited resource:
Regulatory depletion. Journal of Personality and Social Psychology, 74(3), 774-789.
Namkoong, K., DuBenske, L., Shaw, B. R. (2012). Creating a bond between caregives online:
Effect on caregivers’ coping strategies. Journal of Health Communication: International
Perspectives, 17(2), 125-14.
Neiger, B.L., Thackeray, R., Van Wagenen, S.A., Hanson, C.L., West, J.H., Barnes, M.D., &
Fagen, M.C. (2012). Use of social media in health promotion: Purposes, key performance
indicators, and evaluation metrics. Health Promotion Practice, 13: 159-164.
Newman, M.E.J. (2007). Clustering and preferential attachment in growing networks, Physical
Review E, 64(2), 025102.
Oh, S., Yi, Y.J., & Worrall, A. (2012). Quality of health answers in social Q&A. In A. Grove
(Ed.), Proceedings of the 75
th
ASIS&T Annual Meeting: Information, interaction,
innovation: Celebrating the past, constructing the present and creating the future.
Naltimore, MD. Silver Spring, MD: American Society for Information Science and
Technology.

139

Oh, H. J., & Lee, B. (2011). The effect of computer-mediated social support in online
communities on patient empowerment and doctor-patient communication. Health
Communication, 1-12.
Olander, E. K., Fletcher, H., Williams, S., Atkinson, L., Turner, A., & French, D. P. (2013).
What are the most effective techniques in changing obese individuals’ physical activity
self-efficacy and behavior: A systematic review and meta-analysis. International Journal
of Behavioral Nutrition and Physical Activity, 10: 29-44.
Orizio, G., Schulz, P., Gasparotti, C., Caimi, L., & Gelatti. (2010). The World of e-Patients: A
content analysis of online social networks focusing on diseases. Telemedicine and e-
Health, 16(10): 1060-1066. DOI:10.1089/tmj.2010.0085.
Paton, C., Hansen, M., Fernandez-Luque, L., & Lau, Y. S. (2012). Self-tracking, social media
and personal health records for patient empowered self-care. IMIA Yearbook of Medical
Informatics, pp. 16-24. http://www.schattauer.de/de/magazine/uebersicht/zeitschriften-a-
z/imia-yearbook/imiayearbook-2012/issue/special/manuscript/17937/show.html
Pearson, M., Steglich, C., & Snijders, T. (2006). Homophily and assimilation among sport-active
adolescent substance users. Connections, 27(1), 47-63.
Putnam, R.D. (2001). Bowling alone: The collapse and revival of American community (1
st

Touchstone ed.). New York: Simon & Schuster.
Pfeil, U., Zaphiris, P., & Wilson, S. (2010). The role of message-sequences in the sustainability
of an online support community for older people. Journal of Computer-Mediated
Communication, 15, 336-363.
Phua, J.J. (2011). The social groups approach to quitting smoking: An examination of smoking
cessation online and offline through the influence of social norms, social identification,
social capital and social support (Doctoral dissertation). Available from ProQuest
Dissertation and Theses database. (UMI No. 3466083)
Rainie, L., & Wellman, B. (2012). Networked: The new social operating system. Boston, MA:
MIT Press.
Rains, S. A., & Young, V. (2009). A meta-analysis of research on formal computer-mediated
support groups: Examining group characteristics and health outcomes. Human
Communication Research, 35, 309-335.
Rau, P.L.P., Gao, Q., & Ding, Y. (2008). Relationship between the level of intimacy and lurking
in online social network services. Computers in Human behavior, 24, 2757-2770.
Rieder, S., & Ruderman, A. (2007). The development and validation of the weight management
support inventory, Eating Behaviors, 8, 39-47.

140

Ripley, R.M., Snijders, T.A.B., & Preciado, P. (2013). Manual for RSiena. Retrieved from
http://www.stats.ox.ac.uk/~snijders/siena/RSiena_Manual.pdf
Robins, G., Pattison, P., 2005. Interdependencies and social processes: dependence graphs and
generalized dependence structures. In: Carrington, P.J., Scott, J., Wasserman, S. (Eds.),
Models and Methods in Social Network Analysis. Cambridge University Press, New
York.
Robins, G., Pattison, P., & Woolcock, J. (2005). Small and other worlds: Global network
structure from local processes. American Journal of Sociology, 110, 894-936.
Sandler, I.N. (1980). Social support resources, stress, and maladjustment of poor children.
American Journal of Community Psychology, 8(1), 41-52.
Schwarzer, R., & Leppin, A., (1991). Social support and health: A theoretical and empirical
overview. Journal of Social and Personal Relationships, 8: 99.
Sciandra, A., Gioachin, F., & Finos, L. (2013). Package ‘egonet’. Version 1.2, Retrieved from
http://cran.r-project.org/web/packages/egonet/egonet.pdf
Sherbourne CD, Stewart AL (1991) The MOS social support survey. Soc Sci Med 32(6):705–714
Shaw, R. J., & Johnson, C. M., Health information seeking and social media use on the Internet
among people with diabetes. Online Journal of Public Health Informatics, 3(1).
http://firstmonday.org/htbin/cgiwrap/bin/ojs/index.php/ojphi/article/viewFile/3561/3011
Shen, C., & Monge, P. (2011). Who connects with whom? A social network analysis of an online
open source software community. First Monday, 16(6).
Shen, C., Monge, P., & Williams, D. (2012). Virtual brokerage and closure: Network structure
and social capital in a massively multiplayer online game. Communication Research,
DOI: 10.1177/0093650212455197
Shim, M., Cappella, J. N., & Han, J. Y. (2011). How does insightful and emotional disclosure
bring potential health benefits? Study based on online support groups for women with
breast cancer. Journal of Communication, 61, 432-454.
Shumate, M., Fulk, J., & Monge, P. (2005). Predictors of the international HIV/AIDS INGO
network over time. Human Communication Research, 31, 482–510.
Sillence, E. (2013). Giving and receiving peer advice in an online breast cancer support group.
Cyberpsychology, Behavior, and Social Networking, 16(6), 480-485.

141

Simpkins, S.D., Schaefer, D.R., Price, C.D., & Vest, A.E. (2012). Adolescent friendships, BMI,
and physical activity: Untangling selection and influence through longitudinal social
network analysis. Journal of Research on Adolescence, 23(3), 537-549.
Snijders, T.A.B., van de Bunt, G. G., & Steglich, C. E.G. (2010). Introduction to stochastic
actor-based models for network dynamics. Social Networks, 32, 44-60.
Sohn, S., Helms, T.M., Pelleter, J.T., Annett, A.M., Kröttinger, I., Schöffski, O. (2012). Costs
and benefits of personalized healthcare for patients with chronic heart failure in the care
and education program “Telemedicine for the Heart”. Telemedicine and e-Health, 18(3),
198-204. DOI:10.1089/tmj.2011.0134.
Steglich, C., Snijders, T.A.B., & Pearson, M. (2010). Dynamics networks and behavior:
Separating selection from influence. Sociological Methodology, 41(1), 329-393.
Swan, M. (2009). Emerging patient-driven health care models: An examination of health social
networks, consumer personalized medicine and quantified self-tracking. International
Journal of Environmental Research and Public Health, 6, 492-525,
DOI:10.3390/ijerph6020492

Takahashi, N. (2000). Emergence of generalized exchange. American Journal of Sociology,
105(4), 1105-1134.

Tang, E., & Lee, W. (2006). Singapore internet users’ health informatoin search: Motivation,
perception of information sources, and self-efficacy. In M. Murero & R. E. Rice (Eds.),
The Internet and Health Care: Theory, Research, and Practice (pp. 107–126). Mahwah,
NJ: Lawrence Erlbaum Associates.
Tanis, M. (2008). Health-related on-line forums: What’s the big attraction? Journal of Health
Communication, 13: 698-714.
Thackeray, R., Crookston, B.T., West, J.H. (2013). Correlates of health-related social media use
among adults. Journal of Medical Internet Research, 15(1), e21. DOI: 10.2196/jmir.2297
Thoits, P. A. (1995). Stress, coping, and social support: Where are we? What next. Journal of
Health and Social Behavior, 35, 53–79.
Toma, C.L., & Hancock, J. T. (2012). What lies beneath: The linguistic traces of deception in
online dating profiles. Journal of Communication, 62(1), 78-97.
Uchino, B.N., Holt-Lunstad, J., Smith, T.W., & Bloor, L. (2004). Heterogeneity in social
networks: A comparison of different models linking relationships to psychological
outcomes. Journal of Social and Clinical Psychology, 23(2), 123-139.

142

Vaccaro, J. A. Exebio, J. C. Zarini, G. G. & Huffman, F. G. (2014). The role of family/friend
social support in disbetes self-management for minorities with type 2 diabetes. Journal of
Nutrition and Health, 2(1), 1-19.
Valente, T. W., Fujimoto, K., Chou, C-P., & Spruijt-Metz, D. (2009). Adolescent affiliations and
adiposity: A social network analysis of friendships and obesity. Journal of Adolescent
Health, 45(2), 202-204.
Van Der Gaag, M., Snijders, Tom A. B. (2005). The Resource Generator: social capital
quantification with concrete items. Social Networks, 27 (1), 1-29.
Veenstra, R., Dijkstra, J.K., Steglich, C., & Van Zalk, M. (2013). Network-behavior dynamics.
Journal of Research on Adolescence, 23(3), 399-412.
Verheijden, M.W., Bakx, J. C., Weel, C. V., Koelan, M.A., Stavern, W. A.V. (2005). Role of
social support in lifestyle-focused weight management interventions. European Journal
of Clinical Nutrition, 59, s179-186.
Vohs, K. D., & Heatherton, T. F. (2000). Self-regulatory failure: A resource-depletion approach.
Psychological Science, 11: 249.
Wadden, T.A., Crerand, C. E., Brock, J. (2005). Behavioral treatment of obesity. Psychiatric
Clinics of North America, 28(1), 151-170, ix.
Wang, Z., Walther, J.B., Pingree, S., & Hawkins, R.P. (2008). Health information, credibility,
homophily, and influence via the Internet: Web sites versus discussion groups. Health
Communication, 23: 358-368.
Walther, J.B., & boyd, S. (2002). Attraction to computer-mediated social support. In C. A. Lin &
D. Atkin (Eds.), Communication Technology and Society: Audience Adoption and Uses
(pp.153-188). Cresskill, NJ: Hampton Press.
Walther, J.B., Pingree, S., Hawkins, R.P., & Buller, D.B. (2005). Attributes of interactive online
health information systems. Journal of Medical Internet Research, 7(3), e33.
Walther, J.B., Van Der Heide, B., Hamel, L.M., Chulman, H.C. (2009). Self-generated versus
other-generated statements and impressions in Computer-mediated Communication: A
test of warranting theory using Facebook. Communication Research, 36(2), 229-253.
Walther, J.B., Van Der Heide, B., Kim, S-Y., Westerman, D., Tong, S.T. (2008). The role of
friends’ appearance and behavior on evaluations of individuals on Facebook: Are we
known by the company we keep? Human Communication Research, 34(1), 28-49.

143

Wasserman, S., & Faust, K. (1999). Social network analysis. Cambridge University Press,
Cambridge, UL.
Welbourne, J. L., Blanchard, A. L., & Wadsworth, M. B. (2013). Motivations in virtual health
communities and their relationship to community, connectedness and stress. Computers
in Human Behavior, 29(1), 129-139. DOI: http://dx.DOI.org/10.1016/j.chb.2012.07.024
Wellman, B. (1997). An electronic group is virtually a social network. In S. Kiesler (Ed.),
Culture of the Internet (pp. 179–205). Mahwah, NJ: Lawrence Erlbaum.
Wellman, B., Boase, J., & Chen, W. (2002). The networked nature of community: Online and
offline. IT&Society, 1(1), 151-165.
Wellman, B., Quan-Haase, A., Boase, J., Chen, W., Hampton, K., Diaz, I., & Miyata, K. (2006).
The social affordances of the Internet for networked individualism. Journal of Computer-
Mediated Communication, 8(3), DOI: 10.1111/j.1083-6101.2003.tb00216.x
Wellman, B., & Wortley, S. (1990). Different strokes from different folks: Community ties and
social support. American Journal of Sociology, 96(3), 558-588.
Wicks, P., Massagli, M., Frost, J., Brownstein, C., Okun, S., Vaughan, T., Bradley, R.,
Heywood, J. (2010). Sharing health data for better outcomes on PatientsLikeMe. Journal
of Medical Internet Research, 12(2): e19, DOI: 10.2196/jmir.1549
Wills, T. A., & Ainette, M. G. (2012). Social networks and social support. In A. Baum, I.
Revenson, & J. Singer (Eds.), Handbook of Health Psychology (2
nd
) pp. 465-492. New
York, NY: Psychology Press.
Wills, T. A., & Fegan, M. F. (2001). Social networks and social support. In A.Baum, T.
A.Revenson, & J. E.Singer (Eds.), Handbook of Health Psychology (pp. 209-234).
Mahwah, NJ: Lawrence Erlbaum Associates.
Williams, D., Consalvo, M., Caplan, S., 7 Yee, N. (2009). Looking for gender (LFG): Gender
roles and behaviors among online gamers. Journal of Communication, 59, 700-725.
Williams, D., Ducheneaut, N., Xiong, L., Zhang, Y., Yee, N., & Nickell, E. (2006). From tree
hourse to barracks: The social life of guilds in World of Warcraft. Games and Culture,
1(4), 338-361.
Wright, K. B. (2000). Perceptions of on-line support providers: An examination of perceived
homophily, source credibility, communication and social support within on-line support
groups. Communication Quarterly, 48, 44–59.

144

Wright, K. B. (2009). Increasing computer-mediated social support. In J.C. Parker & E. Thorson
(Eds.), Health Communication in New Media Landscape, pp.243-266. NY, New York:
Springer Publish.
Wright, K. B., & Bell, S. B. (2003). Health-related support groups on the Internet: Linking
empirical findings to social support and computer-mediated communication theory.
Journal of Health Psychology, 8(1), 39–54.
Wright, K.B., Rosenberg, J., Egbert, N., Ploeger, N.A., Bernard, D. R., & King, S. (2013).
Communication competence, social support, and depression among college students: A
model of Facebook and Face-to-Face support network influence. Journal of Health
Communication: International Perspective, 18(1), 41-57.
Yakusheva, O., Kapinos, K.A., & Eisenberg, D. (2013). Estimating heterogeneous and
hierarchical peer effects on body weight using roommate assignments as a natural
experiment. The Journal of Human Resources, 49 (1), 234-261.
Yan, L., Tan, Y., & Peng, J. Network dynamics: How can we find patients like us? Retrieved
from SSRN: http://ssrn.com/abstract=1820748 or
http://dx.DOI.org/10.2139/ssrn.1820748
Ye, Y. (2013). The role of illness factors and patient satisfaction in using online health support
groups. Health Communication, DOI: 10.1080/10410236.2012.757526
Yuan, Y.C., & Gay, G. (2006). Homophily of network ties and bonding and bridging social
capital in computer-mediated distributed teams. Journal of Computer-Mediated
Communication, 11(4), 1062-1084.
Zhang, Y. (2014). Beyond quality and accessibility: Source selection in consumer health
information searching. Journal of the Association for Information Science and
Technology, DOI: 10.1002/asi.23023
Zhu, M., Huang, Y., & Contractor, N. S. (2013). Motivations for self-assembling into project
teams. Social Networks, 35, 251-264.
Zhu, X., Woo. S.E., Porter, C., & Brzezinski, M. (2013). Pathways to happiness: From
personality to social networks and perceived support. Social Networks, 35, 382-393.
Zook, K.R., Saksvig, B., Wu, T.T., & Young, D.R. (2014). Physical activity trajectories and
multilevel factors among adolescent girls. Journal of Adolescent Health, 54(1), 74-80

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Mobile self-tracking for health: validating predictors, effects, mediator, moderator, and social influence

PDF

Benefits of social networking in online social support groups

PDF

The social groups approach to quitting smoking: An examination of smoking cessation online and offline through the influence of social norms, social identification, social capital and social support

PDF

Social media and health: social support and social capital on pregnancy-related social networking sites

PDF

Relationship formation and information sharing to promote risky health behavior on social media

PDF

The effect of Twitter social support on health outcomes and its mediators: a randomized controlled trial of a social support intervention on Twitter for patients affected by cancer

PDF

Contagious: social norms about health in work group networks

PDF

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

PDF

A multitheoretical multilevel explication of crowd-enabled organizations: exploration/exploitation, social capital, signaling, and homophily as determinants of associative mechanisms in donation-...

PDF

Building personal wellness communities: meaningful play in the everyday life of a network society

PDF

The role of social control in promoting healthy eating behavior among Chinese immigrants: an ecological approach

PDF

Social network influences on depressive symptoms among Chinese adolescents

PDF

The patterns, effects and evolution of player social networks in online gaming communities

PDF

Effects of economic incentives on creative project-based networks: communication, collaboration and change in the American film industry, 1998-2010

PDF

Leveraging social normative influence to design online platforms for healthy eating

PDF

Social selection and influence of alcohol & marijuana implicit cognitions and behaviors: a longitudinal investigation of peer social network dynamics

PDF

The evolution of multidimensional and multilevel networks in online crowdsourcing

PDF

Network influences in health initiatives: multimedia games for youth in Peru

PDF

The power of social media narratives in raising mental health awareness for anti-stigma campaigns

PDF

The influence of social networking sites on high school students' social and academic development

Asset Metadata

Creator Meng, Jingbo (author)

Core Title The formation and influence of online health social networks on social support, self-tracking behavior and weight loss outcomes

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Publication Date 06/13/2014

Defense Date 03/25/2014

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

Tag homophily,OAI-PMH Harvest,online health social networks,personal health information,self-tracking,social influence,social support

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor McLaughlin, Margaret L. (committee chair), Fulk, Janet (committee member), Jordan-Marsh, Maryalice (committee member), Monge, Peter R. (committee member)

Creator Email jingbome@msu.edu,jingbomeng@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-418289

Unique identifier UC11296195

Identifier etd-MengJingbo-2547.pdf (filename),usctheses-c3-418289 (legacy record id)

Legacy Identifier etd-MengJingbo-2547.pdf

Dmrecord 418289

Document Type Dissertation

Format application/pdf (imt)

Rights Meng, Jingbo

Type texts

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

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

Repository Name University of Southern California Digital Library

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