University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
RELATIONSHIP FORMATION AND INFORMATION SHARING
TO PROMOTE RISKY HEALTH BEHAVIOR ON SOCIAL MEDIA

by

Mina Park

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

December 2017

Copyright 2017 Mina Park

i
DEDICATION
To my family and Peggy McLaughlin.

ii
ACKNOWLEDGMENTS
Foremost, I would like to sincerely and gratefully thank my advisor, Dr. Margaret
McLaughlin for her continuous support, caring, and endless encouragement. She has
guided me along the path of scientific research and inspired me to pursue research that
contributes to the public good. She has been an excellent academic advisor and a great
mentor over the course of my PhD program. Writing this dissertation and completing my
PhD would not have been possible without her.
I also express gratitude to my other committee members. I thank Dr. Michael
Cody for his support, academically and socially. He provided me valuable feedback and
encouraged me throughout my PhD study. Dr. Tom Valente inspired my research interest
in social networks. He was committed to guiding my dissertation and offered many
insights along the way. In addition to my committee members, I want to thank Dr. Lynn
Miller for being a wonderful supporter and a great collaborator. I have learned greatly
from all the conversations we’ve had and the projects we’ve collaborated on. I would also
like to thank Dr. Ken Sereno, Dr. Lian Jian, and Dr. Christina Dunbar-Hester for their
continuous support. It was fortunate to have opportunities to work with them on teaching
foundational courses. Thanks to their guidance and examples, I learned the importance of
becoming a good teacher-scholar.
I would also like to thank my friends and colleagues at Annenberg for providing
wonderful companionship and valuable feedback on my research during my PhD life. I
am grateful to Komathi Ale, Janeane Anderson, Jeeyun Baik, Amanda Beacom, Yomna
Elsayed, Renyi Hong, Jove Hou, Selene Hu, Young Ji Kim, Do Own Kim, Hye Min Kim,
Hyun Tae Kim, Nahoi Koo, Yujung Nam, Poong Oh, Yao Sun, Jieun Shin, Minhee Son,
iii
Nathan Walter, Rong Wang, Wei Wang, Yusi Xu, Lin Zhang, and Nan Zhao. They have
each filled my graduate school life with love and unforgettable memories.
Last but certainly not least, I want to acknowledge my family. I am grateful for
their endless encouragement and incredible support. They provided the primary
motivation for my pursuing a PhD before and throughout my graduate studies. And to my
husband, Joungwook, whom I am blessed to have in my life, thank you always for your
love, patience, and support.

iv
TABLE OF CONTENTS
DEDICATION i
ACKNOWLEDGMENTS ii
LIST OF TABLES vi
LIST OF FIGURES vii
ABSTRACT viii
CHAPTER 1: INTRODUCTION

Overview 1
Chapter Summaries 9

CHAPTER 2: UNDERSTANDING RISKY HEALTH BEHAVIOR
ON SOCIAL MEDIA 11

Understanding Risky Health Behavior 11
Defining Risky Health Behavior 11
Defining Individuals Who Promote Risky Health Behavior 14
Risky Health Behaviors on Social Media 17
Social Influence of Risky Health Behavior on Social Media 21

CHAPTER 3: STUDY I. INDIVIDUALS WHO PROMOTE
RISKY HEALTH BEHAVIOR AND THEIR RELATIONSHIP FORMATION
ON SOCIAL MEDIA 27

Motivations of Relationships Formation for Individuals Who Promote
Risky Health Behavior 27
Stigma 27
Social Identity 28
Social Capital in Social Media and Formation of Social Networks
Promoting Risky Health Behavior 32

CHAPTER 4: STUDY II. INFORMATION SHARING TO PROMOTE RIKSY
HEALTH BEHAVIOR ON SOCIAL MEDIA 36

Dissemination of Information with Social Support 37
The Role of Social Support in Information Sharing 37
Emotional Support for Individuals Who Promote Risky Health
Behavior 38
Dissemination of Information with Affective Tone 39
The Role of Affective Tone in Information Sharing 39
v
Positive and Negative Affective Tones for Individuals Who
Promote Risky Health Behavior 41
Focus of the Current Study 43

CHAPTER 5: METHODS AND RESULTS 46

Study I 47
Data Collection 47
Measurement 51
Smoking-related following networks
Source characteristics and user categories
Network measures
Analysis 54
Results 54

Study II 63
Data Collection 63
Data Cleaning 63
Measurement 65
Information dissemination
Social support
Affective tone
Analysis 66
Results 66

CHAPTER 6: DISCUSSION AND CONCLUSION 77
Study I 77
Study II 83
Limitations and Future Research 87
Conclusion 91

REFERENCES 93
vi
LIST OF TABLES

Table 1: Frequency and Percentage of Sources Characteristics in Social
Networks from the Pro-smoking and Anti-smoking Seed
Accounts 55

Table 2: Frequencies and Percentages of Sources Characteristics in Pro-
smoking and Anti-smoking Networks, Users with Explicit
Positions on Smoking 57

Table 3: Network-Level Measures of Pro-smoking, Anti-smoking, and
Combined Networks 59

Table 4: Bridging Ties between Pro-smoking and Anti-smoking Networks 61

Table 5: Source Characteristics of Bridging Users 62

Table 6: Content Analysis of Tweets in Pro-smoking and Anti-smoking
Groups 67

Table 7: Frequency of Social Support Relevance by Level of
Dissemination 69

Table 8: Observed and Expected Frequencies of Tweets by Social Support
Type and Group 70

Table 9: Observed and Expected Frequencies of Tweets by Affective
Tone and Dissemination Level 72

Table 10: Observed and Expected Frequencies of Tweets by Affective
Tone and Group 74

Table 11: Chi-squared Difference Tests on Social Support, Affective Tone,
and Tweets Dissemination 75

Table 12: Chi-squared Difference Tests on Stance and Types of Social
Support and Affective Tone in Disseminated Tweets 75

Table 13: A Summary of the Hypotheses Testing Results 76

vii
LIST OF FIGURES

Figure 1: Images of risky health behavior on social media 18

Figure 2: An example of online communities promoting extreme dieting 20

Figure 3: Visualization of the Network Using Gephi. Red dots represent pro-
smoking users while green dots represent anti-smoking users.
Yellow links represent ‘pro-to-anti’ ties while blue links represent
‘anti-to-pro’ ties. 60

Figure 4: Number of Tweets by social support and dissemination level 68

Figure 5: Number of Tweets by social support type in pro-smoking and
anti-smoking groups 71

Figure 6: Number of Tweets by affective tone dissemination level 73

Figure 7: Number of highly-disseminated Tweets by affective tone and
group 74

viii
ABSTRACT
While social media can serve as beneficial channels for exchanging health
information and social support, they have also been blamed for promoting risky health
behavior. Social media platforms can facilitate online relationship formation and
information sharing among people who engage in risky health behavior such as smoking
or unprotected sex. The purpose of this project is to gain a better understanding of how
individuals who promote risky health behavior form networks on social media and what
kinds of information they are motivated to share.
Focusing on smoking behavior, the first study investigates the theoretical
mechanisms that drive social network formation among pro-smoking users, and examines
an empirical instance of one such network structure on Twitter. Consistent with the social
identity framework, the first study finds that risk-promoting networks manifest higher
stance homophily (pro-smoking vs. anti-smoking), higher network cohesion, and lower
community exclusivity than health-promoting networks. Most pro-smoking users who
had social ties with anti-smoking users were found to be individuals, whereas most anti-
smoking users who had social ties with pro-smoking users were non-profit/academic
groups or individuals. Bridging users on both sides tended to be linked to pressure groups.
The second study drew from literature on social support, emotion, and
information sharing. It found that information conveying social support and an affective
tone was disseminated at higher rates. Emotional support and a negative affective tone
were especially prevalent features of information sharing in the pro-smoking network
compared to the anti-smoking network. These findings have valuable implications for
ix
designing effective health campaigns. These implications are discussed along with
limitations and directions for future research.

Keywords: Risky health behavior, Social media, Social identity, Stigma,
Relationship formation, Social support, Information sharing, Smoking.

1
CHAPTER 1: INTRODUCTION
Overview
Communication technologies have become essential tools for seeking and sharing
health information. People may experience uncertainty about appropriate diets,
ambiguous symptoms, costs of medical treatment (Brashers, Neidig, & Goldsmith, 2004;
Rains & Tukachinsky, 2015). Studies have shown that communication technologies play
a critical role in this uncertainty management process by providing convenient and
effective tools for accessing health information (Rains & Tukachinsky, 2015). According
to a recent report, about 72 % of Internet users have looked online for health information
and 40% of Internet users have shared their personal health experiences (Fox & Duggan,
2013). In particular, about 34% of online health information seekers use social media to
access and share health information (Korda & Itani, 2013). Given that nearly 80% of
online users engage on social media (Greenwood, Perrin, & Duggan, 2016), use of social
media in healthcare is expected to increase even more.
Social media have become popular ways not just for sharing health information,
but for exchanging social support (Chung, 2013; Heather, Murphy, & Valente, 2014;
McLaughlin et al., 2012). It has been shown that people feel uncomfortable discussing
their health issues with close others (Albrecht, Goldsmith, & Thompson, 2003; Brashers
et al., 2004; Wright, 2009). One major reason is that they feel stigmatized or judged when
talking about such a sensitive topic. For example, Tong, Heinemann-LaFave, Jeon,
Kolodziej-Smith, & Warshay (2013) found that when individuals engaging in extreme
dieting discussed their health problems with others close to them, they often experienced
negative reactions rather than social support. As a result, they conceal their health issues
2
from close ones in order to avoid being hurt. Another major reason is that close ties lack
experience with specific health problems or persons who have those problems (Brashers
et al., 2004). Family or friends, in general, have limited information about a specific
health problem and thus, may not provide a necessary solution (Wright, 2009). They may
misunderstand the person with a health problem (Brashers et al., 2004). For example,
close ones may provide too much information (e.g. sending articles) when what the
person actually wants is to avoid information and instead receive emotional support. Thus,
for individuals who concern about these, the costs of sharing one’s health issue with close
ones outweigh the benefits.
A social media platform provides a place where people can more actively
participate in a conversation around their health issues and exchange social support
(Mclaughlin, Park, & Sun, 2015). Most social media do not require personal information
such as users’ real names. Social media users can create multiple accounts with various
names. They also can connect with different social groups using different accounts. This
anonymous feature provides a comfortable environment for talking about health problems
without the user feeling stigmatized or judged. In addition, on social media, it is easy for
individuals to find like-minded people with whom to communicate and share information
regardless of constraints of geography and time (Hether, Murphy, & Valente, 2016). With
a few searches and clicks on computers or smartphones, individuals who share similar
health issues can be easily found. They may exchange more useful and necessary social
support with a great understanding of each other’s needs.
While social media are found to enhance quality of life by facilitating the
exchange of health information and social support (Klemm et al., 2003; Vilhauer,
3
McClintock, & Matthews, 2010), they have also been blamed for promoting risky health
behavior (Arseniev-Koehler, Lee, McCormick, & Moreno, 2016; Oksanen et al., 2015;
Park, Sun, & McLaughlin, 2017). Similar to creating social circles for a healthier life,
individuals who engage in stigmatized risky health behaviors, such as smoking or
unprotected sex, can find social media users who share their same unhealthy behaviors.
Social media platforms have contributed to an environment where these individuals can
get together and share social support (Bert, Gualano, Camussi, & Siliquini, 2016;
Chhabra & Bryant, 2016; Wang et al., 2015). On social media such as Twitter, YouTube,
and Instagram, these users can anonymously share tips and information about committing
suicide, getting drugs, and hiding eating disorders through texts, videos, and images.
Users engaged in unhealthy behaviors also leave comments on such content and support
each other by justifying, maintaining, or promoting risky behaviors (Arseniev-Koehler et
al., 2016; Bert et al., 2016; Paek, Kim, Hove, & Huh, 2014; M. Park et al., 2017; van der
Tempel et al., 2016).
Previous research has studied the deleterious effects of content promoting risky
health behaviors on the health-related attitudes and behavior of receivers (Huang et al.,
2014; Yoo, Yang, & Cho, 2016). Specifically, studies have found that sharing or being
exposed to such content increased receivers’ intentions to engage in those behaviors,
positive attitudes towards those behaviors, and the frequency of actually engaging in
those behaviors (Dunlop, More, & Romer, 2011; Huang et al., 2014; Noar, Myrick,
Morales-Pico, & Thomas, 2014; Oksanen et al., 2015; Yoo et al., 2016). While scholars
have paid greater attention to the effects of such social media content on health behavior,
the question of how information senders’ behavior on social media contributes to these
4
effects has received little attention (although see Yom-Tov, Fernandez-Luque, Weber, &
Crain, 2012). Their relationship formation on social media may affect the boundaries of
the information receivers. For example, if individuals who advocate risky health
behaviors form highly cohesive networks and promote bonding functions, such
information will be mostly reached only by the network members. Also, their information
sharing behavior may affect the influential power of such content. For instance, if they
like to disseminate a particular type of information, such information will be widely
exposed to social media users and have great influences. Thus, investigating their
behaviors provides a better understanding of the influence mechanisms of risky health
behavior-related content on social media. To advance our understanding of individuals
who engage in risky health behaviors and their roles in the promotion of such behaviors
on social media, this dissertation examines the social relationships of social media users
who engage in such behaviors and their information sharing behavior within their
respective social circle.
As discussed above, people are less willing to talk about health concerns with
close ties because they are afraid of being stigmatized and receiving undesired support.
Risky health behaviors including smoking, self-harm, extreme diet, etc. are not welcomed
by society. Individuals who engage in those behaviors may be very reluctant to share
these issues with close ones. According to Davies and Lipsey (2003), people with eating
disorders seek tips on how to avoid detection by family members and friends when
skipping meals.
Due to this fact, such individuals have little chance of finding supportive peers in
the offline world. Typically, individuals who share similar health interests exchange
5
information about how to cope, share health related perspectives, and empathize with
each other’s personal stories (Barak, Boniel-Nissim, & Suler, 2008). For example,
marathon runners can easily obtain tips on effective running through other marathon
runners in their social network. Likewise, it is not difficult for a cancer patient to find
other patients through support groups. By contrast, it is difficult to find people or groups
that encourage risky health behaviors. The difficulty of meeting fellow risk-takers offline
may cause risky behavior advocates to remain ignorant of the seriousness of their
behaviors and coping methods. It also prevents them from gaining social support
including emotional support (e.g. sympathy, understanding, encouragement, and physical
affection), informational support (e.g. situation appraisal, advice, and teaching), and
esteem support (e.g. validation, compliment, and relief of blame) (Braithwaite, Waldron,
& Finn, 1999).
With fewer chances of finding those with whom they can discuss their risky
health behaviors, individuals engaging in these behaviors may experience strong feelings
of uncertainty about their identity. For example, a girl with an eating disorder may
believe ‘attractiveness’ is an important factor of being loved, and may believe that having
an extremely thin body makes her more attractive. At the same time, she may also know
people around her think that an eating disorder is a disease and not a lifestyle choice.
Thus, the conflicts between her internal belief and the negative views of others would
increase her uncertainty about herself as she engages in the risky behavior.
The social identity perspective provides a useful framework to predict behavioral
patterns of such risky health behavior advocates on social media. This approach
addresses the self-conception of group members and related phenomena including
6
differentiation within groups, deviance, group culture, group decision making, and
collective action (Hogg, Abrams, Otten, & Hinkle, 2004). According to the social identity
perspective, when individuals feel uncertain about things that reflect on self or self-
identity, they are motivated to reduce that uncertainty because the feeling is
psychologically uncomfortable (Hogg & Blaylock, 2011). People want to know who they
are, how to behave, who others are, and how others might behave and think (Hogg &
Blaylock, 2011). High uncertainty about one’s identity also causes low self-efficacy,
which give a feeling of helplessness (Glanz, Rimer, & Viswanath, 2008). Thus,
individuals who engage in risky health behaviors will strongly desire to reduce
uncertainty and protect their self-efficacy.
The social identity perspective demonstrates that individuals reduce feelings of
uncertainty by identifying themselves with groups of similar peers. Individuals who
engage in risky health behaviors may desire to find similar peers and support groups in
order to reduce their feelings of uncertainty about themselves and their behaviors. As
discussed above, however, it is difficult for them to find supportive others offline. Thus,
social media play an important role for them as it provides a place where they can easily
get together and create social networks to share their experiences without feeling
stigmatized. Within these social circles, users can share more sensitive information and
their honest feelings. Through this process, they may be able to reduce uncertainty about
the worldviews held by others like them and learn how others cope with the effects of
their risky health behaviors. On the other hand, individuals who engage in and advocate
healthy behaviors will have less uncertainty about their behavior and their identity. They
can easily access information about their health concerns both offline and online. They
7
also have lower barriers to talking about their health issues with close ties. For them,
online social networks do not constitute their primary support community but rather
provide supplemental support and information. Thus, their motivations to avoid stigma or
to reduce uncertainty may be much lower than those engaging in risky health behaviors.
The first study of this dissertation examines the social relationships of individuals
who engage in risky health behavior. As their motivations to find others on social media
may differ from those of individuals who advocate healthy behavior, the patterns of
social network formation and maintenance might be different. Social networks are
important to look at as they have been shown to strongly influence individuals’ health-
related behavior (Valente, 2012). According to Hanson et al. (2013), relationships
embedded in one’s social network are a significant contributor to health behavior, even
beyond individual attributes such as age, sex, education level, income, and occupation.
Studies have found that social relationships play a decisive role in influencing substance
abuse and smoking (Valente, Fujimoto, Unger, Soto, & Meeker, 2013; Valente, Hoffman,
Ritt-Olson, Lichtman, & Johnson, 2003). Valente et al. (2003) found that social network-
based tobacco interventions were more effective than random networks at structuring
heath prevention programs. The purpose of Study I is to understand the difference in
relational influence mechanisms among people advocating risky health behaviors
compared to those advocating healthy behaviors thorough examinations of their social
networks.
The second study focuses on information sharing behaviors. Information sharing
behaviors may differ depending on motivations people have to find peers with whom to
talk about their health issues. For example, information promoting risky behavior may be
8
paid greater attention and disseminated more widely within the social networks of those
advocating risky health behaviors. Park et al., (2017) found that information promoting
healthy eating was less likely to be shared among people with eating disorders. In other
words, although social media posts promoting healthy eating outnumbered posts
promoting eating disorders, people with eating disorders had lower chances of being
exposed to helpful content. Thus, examining information sharing behavior may provide a
critical clue to enhancing the persuasive power and effectiveness of social media
interventions. Study II aims to understand how informational influence mechanisms
differ among people advocating risky health behaviors and those advocating healthy
behaviors.
This dissertation will contribute to advance social media interventions by
providing clues to understand individuals who engage in risky health behavior compared
to those who engage in healthy behavior. Social media interventions designed for healthy
behavior advocates such as weight loss programs were found to be very effective
(Napolitano, Hayes, Bennett, Ives, & Foster, 2013). However, most social media
interventions designed for individuals who engage in risky health behavior such as
unprotected sex or binge drinking have been less effective or completely ineffective (Bull,
Levine, Black, Schmiege, & Santelli, 2012; Maher et al., 2014; Moreno, Grant,
Kacvinsky, Egan, & Fleming, 2012; Shaw, Mitchell, Welch, & Williamson, 2015). If
risky behavior advocates show different behavioral patterns on social media compared to
individuals who engage in healthy behaviors, a well-designed intervention for healthy
behavior may not work for them. This dissertation will help tease apart these mixed
9
results by examining risky health behaviors on social media from a social identity
perspective.
Chapter Summaries
This dissertation is organized as follows. The current chapter provides broad
background knowledge for this project. It argues that individuals engaging in risky health
behaviors have different motivations for finding similar others on social media compared
to individuals engaging in healthy behaviors. It also argues that different motivations may
influence patterns of relationship formation and information sharing behavior on social
media.
Chapter 2 provides conceptualizations of risky health behaviors focusing on
concepts of health. It reviews current literature on risky health behavior on social media.
Based on that it addresses problems of the phenomenon and implications of the current
study.
Chapter 3 proposes the theoretical framework and hypotheses for the first study. It
begins by introducing the Social Identity Perspective and applies it to the relationship
development mechanisms of people engaging in risky health behaviors. By comparing
the latter to people engaging in healthy behaviors, and drawing from literature on social
identification and social capital, two research questions and three hypotheses are
proposed.
Chapter 4 describes the theoretical background and proposes hypotheses for the
second study. It focuses on principal features of social media posts that attract advocates
of risky health behaviors. After reviewing literature on social support and affective tone,
one research question and three hypotheses are proposed.
10
Chapter 5 describes the methods used to analyze data and test hypotheses for the
two studies. It discusses the data collection processes, sampling strategies, data cleaning
procedures, measurements, and data analysis for each study. Then it reports descriptive
statistics on data collected and the results of hypothesis tests. Finally, Chapter 6 includes
a discussion of the findings and the theoretical and practical implications of those
findings. Limitations and future research are also discussed.

11
CHAPTER 2: UNDERSTANDING RISKY HEALTH BEHAVIOR ON SOCIAL
MEDIA
In this chapter, I outline the basic concept of risky health behavior and review
research on such behavior in social media. First, I describe risky health behavior based on
definitions of health. Then I explain how to understand individuals who engage in and
promote such behavior. Second, I review social media research with a focus on those
risky health behaviors. Finally, I discuss problems associated with these behaviors on
social media and highlight implications for the current study.
Understanding Risky Health Behavior
Defining Risky Health Behavior
Studies on behaviors that imperil one’s health, such as smoking and self-harm
have referred to these behaviors using the terms ‘risky health behavior’ (Jonah Berger &
Rand, 2008; Hoyt, Chase-Lansdale, McDade, & Adam, 2012; Klosky et al., 2012),
‘health-risk behaviors’ (Borawski, Ievers-Landis, Lovegreen, & Trapl, 2003; Williams,
Hedberg, Cox, & Deci, 2000) and ‘unhealthy behavior’ (Alexandrov, Lilly, & Babakus,
2013; Luxton, June, & Fairall, 2012). As there is little consensus among health
professionals and scholars, literature on these behaviors can sometimes be confusing
(Alexandrov et al., 2013; Keelan, Pavri-Garcia, Tomlinson, & Wilson, 2007; Luxton et
al., 2012).
The current dissertation will use ‘risky health behavior’ to refer to health-related
behaviors that may cause critical health problems including death. ‘Health-risk behaviors’
is a widely used term to refer to problematic health behaviors including excessive alcohol
use and unprotected sex with multiple partners. However, this term often includes general
12
behaviors that may indirectly lead to health problems such as violent crime and
dangerous driving habits (Williams et al., 2000). This dissertation focuses on health
behaviors that are directly related to one’s health rather than general behaviors that might
indirectly effect health. Unhealthy behavior focuses more on specific health-related
behaviors rather than general behavior that may have side effects on health. It
encompasses risky behaviors such as smoking, as well as less healthy behaviors such as
the consumption of fast food. Thus, ‘risky health behavior’ would be an appropriate term
to refer to health behaviors that directly imperil one’s health. Clarifying the concept of
‘health’ would help better define the concept of risky health behavior and avoid
unnecessary confusion.
One of the most influential definitions of health is the one established by the
World Health Organization (1946): “health is a state of complete physical, mental and
social well-being and not merely the absence of disease or infirmity.” The International
Epidemiological Association takes a broad view of health: “a state characterized by
anatomical, physiological and psychological integrity, ability to perform personally
valued family, work, and community roles; ability to deal with physical, biological,
psychological and social stress; a feeling of well-being; and freedom from the risk of
disease and untimely death (Jonas, Goldsteen, Goldsteen, & Jonas, 2013, p.3).” Thus,
health can be broadly understood as a state of complete physical and psychological well-
being, ensuring one’s ability to perform social roles free from the risk of disease or
infirmity.
As health is viewed in multiple levels ranging from a personal state to social roles,
health behavior broadly refers to “the actions of individuals, groups, and organizations, as
13
well as their determinants, correlates, and consequences, including social change, policy
development and implementation, improved coping skills, and enhanced quality of life
(Glanz, Rimer, & Viswanath, 2008, p.12).” Focusing more on individual aspects, three
categories of health behavior are developed: “any activity undertaken by a person
believing himself to be healthy, for the purpose of preventing disease or detecting it in an
asymptomatic stage,” “any activity undertaken by an individual who perceived himself to
be ill, to define the state of his health and to discover a suitable remedy,” and “any
activity undertaken by those who consider themselves ill, for the purpose of getting well.
(Kasl & Cobb, 1966, p. 246).” Consistent with these individual-focused categories, health
behavior is defined as “those personal attributes such as beliefs, expectations, motives,
values, perceptions, and other cognitive elements; personality characteristics, including
affective and emotional states and traits; and overt behavior patterns, actions, and habits
that relate to health maintenance, to health restoration, and to health improvement
(Gochman, 1998, p.3).”
Considering these definitions of health behaviors, any activity undertaken by
individuals to harm health improvement, or personal attributes that promote negative
health states can be construed as unhealthy behavior. In this dissertation, unhealthy
behaviors that contribute to the leading causes of death, disability, and social problems
will be viewed as risky health behaviors. These behaviors include suicidal or self-
harming behaviors that contribute to injuries and violence, unprotected sexual behaviors
that cause unintended pregnancy and sexually transmitted infections, alcohol and other
drug use, tobacco use, unhealthy dietary behaviors, and inadequate physical activity
(CDC, 2016).
14
This project focuses on people who promote risky health behavior. Some people
who engage in risky health behavior may not like their behaviors. For example, a smoker
may want to quick smoking and a person who engage in extreme dieting may want to
recover from it. Some others may enjoy the risky health behaviors in which they are
engaged. An e-cigarette smoker may try different flavors to find his/her favorite. A
person who engages in extreme dieting may want to have an even more extremely thin
body. The next section will explain how to understand people who engage in and
promote risky health behavior.
Defining Individuals Who Promote Risky Health Behavior
The current project proposes two dimensions to define individuals who promote
risky health behaviors and individuals who promote healthy behaviors: behavioral types
(risky vs. healthy) and stance towards the health behavior (pro vs. anti). It defines
individuals who promote risky heath behaviors as those who engage in and encourage
risky health behavior such as smoking and self-harm. Individuals who promote healthy
behavior are those who encourage healthy behavior, such as exercise, or discourage risky
health behavior, such as smoking cessation. ‘Behavior engagement’ and ‘health status
caused by behavior engagement’ must be distinguished. Having a sexually transmitted
disease is health status caused by risky sexual behavior. A person living with HIV is
considered as a person who promote risky health behavior if the person engages in and
encourages unprotected sex. The person is considered to be one who promote healthy
behavior if the person engages in and promotes safe sex.
It is easy to find people who promote healthy behavior or who oppose risky health
behaviors. Those who promote healthy behavior may share information about the benefits
15
of healthy behavior and advice on performing those behaviors (Albrecht, Goldsmith, &
Thompson, 2003). Examples include tips on eating healthy foods, advice on joining a
running group, tips on starting exercise, and the benefits of sleep. They may share
information or emotional support with individuals who are ill to help them enhance their
physical or mental health (Braithwaite, Waldron, & Finn, 1999). They usually want to
help enhance individuals’ knowledge of health, positive attitude towards health, and
motivation to lead a healthier life.
They also share information to reduce or prevent others engaging in risky health
behaviors (Paek, Kim, & Hove, 2010; Yom-Tov, Fernandez-Luque, Weber, & Crain,
2012). Such information may include the negative consequences of risky health behaviors
and advice on stopping unhealthy habits. For example, they may share messages about
the fatal effects of third-hand smoke and the positive consequences of quitting smoking
in order to encourage individuals to quit smoking. Sharing information about suicide
warning signs may help individuals to detect family or friends at risk for suicide. Drug
addiction recovery advice may help individuals to overcome drug dependence.
Similarly, individuals who promote risky health behaviors may share information
regarding the benefits of risky health behaviors. This may include cool-looking images of
people smoking or personal experiences about extreme weight loss and enhanced
confidence. They may provide practical information about how to maintain those
behaviors. Examples include tips on vomiting to lose extreme weight and guidance on
hiding self-harm from family members (Dunlop et al., 2011; Forsyth & Malone, 2010;
Kim, Paek, & Lynn, 2010; Lapinski, 2006).
16
This project differentiates people who consciously engage in risky behaviors from
people who have mistaken beliefs and, as a result, do not engage in healthy behavior. For
instance, there are individuals who believe childhood vaccines cause autism, although
vaccines are necessary for children to develop immunity (Briones, Nan, Madden, &
Waks, 2012). Misinformed individuals may share their inaccurate beliefs with others and
convince them not to vaccinate their children. In this way they may negatively influence
the health behaviors of others. Another example is myths about hand sanitizer. Some
people believe that hand sanitizers are not effective in killing germs, and can even irritate
the skin. Thus, they share anti-hand sanitizer information to discourage others from using
hand sanitizer. However, it is a scientific fact that hand sanitizers are effective at killing
flu viruses and make people less vulnerable to catching the flu. Studies also show that
when it comes to killing germs, alcohol-based hand sanitizers are gentler on the hands
compared to soap and water (Grisham, 2014). Another example of a misconception is
that milk is not good for a runny nose because it produces more mucous. The fact is that
drinking milk might actually be helpful for runny noses because the vitamin D in milk
can help keep people’s immune system running optimally (Lai & Kardos, 2013). People
in the examples above hold a misconception that their behavior – not engaging in health-
promoting activities– actually promotes health.
This dissertation does not deal with misconception about health, but rather
focuses on advocacy of risky health behaviors. Individuals who promote risky health
behaviors are generally aware that the behaviors they encourage are risky and stigmatized.
Thus, they are motivated to protect themselves from social stigma. The characteristics of
these individuals will be addressed in Chapter 3. The following section will review
17
current literature on the activities of those who promote risky health behavior on social
media, namely the activities of relationship formation and information dissemination.
Risky Health Behaviors on Social Media
Communication technologies have been regarded as important corrective to the
loss of community and the weakening of face-to-face relations (Norris, 2004). In
particular, social media are believed to be a valuable technology in facilitating social
connections. A social media has been defined as “a networked communication platform
in which participants 1) have uniquely identifiable profiles that consist of user-supplied
content, content provided by other users, and/or system-provided data; 2) can publicly
articulate connections that can be viewed and traversed by others; and 3) can consume,
produce, and/or interact with streams of user-generated content provided by their
connections on the site (Ellison & boyd, 2013, p.158).” As communication technology
develops, participation in social media has become an increasingly popular practice
among Internet users. Many social media users reported feeling as strongly about
members of online communities as those in the offline world (Center for the Digital
Future, 2008).
While social media have received much positive attention, recent research has
also found that they are frequently used as communication channels for promoting risky
health behaviors (see Figure 1) and that such content easily reaches social media users
(Hanson et al., 2013; Mclaughlin et al., 2015). For example, user-generated videos on
YouTube which portrayed smoking in a positive light far outnumbered those portraying
smoking negatively (Forsyth & Malone, 2010; Paek, Kim, Hove, & Huh, 2014; Ribisl,
2003). It is also found that the pro-smoking messages received greater numbers of views
18
(Forsyth & Malone, 2010). Among adolescents, videos that positively described
smokeless tobacco have been widely shared via YouTube (Seidenberg, Rodgers, Rees, &
Connolly, 2012). Through content analysis, researchers have found that a large number of
adolescents uploaded videos showing their smoking behaviors and sharing tips on
chewing smokeless tobacco. According to Kim, Paek, and Lynn (2010), tobacco-related
social media posts capture users’ eyes as they usually feature sexy, young, and healthy-
looking females’ smoking.

Figure 1. Images of risky health behavior on social media

People not only share such content on social media, but also build online
communities to exchange information and to gain social support (see Figure 2). For
example, researchers found that like-minded people formed pro-smoking online
19
communities and created virtual smoking clubs (Ribisl, 2003; Wang et al., 2015). In these
communities and clubs, members share pro-smoking messages, pictures of celebrity
smokers, vaping experiences, opinions about e-cigarette flavors, and information about
various e-cigarettes brands (Ribisl, 2003; Wang et al., 2015). Pro-extreme dieting
community members commonly share information about effective weight loss including
the use of weight loss drugs (Chancellor, Mitra, & De Choudhury, 2016; Davies &
Lipsey, 2003; Lapinski, 2006; Tong et al., 2013; Yeshua-Katz & Martins, 2013; Yom-
Tov et al., 2012). They share photos of thin celebrities as motivation and obese people as
warnings. They also support each other in their weight-loss targets and share tips on
avoiding detection by professionals and family members when skipping meals.
Researchers concluded that overweight people and people of normal weight were seen as
being weak in the pro-extreme dieting community, and community members believed
that strength and beauty were attained through thinness (Davies & Lipsey, 2003;
Lapinski, 2006; Park et al., 2017; Tong et al., 2013). Also, a pro-smoking community,
called ‘Jenny Teen Smoking Page,’ provided information about smoker’s rights. Many of
the community members were influenced by this information and shared their stories
about their initiations to smoking. Many of them kept posting as their smoking habit
developed (Ribisl, Williams, & Kim, 2003).
More extreme risky behaviors such as self-harm or suicide are also commonly
observed. For example, there are numerous reports of non-suicidal self-injury videos and
suicide announcements on social media (Biddle, Donovan, Hawton, Kapur, & Gunnell,
2008; Chhabra & Bryant, 2016; Lewis, Heath, St Denis, & Noble, 2011; Lewis, Heath,
Sornberger, & Arbuthnott, 2012; Ruder, Hatch, Ampanozi, Thali, & Fischer, 2011).
20
According to Luxton et al., (2012), among suicide-related websites that were available on
social media, approximately half were pro-suicide sites. In pro-suicide communities,
people shared information on effective methods or stories about persons who committed
suicide (Bell, 2007; Dunlop et al., 2011). Some individuals even made ‘suicide pacts’ to
commit suicide together.

Figure 2. An example of online communities promoting extreme dieting.

21
Social Influence of Risky Health Behavior on Social Media
As mentioned above, information promoting risky health behavior is prevalent on
social media. It is not difficult for individuals who engage in or who are interested in
risky health behavior to find people willing to share information and their experiences.
Theories of social influence provide a useful framework for understanding the effects of
risky health content on social media users. They imply that the content not only
influences one’s health behavior, but also normalizes those risky health behaviors.
Social comparison theory (Festinger, 1954) allows researchers to understand how
perceptions of others within a referent group may influence individuals’ viewpoint on
health information and decision-making processes on health related issues. According to
social comparison theory, individuals compare their health and coping mechanisms with
those of others in their referent group in order to make assessments and decisions about
their own health behavior (Wright, 2009). In the context of risky health behavior, social
media may facilitate social comparisons by allowing individuals to see other social media
users engage in risky health behavior.
Comparing oneself to others can lead to positive or negative self-assessments
(Wright, 2009). For example, when a person perceives that others are coping with
problems more effectively than himself or herself, an upward comparison is created and
this may either produce a feeling of frustration, or serve as inspiration to learn and adopt
the effective behaviors. On the other hand, when a person feels that s/he is dealing with
health issues better than other members, downward comparisons in the network could
lead to a positive self-assessment or to feelings that other members are unhelpful (Wright,
2009). Since participating in or withdrawing from social media groups is relatively
22
painless compared to face-to-face withdrawing from offline networks, individuals who
feel that the group is frustrating or unhelpful may withdraw from that network and find
another network that makes them feel more inspired and confident. Ultimately,
individuals only remain in online networks that make them feel positive about themselves
and other members in the network.
Member interactions resulting in positive self-assessments and positive
assessments of others may validate their risky health behaviors. Discussions between
others in their social network may also increase their confidence in their own behaviors.
For example, a smoker may begin to believe that smoking does not lead to lung cancer as
they communicate with fellow risk-takers via social media. Also, the smoker will feel
safe by thinking that s/he is not the only one who has the same interests. Through this
process, smokers may reduce uncertainty about their own beliefs and relieve stress
arising from negative social views about smoking behavior.
Social cognitive theory helps explain the process by which risky health behaviors
are normalized on social media. According to Bandura (1995), people can learn behaviors
not only from their own direct experience, but also by observing and modeling other
people’s behavior and the results of their actions. From the perspective of social learning
theory, risky health behaviors can be reinforced by positive rewards such as peer
acceptance within a community. Furthermore, social cognitive theory emphasizes that
individuals’ behavior can also be learned and reinforced by observation of others’
behaviors. That is, a person would learn to adopt a behavior after seeing that his/her
friend receive rewards from that behavior (Kim et al., 2010). In this sense, behaviors
rewarded online are powerful socializing forces, shaping views of what is ‘cool’ and
23
‘attractive’ (Bandura, 2001). For example, in a pro-eating disorder community, a person
may observe that when a member of the community shares a ‘selfie’ showing her
extremely thin body, she receives numerous ‘likes’ and positive comments from other
community members. By contrast, when another member shares a ‘selfie’ showing her
normal weight, it can be observed that she does not receive as much attention from other
community members. Comparing these two cases, the observer would learn that the
community values extremely thin bodies more than normal ones. This phenomenon,
referred to as ‘outcome expectation’, is the primary determinant influencing one’s
perception of a particular behavior in social cognitive theory. This process is also related
to the formation of positive social images (Forsyth & Malone, 2010; Kim et al., 2010).
On social media, pictures of underweight people would likely be depicted as more
sexually attractive and cool compared to pictures of people of normal weight. Thus,
observers would not only learn that risky health behaviors yield positive outcomes, but
they would also associate positive social images with those behaviors. This process may
help reinforce risky behaviors by normalizing them.
This social learning process may increase self-efficacy of engaging in risky health
behavior. Self-efficacy is “a person’s beliefs about her capacity to influence the quality of
functioning and the events that affect her life (Glanz, Rimer, & Viswanath, 2008, p.
172)”. Bandura (1995) emphasizes that self-efficacy, when combined with outcome
expectations, can lead to actual behavior. Social media content promoting risky health
behavior may contribute to enhancing one’s self-efficacy for risky behavior by providing
tips on initiating or maintaining such behavior (Tong et al., 2013). Throughout this
24
process, community members are more likely to engage in risky behavior without feeling
uncomfortable.
Theory of reasoned action (TRA) and Theory of planned behavior (TPB) also
provide a framework to understand the normalization process. They suggest that
behavioral intention is mainly determined by attitudes toward performing the behavior,
subjective norm associated with the behavior, and perceived control over the behavior.
And the behavioral intention predicts a number of different behaviors including health
behaviors (Montano & Kasprzyk, 2008).
Individual’s subjective norm is determined by “his or her normative beliefs, that
is whether important referent individuals approve or disapprove of performing the
behavior, weighted by his or her motivation to comply with those referents (Montano &
Kasprzyk, 2008, p. 71).” Thus, when a person believes that a certain behavior is accepted
by referent groups, he or she will hold a positive subjective norm towards the behavior.
For example, if a person believes that his friends think he should try drug injection, he
may hold positive subjective norm towards drug injection, and this may influence
intention to engage in drug injection. Perceived control is determined by “control beliefs
concerning the presence or absence of facilitators and barriers to behavioral performance,
weighted by their perceived power or the impact of each control factor to facilitate or
inhibit the behavior (Montano & Kasprzyk, 2008, p. 71).” A person who perceives s/he
has a low barrier to engage in a certain behavior will have greater perceived control. For
example, an athlete may have negatively perceived control towards drug injection
because athletes frequently encounter drug testing.
25
In social media communities promoting risky health behaviors, individuals would
be more likely to form positive attitude towards risky health behaviors as they see that
such behaviors are encouraged. Informational support in these communities may also
lower the barriers and increase members’ perceived control.
Concepts in cultivation studies are helpful for understanding the difference
between heavy users and light users in social media in terms of normalization (Williams,
2006). Cultivation studies propose that heavy television viewers are more likely than
light viewers to perceive reality as television portrays it (Gerbner, Gross, Morgan,
Signorielli, & Shanahan, 2002). For example, when frequent violence is shown on
television, heavy television viewers are likely to perceive that the world is more violent
than it really is, to be less trusting of others, and to overestimate the number of people
employed in law enforcement. This cultivation process is believed to occur through two
different mechanisms: mainstreaming and resonance. Mainstreaming occurs when heavy
viewers tend to share a similar worldview regardless of their socio-demographic
background. Resonance occurs when what is seen on television is congruent with the
experience of people’s lives, in which case they receive a double dose of the cultivation
effect (Gerbner et al., 2002).
In social media communities promoting risky health behavior, heavy users have
more chance to be exposed to information about risky health behavior than light users. In
other words, heavy users’ estimates of the prevalence of individuals engaging in risky
health behavior will be higher than light users’ estimates. In addition to individual
processes of mainstreaming and resonance, heavy users’ beliefs may collectively
converge, regardless of user socio-demographics, to a shared perception that risky health
26
behaviors are common and not extreme. In this process, the more users participate in
online conversation with others engaging in risky health behaviors, the more likely they
are to normalize these behaviors.
As theories of social influence imply, social media users who engage in risky
health behaviors are influenced by other users in their social network through information
sharing and social support. Previous studies of social media have advanced our
understanding of how the harmful behaviors of some users distort the health beliefs and
health behaviors of others. For example, pro-smoking messages on social media tend to
increase positive attitudes about smoking and users’ intentions to smoke (Yoo et al.,
2016). It has also been found that pro-eating disorder content exacerbates or maintains
users’ eating disorder symptoms (Rouleau & von Ranson, 2011). However, the
underlying mechanisms of the way people influence each other have yet to be examined.

27
CHAPTER 3: STUDY I. INDIVIDUALS WHO PROMOTE RISKY HEALTH
BEHAVIOR AND THEIR RELATIONSHIP FORMATION ON SOCIAL MEDIA
This chapter aims to examine how individuals promoting risky health behavior
form and maintain relationships on social media. Understanding relationship formation is
a first step to learning how individuals who promote risky health behavior influence each
other. The first study adopts the social identity frameworks for theoretical guidance.
Specifically, literature on social identification processes is reviewed to understand the
motivation to be connected to similar others. The motivations of group members
influence the type of networks they form and the type of social capital that may be
derived from these networks. Thus, literature on social capital will be discussed to learn
how those motivations would affect relationship formation. From this, hypotheses and
research questions will be derived.
Motivations of Relationships Formation for Individuals Who Promote Risky Health
Behavior
Stigma
Goffman (2009) argues that stigmatized individuals in an unaccepting world may
search for others who bear the same stigma and are therefore ready to adopt or accept an
alternative worldview. When those stigmatized individuals join together as a group, they
provide one another with a circle of empathy and a sense of belonging. However, it is
hard for stigmatized individuals to find one another in ‘face-to-face’ social networks
because they are socially marginalized (McKenna, Bargh, Levine, & Moreland, 2004).
Social media environments, by contrast, provide greater opportunities to find those who
are experiencing the same frustrations and anxiety. Consistent with these reasons,
28
stigmatized individuals have been found more likely to go online in order to fulfill
important social and psychological needs and to express their worldviews freely
(McKenna et al., 2004; Tong et al., 2013).
Risky health behaviors, including smoking, self-harm, and extreme dieting, can be
construed as stigmatized health behaviors. Such behaviors are not promoted in public.
Instead, family, friends, non-profit organizations, and pressure groups encourage
recovery from the behavior. It is not easy for people engaging in risky heath behaviors to
discuss their behavior with others due to negative social reactions (Tong et al., 2013).
Consequently, they are motivated to hide their risky health behaviors from others,
especially from those to whom they are emotionally attached.
Social Identity
The formation of social relationships on social media may differ depending on
users’ motivations for finding peers. Accordingly, the mechanisms of relationship
formation and social influence among those promoting risky health behaviors may differ
from the mechanisms among individuals promoting healthy behavior. The social identity
perspective provides a useful framework for understanding these potentially different
patterns of social relationship formation.
The social identity perspective addresses the self-conception of members of a
social group, including differentiation within groups, deviance, group culture, group
decision making, and collective action (Hogg et al., 2004). The concept of group, in the
social identity perspective, includes actual groups (e.g. small groups or organizations) as
well as socially perceived groups based on gender, ethnicity, education, etc. From the
social identity perspective, individuals engaging in risky health behavior are most likely
29
to identify with groups of people who engage in similar behavior. Strong group
identification may contribute to the formation of cohesive social networks as members
learn and follow group norms and others’ beliefs. The processes involved in developing
strong group identification can be explained by the concept of social identity.
Social identity is a widely employed concept addressing the relationship of the
individual to the group and the emergence of collective phenomena from individual
cognitions (Brown, 2000). Tajfel (1982) defined social identity as “the part of the
individuals’ self-concept which derives from their knowledge of their membership of a
social group (or groups) together with the value and emotional significance of that
membership (Tajfel, 1982, p. 24).” Thus, while personal identity is “a self-construal in
terms of idiosyncratic personality attributes that are not shared with other people (“I”) or
close personal relationships that are tied entirely to the specific other person in the dyadic
relationship (“me” and “you”) (Hogg et al., 2004, p. 251)”, social identity is a collective
self-construal such as “we,” “us,” and “them” based on the way group members identify
themselves and define who they are, attributes group members have, and the way group
members are related to and differ from specific out-groups (Hogg et al., 2004).
Thus, social identification, the process of building social identity, is derived
primarily from group memberships. Social identification begins with social
categorization. Social categorization is the process of perceiving, defining, and
recognizing both the self and others as members of a distinct social group. For example,
when a person identifies himself/herself as a member of the group of people who smokes,
his/her membership would become salient when he/she meets an anti-smoking activist.
Thus, social categorization enables individuals to predict their own and others’ behavior,
30
and instructs them how they should feel and behave. Once social categorization has
occurred, individuals define and distinguish their social circle from others based on
common and representative attributes. Finally, individuals attribute perceived
characteristics of the in-group to the self, a process called ‘self-stereotyping’ (Mackie,
1986). This social identification process clarifies the self-concepts and social norms held
by members of a particular group. For example, identifying oneself as a member of the
group of smokers clarifies what beliefs ought to be important to him and how he should
behave when meeting with smoking advocates and anti-smoking activists.
Based on the psychological mechanism involved in social identification processes,
previous research has shown that individuals who have high uncertainty about themselves
and their behaviors are more motivated to identify, and to identify more strongly with
groups (Hogg & Reid, 2006). Uncertainty-identity theory, in particular, explains how
individuals who have high uncertainty about themselves are motivated to engage in social
identification processes. According to uncertainty-identity theory, “people are motivated
to reduce feelings of uncertainty about or related to themselves; identifying with a group
reduces self uncertainty because the group’s attributes are cognitively internalized as a
prototype that describes and prescribes one’s own attitudes, feelings, and behavior, and
these attributes are consensually validated by fellow group members (Hogg & Blaylock,
2011, p. 20)”.
Moreover, uncertainty-identity theory proposes that people with high self
uncertainty are likely to identify more strongly with groups that are more distinctive,
more clearly structured, and associated with a clearer prototype (Hogg & Reid, 2006).
People seek to join high entitivity groups because clearly defined groups are most
31
effective at reducing self-uncertainty. For example, a woman who has high uncertainty
about her smoking behavior may want to identify with groups of people who encourage
and support smoking behavior rather than groups of people who share information about
tobacco industries. When individuals with high uncertainty cannot find those groups, they
may even transform existing communities to be more entitative (Hogg & Blaylock, 2011).
In addition, the social identification perspective suggests that when individuals
feel their security, lifestyle, and status are under threat, they seek to identify strongly with
a group to remove or buffer the threat (Branscombe, Schmitt, & Harvey, 1999; Hogg &
Blaylock, 2011). Individuals perceiving themselves as victims of social prejudice may
tend to identify strongly with pro-smoking groups because these groups minimize the
extent to which people are victimized. It has also been established that the identification
strength among low-status/devalued group members is greater compared to lower-
status/devalued group members (Branscombe et al., 1999). Also, recognizing threats to
the minority groups’ status, such as the minority groups’ experience of discrimination
and prejudice, will lead to minority group members’ identification with the in-group
because they become strongly reliant on their group as a means of building a meaningful
and positive self-concept (Branscombe et al., 1999).
Therefore, compared with individuals engaging in healthy behavior, individuals
engaging in risky health behavior will be more likely to identify, and to identify more
strongly, with their social groups because they are feeling uncertain about their beliefs
and behaviors that are relevant to self-concept. Once individuals identify with a group
promoting risky health behavior, they might expect their social group to be categorized as
32
a socially devalued group. This will also reinforce individuals engagement with group
identification.
Thus, the in-group cohesion of individuals engaging in risky health behavior may
be enhanced over time. Since social media users who promote risky behaviors provide a
clear prototype, other pro-smoking users can easily grasp group norms and behave
accordingly. Through this process, pro-smokers form relationships with other pro-
smokers on social media in a way that reinforces in-group coherence.
Social Capital in Social Media and Formation of Social Networks Promoting
Risky Health Behavior
A major function of social media is to form and develop social relations among
users, leading to the emergence of large-scale social networks (Kwak, Lee, Park, & Moon,
2010). Social networks can be understood by looking at different types of social functions
that a social network serves: social networks that mainly serve bridging functions and
social networks that mainly serve bonding functions. Putnam (2000) focused on social
capital and explained the functions social networks have. Bridging social capital brings
together people of different sorts. It is beneficial for building social capital, interpersonal
trust, and reinforcing community ties. Bridging networks expose people to new ideas and
various pieces of information. On the other hand, bonding networks bring together people
of a similar sort and encourages similar beliefs and values (Norris, 2004). Thus, bonding
social capital reinforces the ideas and information people already have or believe in
common.
Individuals who promote risky health behavior are more likely to form social
networks that serve strong bonding functions. As described in the previous section, they
33
are more likely to develop social relationships in ways that reinforce in-group coherence.
By contrast, those who promote healthy behavior have less need to reinforce their own
beliefs about healthy behaviors, and therefore have less motivation to identify with
groups of people who engage in healthy behaviors. Individuals who promote risky health
behavior also have limited access to information encouraging the behaviors that interest
them. Moreover it is difficult to find risk-taking peers in offline networks. Previous
studies indicated that risk-takers actively search for peers and develop social relationships
in order to compensate for their unfulfilled social needs and in order to freely express
their views of health issues (McKenna et al., 2004; Tong et al., 2013). Those who
promote healthy behavior, on the other hand, can easily obtain social support from both
their offline and online networks. Information promoting health behavior is prevalent and
easily accessible. They also have less difficulty finding healthy peers through their social
networks. Thus, they might use social media to search for additional information rather
than to meet peers.
Individuals who promote risky health behaviors will also be more likely to
strongly bond together when forming a social network. These strong social ties will
increase with homophily and contribute to forming a highly cohesive social network
(Granovetter, 1973). On the other hand, those who promote healthy behaviors will be less
likely to strongly bond together and, thus, will be less likely to have a cohesive social
network.
Bonding social networks promoting unhealthy behavior may serve a positive
function in terms of providing emotional support, but it is also dangerous because the
network may force group members to share unhealthy beliefs and engage in risky health
34
behaviors (Norris, 2004). Thus, social media promoting deleterious health beliefs may
have greater influence on members’ health behaviors compared to a positive health-based
social networks.
Drawing on the social identity and social capital perspectives, the following
hypotheses may be proposed:

H1: Individuals are more likely to form social ties with those who share their
stance towards risky health behavior than those who have different stance on social
media.
H2: Individuals who promote risky health behavior will form more cohesive
networks than those who promote healthy behavior on social media.
H3: Social ties to individuals holding the opposite stance will be more frequently
observed among individuals who promote healthy behavior than among those who
promote risky health behavior on social media.

In addition, this study examined characteristics of the sources of information (i.e.
individuals, non-profit or academic groups, media, and commercial entities) to advance
our understanding of social networks promoting risky health behavior. Studies have
found that various sources of information differently characterized health issues
(McLaughlin et al., 2016). By investigating source characteristics, we can learn how
health information would be framed in such social networks Thus, the following research
questions are proposed:

35
RQ1: Are source characteristics (i.e. individuals, non-profit or academic groups,
media, and commercial entities) in networks promoting risky health behavior different
from those in networks promoting healthy behavior?
RQ2: What are the characteristics of individuals who have social ties with those
holding an opposite stance?

36
CHAPTER 4: STUDY II. INFORMATION SHARING TO PROMOTE RISKY
HEALTH BEAHVIOR ON SOCIAL MEDIA
This chapter aims to examine how individuals who promote risky health behavior
share information on social media. Understanding information sharing is the next step to
learning how these individuals influence each other on social media. A large number of
posts promoting risky health behavior are posted on social media, but not all posts are
widely disseminated. Disseminated information is more likely to influence social media
users as users may not be exposed to non-disseminated posts. Thus, what kinds of
information they pay attention to and spread is necessary to examine the informational
influence mechanisms.
To understand how information is shared, the second study focuses on principal
features of social media posts that attract individuals who promote risky health behavior.
Literature on social identity implies that these individuals have motivations to gain
support and protect their identity. There are different types of social support including
informational and emotional support. Examination of types of social support that attract
such individuals will unfold their specific needs and information sharing behavior on
social media. Their motivations are also related to encouragement or defense mechanisms
which arouse positive or negative emotions. Social media posts with positive or negative
affective tones would evoke such individuals’ act on information sharing. Based on that,
literature on social support and affective tone are reviewed. From these, hypotheses and
research questions will be derived.

37
Dissemination of Information with Social Support
The Role of Social Support in Information Sharing
People use media to gain information that they want (Rubin, 2009). Thus,
messages are more likely to become popular when the content meets individuals’ needs.
Social media studies find that the most frequently cited motivations of individuals
searching for peers online are informational support and emotional support (Park, Kee, &
Valenzuela, 2009; Ridings & Gefen, 2004; Wasko & Faraj, 2000). Considering that
individuals who promote risky health behavior have little information and support offline,
they may have stronger motivation to gain social support online than do healthy behavior
advocates. For example, practical information about getting drugs and messages
supporting smoking may be of great interest and therefore may disseminate widely
among advocates of risky health behavior.
Social support is defined as “the social resources that persons perceive to be
available or that are actually provided to them by nonprofessionals in the context of both
formal support groups and informal helping relationships (Cohen, Underwood, Gottlieb,
& Institute, 2000, p.4)” Social support gives people the feeling of being together with
others, the notion of being part of a group, spending time together, companionship, and
networking (Ridings et al., 2006). According to Herring (1996), the freedom to express
views and to receive social support is the main reason for individuals to join and use
online networks. Studies have found that social media provide emotional support and
sociability (Chung, 2013; Hiltz & Wellman, 1997; Parks & Floyd, 1996).
The major motivation both for individuals who engage in healthy behavior and
risky health behavior to engage in social media is access to information (Park et al., 2009;
38
Ridings et al., 2006). Information about tips, skills, or solutions to maintain or promote
health behaviors are significant factors affecting information popularity on social media
(Ruder et al., 2011; Yom-Tov et al., 2012).
Emotional Support for Individuals Who Promote Risky Health Behavior
While informational support plays a significant role in information sharing
behavior in general, individuals engaging in risky health behaviors may benefit more
from emotional support than people who engage in healthy behavior. For example, when
an individual’s healthy behavior is known to others, s/he is likely to receive emotional
support that encourages this behavior. By contrast, few people would support risky health
behavior and share the same feelings. Most people would express negative reactions and
discourage the behavior. Thus, they may have fewer opportunities to gain emotional
support such as sympathy, understanding, encouragement, and physical affection than
healthy behavior advocates. Studies that have focused on the emotional support functions
of social media found that among survivors of childhood cancers, individuals with little
emotional support from offline social networks had higher rates of participation in social
networks with other survivors (McLaughlin et al., 2012). The buffering hypothesis also
supports the emotional support function of social media, positing that a social network
helps to shield individuals from major crises and the everyday sources of stress that they
may experience (Cohen & Wills, 1985). Therefore, supportive messages that provide
emotional acceptance, allow cathartic ventilation, and encourage perspective shifting
might fulfill needs that are not being met in individuals’ offline lives, and create a place
of safety apart from offline social networks that make them feel stigmatized (McKenna et
al., 2004; Tong et al., 2013; Wright, 2009).
39
Individuals engaging in risky health behaviors desire to meet risk-taking peers in
order to exchange social support. Online social networks supporting their risky health
behaviors provide a place where users feel less stigmatized about their behavioral choices,
share more sensitive information and honest feelings, and can reinforce their perceived
self-efficacy for engaging in risky behaviors. Consequently, they may pay closer
attention to information that fulfills their needs and may also be more likely to
disseminate such information.
These intuitions are summarized in the following set of hypotheses:

H4. Information with social support will be more prominent among highly
disseminated posts than among non-disseminated posts both for individuals who promote
risky health behaviors and those who promote healthy behaviors.
H5. Information relevant to emotional support is more likely to be disseminated
among individuals who promote risky health behavior than among those who promote
healthy behavior.

Dissemination of Information with Affective Tone
The Role of Affective Tone in Information Sharing
The affective tone of messages has been found to be a critical factor in
information sharing behavior (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001;
Thelwall, Buckley, & Paltoglou, 2011; Walther, Heide, Hamel, & Shulman, 2009). Early
work indicated that affective stimuli are more efficient in promoting humans’ survival
fitness (e.g. Bradley, 2014; Shoemaker, 1996). Recent research has also confirmed that
40
affective messages increase cognitive involvement and the level of activation (Bayer,
Sommer, & Schacht, 2012; Berger, 2011; Heath, 1996; Rime, 2009), and thus motivate
information seeking and sharing behavior (Berger, 2011; Berger & Milkman, 2012). By
the same token, Huffaker (2010) provided empirical evidence that affective messages
were more likely to receive feedback in online discussion forums.
In line with these findings, salient affective tones have been detected in posts
disseminated on social media (Pang & Lee, 2008). Conducting sentiment analysis on a
month’s worth of tweets, Thelwall et al., (2011) found that tweets expressing sentiment
were most likely to be propagated. Similarly, in a study of 165,000 tweets about German
politics, Stieglitz and Dang-Xuan (2013) demonstrated that messages containing
emotional cues get disseminated more quickly and frequently compared to those with
neutral cues.
Affective tone is expected in posts disseminated both among individuals who
promote risky health behavior and individuals who promote healthy behavior. However,
it is uncertain whether messages with positive or negative tones will be more frequently
observed among people who advocate risky health behaviors. Valenced messages could
arouse positive emotions, such as joy, love, and interest, as well as negative emotions,
such as anxiety, sadness, and anger. Depending on group members’ interests or social
experiences, they may pay greater attention to messages with a specific affective tone. In
other words, messages arousing positive emotion may have greater influence on group
members than messages arousing negative emotion, or the other way around.

41
Positive and Negative Affective Tones for Individuals Who Promote Risky Health
Behavior
Individuals promoting healthy behavior are expected to share information with a
positive affective tone. They need self-efficacy and supportive interactions with close
others to promote and maintain their healthy behavior. Relevant literature suggests that
they use social media as a way to fulfill needs that were not being met within their offline
social circle including family and friends (Chung, 2013; McLaughlin et al., 2012). When
they do not get enough support, they find others who share similar interests on social
media to receive encouragement about their healthy behavior. Thus, they would pay more
attention to information with a positive affective tone and therefore, share such
information with others to exchange social support on social media.
Individuals who engage in risky health behavior may show similar patterns in
terms of sharing information with affective tones to those who advocate healthy behavior.
They often experience stigma – blemishes of individual character – from members of
their face-to-face social network (Yeshua-Katz & Martins, 2013). Stigma is “the situation
of the individual who is disqualified from full social acceptance” and stigmatized
individuals are “perceived as [having] weak will, domineering or unnatural passions,
treacherous and rigid beliefs, and dishonesty, these being inferred from a known record
of, for example, mental disorder, imprisonment, addiction, alcoholism, homosexuality,
unemployment, suicidal attempts, and radical political behavior (Goffman, 1986, p.16).”
Goffman (1986) argues that stigmatized individuals in an unaccepting social network
may search for similar others who are likely to adopt their worldview. Similarly,
individuals who promote risky health behaviors may pay greater attention to supportive
42
messages with a positive affective tone in order to alleviate negative experiences such as
isolation and frustration.
On the other hand, it could be possible that posts with a negative affective tone
are more likely to be shared among individuals who promote risky health behavior than
among those who promote healthy behavior. When experiencing stigma that violates the
preferred view of self, people tend to defend their self-esteem from the threatening event
(Baumeister, Dale, & Sommer, 1998; Freud, 1961). This defense mechanism usually
leads to exaggerated or extreme reactions (Baumeister et al., 1998). For example, having
been accused of having socially undesirable attitudes, people tend to aggressively assert
the opposite and attempt to prove it. Similarly, people who perceive stigma about their
risky health behaviors may pay greater attention to exaggerated or extreme posts
defending their position. As they defend their self-concept from those who would isolate
and frustrate them, the exaggerated or extreme posts are likely to convey negative
affective tone. For instance, a social media post strongly asserts that risky health behavior
has nothing to do with a weak will or an unnatural passion and is instead a life style
choice, the post is likely to assume a defensive stance and express negative emotions
such as anger and fear.
To sum up, the affective tone accompanying information plays a significant role
in information sharing behavior both among advocates of risky health behavior and
healthy behavior. However, the role of different affective tones must be explored if we
are to understand how information sharing differs among these two groups.
Taken together, this study explores the following hypothesis and research
question:
43

H6. Posts with an affective tone will be more prominent among highly
disseminated posts than among non-disseminated posts both for individuals who promote
risky health behavior and those who promote healthy behavior.
RQ3. Which affective tones (positive or negative) contribute more to post
dissemination among individuals who promote risky health behaviors compared to those
who promote healthy behaviors?

Focus of the Current Study
This dissertation examines smoking, the leading preventable cause of death in the
United States, as an empirical case of risky health behaviors. According to the U.S.
Department of Health and Human Services (2016), tobacco use causes more than 480,000
deaths each year in the United States. More than ten times as many people have died
prematurely from it than have died in all the wars fought by the United States. Tobacco
use harms nearly every organ of the body, causes many diseases, and reduces the health
of smokers in general. Smoking can cause cancer almost anywhere in one’s body and
smokers are from two to twenty-five times more likely than nonsmokers to develop heart
disease, stroke, and lung cancer (Center for Disease Control and Prevention (CDC),
2017).
After electronic cigarettes (e-cigarettes) were introduced, many adults – and even
youth– started to use e-cigarettes believing that they are harmless “juice” and help
smokers quit smoking. E-cigarettes, however, are not approved by the Food and Drug
Administration (FDA) to help people quick smoking. Rather, current research shows that
44
the liquid in e-cigarettes contains nicotine and possibly other chemicals. Nicotine is a
health danger as it is highly addictive and has negative effects on brain development. It is
also known to harm pregnant women’s and babies’ health. Moreover, it is found that
children and adults have been poisoned by swallowing, breathing, or absorbing the liquid
through their skin or eyes (CDC, 2015). Like smoking, e-cigarettes, so-called “vaping” is
also a risky health behavior.
Nevertheless, about 36.5 million adults in the United States still currently use
tobacco and more than 16 million Americans are living with a smoking-related disease
(CDC, 2017). Also, nearly 4 million youth under the age of 18 currently use tobacco
projects and e-cigarette use has increased considerably as teens are exposed to e-cigarette
advertisements from the Internet, TV, magazines, and retailo shops (CDC, 2017).
On social media, smokers have started to share tobacco or e-cigarettes-related
information including their positive experiences, reactions to a specific flavor, smoking
fetish pictures, and defenses against health campaigns that threaten smokers’ rights
(Forsyth & Malone, 2010; Kim et al., 2010; Paek et al., 2014). Pro-smoking content is
becoming more problematic as positive portrayals of smoking in social media have been
found to far outnumber negative portrayals (Paek et al., 2014). It has also been found that
social media users are overwhelmingly exposed to positive messages about e-cigarettes
(van der Tempel et al., 2016). That is, social media users are more likely to be exposed to
pro-smoking content than anti-smoking content.
Studies have found that pro-smoking content on social media influences receivers’
smoking-related attitudes and behavior (Huang et al., 2014; Yoo et al., 2016). According
to Yoo et al., (2016), expressing and receiving pro-smoking messages increased smoking
45
intentions and positive attitudes towards smoking. Huang et al., (2014) also found that
friends’ pictures of risky health behavior on social media were associated with
adolescents’ smoking. The remarkable thing was that adolescents who did not have
friends engaging in risky health behavior were also affected by higher exposure to risky
online pictures.
To understand the influence mechanisms of smoking promotion on social media,
the current study focuses on social networks of individuals who encourage tobacco or e-
cigarette use (pro-smoking users) and their information sharing behaviors on Twitter.
This study focused on Twitter because it is the most commonly used social media
application in public health and provides relatively reliable information about users’
social networks (Harris, Snider, & Mueller, 2013; Thackeray, Neiger, Smith, & Van
Wagenen, 2012). In Study I, I will compare pro-smoking users’ network structures (pro-
smoking networks) to anti-smoking users’ network structures (anti-smoking networks) on
Twitter to understand how pro-smoking users use social media to build and maintain
social relationships. Then, in Study II, the health information sharing behavior of pro-
smoking users will be compared to anti-smoking users within their respective
communities on Twitter. I also compare highly-disseminated and non-disseminated
messages in a pro-smoking group and an anti-smoking group in order to examine what
kinds of information are more influential among pro-smoking and anti-smoking users.

46
CHAPTER 5: METHODS AND RESULTS
The current dissertation employs computer programming techniques to collect
data from social media. Data extracted from social media have several advantages as a
source for studying health-related behavior. Unlike survey methodology or controlled
laboratory experiments (Chung, 2013; Olander et al., 2013), this approach allows direct
assessments of human behavior in natural settings. The naturalistic settings are especially
an advantage in the current study of risky health behavior. As people are reluctant to
share their socially undesirable behaviors with others, self-reports of risky health
behavior data has been questioned (Weinhardt, Forsyth, Carey, Jaworski, & Durant,
1998). Diverse assessment techniques have been developed to enhance reliability, but
there still are fundamental measurement errors when assessing risky health behavior.
Thus, this project extracted data from social media in order to provide accurate and valid
results.
This dissertation uses data from Twitter in order to test the hypotheses and
examine the research questions posed above. Twitter is the most commonly used social
media application in public health (Harris et al., 2013; Thackeray et al., 2012). Health-
related individuals, groups, and organizations have used Twitter as an important method
for information sharing (Chung, 2016; McLaughlin et al., 2016; van der Tempel et al.,
2016). Especially, individuals are found to use Twitter to share their personal experience
of stigma and attitudes towards stigmatized health issues, such as mental illness (Joseph
et al., 2015; Reavley & Pilkington, 2014). Twitter also provides an “access point” to
individuals who engage in risky health behavior (e.g. drug abuse) and they may reveal
more information to talk about their experience than they do offline (Stoddard et al.,
47
2012). In addition, social networks on Twitter are relatively reliable as social ties
between Twitter users (e.g. followings and followers) are publicly visible. By
investigating risky health behavior-related social networks and information sharing
behavior on Twitter, this project aims to contribute to the health communication literature.

Study I
The main purpose of Study I is to examine the relational influence mechanisms
among individuals who promote risky health behavior by comparing their social network
structure to that of individuals who promote healthy behavior. Study I tests the degree of
homophily with respect to attitudes about smoking in social networks on Twitter. Then,
pro-smoking networks and anti-smoking networks are compared in terms of network
cohesiveness and member exclusiveness. Lastly, I explore source characteristics (e.g.
individual, non-profit or academic, media, and commercial entities) of people in each
network as well as those bridging the two networks.
Data Collection
Data were collected from March 2017 to April 2017 after receiving approval from
the University of Southern California Institutional Review Board (IRB). Three steps were
involved in data collection. The first step involved extraction of pro-smokers’ and anti-
smokers’ following networks on Twitter. The current study identified initial seeds in the
smoking-related social network using a snowball sampling strategy, which is a widely
used method for sampling networks in large-scale social media (Ahn, Han, Kwak, Moon,
& Jeong, 2007). Under normal conditions, random sampling strategies would generate
less bias than snowball sampling. However, given the large number of automated
48
account-creating programs, known as bots on Twitter, random sampling on Twitter is
more likely to select bots. The large numbers of positive and negative bots on Twitter are
due to the platform’s popularity and open structure. Legitimate bots are beneficial as they
deliver news and update feeds. Malicious bots, on the other hand, spread spam, fake news,
or content that is harmful for commercial, political, or any other reasons (Chu,
Gianvecchio, Wang, & Jajodia, 2012). Lienemann, Unger, Cruz, and Chu (2017)
recommended identifying bots when studying smoking content on Twitter in order to
reduce bias. According to them, studies coding for bots reported that anywhere from 6.9%
to 80.7% of tweets were generated by bots. Detecting bots, however, is becoming more
difficult as more advanced bots are developed. Ferrara, Varol, Davis, Menczer, and
Flammini (2016) argued that it is currently impossible to accurately detect bots. Thus, in
the context of pro-smoking content on social media, snowball sampling is a much more
trustworthy and efficient sampling method.
Seed pro-smoking and anti-smoking accounts were selected in order to conduct
snowball sampling. The first trial involved in collecting smoking-related accounts on
Twitter using search terms “pro,” “anti,” “tobacco,” and “smoking.” A total of 2,618
accounts that posted the search term-related Tweets on one day were identified. Then, I
conducted personal investigation to verify human-generated accounts. Although several
bots were identified, many of the suspicious accounts were difficult to distinguish. As
Ferrara, Varol, Davis, Menczer, and Flammini (2016) suggested, it was almost
impossible to accurately detect bots.
Thus, I choose to identify pro- and anti-smoking groups’ websites on Google and
find their Twitter accounts. Pro-smoking users were identified using the search terms
49
“pro,” “tobacco,” “groups,” “smoker,” and “rights” and anti-smoking users were
identified using the terms “anti,” “tobacco,” “groups,” and “rights” on Google. The
search results returned seven pro-smoking groups and thirteen anti-smoking groups in the
first ten pages. Among them, nine were general health campaign groups and three were
government health-related organizations such as the Centers for Disease Control and
Prevention (CDC). One group was revealed to be supported by a tobacco company. In
order to focus on explicitly pro-smoking and anti-smoking users, these seven more
general groups and organizations were excluded during the process of initial seed
selection.
Official Twitter accounts were identified for the remaining three pro-smoking
groups and four anti-smoking groups. Accounts included their official group account or
the account of a group’s founder or chair. The founder or chair having the highest number
of followings was identified in each of the pro-smoking and anti-smoking related groups.
Likewise, the group with the highest number of followings was also selected from each
camp. Out of this process, four Twitter accounts were identified: a pro-smoking advocacy
group (PG; 142 following), a founder of a pro-smoking advocacy group (PI; 342
following), an anti-smoking pressure group (AG; 96 following), and a chair of an anti-
smoking pressure group (AI; 337 following).
Next, online media organizations associated with these four accounts were
identified in order to discover social media users who had no formal connection to these
groups offline, but followed pro-smoking or anti-smoking information. Three pro-
smoking and six anti-smoking media organizations were found to be associated with the
advocacy. The media organization with the highest following in each camp was included
50
among the seed accounts: one pro-smoking media organization (PM; 94 following) and
one anti-smoking media organization (AM; 35 following). At the conclusion of this
process, six Twitter accounts remained, one advocacy organization, one leader of an
advocacy organization, and one media organization account for each pro-smoking and
anti-smoking camp. These six accounts constituted the initial seed accounts.
A web-crawling program, written in Python, was used to uncover the social
networks derived from these initial seed accounts. After a few pilot tests of crawling on
Twitter, the web-crawling program was run on the six seed accounts. This process
revealed a total of 1046 Twitter users: 578 followings from the initial pro-smoking
accounts and 468 followings from the initial anti-smoking accounts. Users overlapping
multiple seed accounts were identified during the scraping process in order not to
duplicate scraping attempts. Among the pro-smoking followings, forty-four were
followed by two or more pro-smoking accounts. Of the anti-smoking followings, seven
were followed by two or more anti-smoking seed accounts. Thus, 530 and 461 unique
users were identified from the initial pro- and anti-smoking accounts respectively. In
addition, five accounts were followed both by a pro-smoking and anti-smoking seed
account. From the initial seed accounts, a total of 986 unique users were included in the
final list of pro- and anti-smoking networks.
The second step involved crawling these 986 users’ information and content
coding this information. User information included number of followers, followings,
tweets, and account users’ self-generated descriptions. This information was also
collected using a python web-scraping technique. Using the account users’ self-
descriptions, it was possible to code users’ explicit stances towards smoking and
51
characteristics of the sources including account types (e.g. individual, non-profit or
academic, media, and commercial entities), and industry/occupation of users (e.g.
pressure groups/activists, research, medical, journalists, political figures, consultants,
etc.).
The third step involved identifying following/follower relationships among the
986 Twitter users in order to model ties between network members. Using the Twitter
API, following and follower relationships between two users can be examined. Due to the
Twitter API rate limit for checking user ties, this process took about two months. The
initial seed accounts were excluded in the second and the third steps of this process
because all of the accounts were connected to at least one of the seed accounts, and
including these accounts would have distorted the entire structure of the network.
Through the data collection process, a total of 986 users and 23,492 following-follower
links between the users were identified and stored.
Among 986 unique users selected, a total of 285 users explicitly expressed their
stance towards smoking behavior in their profile descriptions. Specifically, 136 pro-
smoking users and 149 anti-smoking users were identified. The 285 users were used in
the analysis of H2, H3, RQ1, and RQ2.
Measurement
Smoking-related following networks.
Users’ ego network information was stored in a csv file, in the form of an edge
list. In a separate csv file, user account information obtained from the data crawling and
content coding was stored in the form of a node list. Network data cleaning was
performed using R.
52
Account types and industry/occupation of users.
As mentioned in the data collection section, characteristics of the sources
including account types and industry/occupation of users were coded based on self-
generated descriptions on users’ Twitter profile pages. The account types of Tweet
creators were classified among the following categories: ‘individual users’, ‘non-profit or
academic’, ‘media’, and ‘commercial entities’. Industry/occupation of users included
‘pressure groups/activists/foundations’, ‘academic/research/forums/conferences’,
‘medical’, ‘news/news agents/journalists/news sharing platforms’, ‘tobacco/e-cigarette
industries’, ‘political/marketing/consultant/life coaches’, ‘entertainments’, ‘technologies’,
and ‘artists/writers/travelers/other’. Users were classified as ‘not identified’ when they
did not have profile page descriptions or when they did not share information related to
source characteristics.
Network measures.
Several network-wide measures were calculated in order to compare the pro- and
anti-smoking networks. I used an R package called igraph (Csardi & Nepusz, 2006) to
generate network-level measures. The igraph package supplies a convenient toolkit of
network-level measures including density, reciprocity, transitivity, diameter, and average
path length (Csardi & Nepusz, 2006).Density is an especially key attribute of a network
and represents “the number of connections in the network reported as a fraction of the
total links possible (Valente, 2012, p.129).” The density of a network is calculated as:
! =
$
%(%−1)

where l is number of links and n is the total number nodes in the network.
53
Reciprocity and transitivity at the network level is calculated as the proportion of
reciprocated links or the proportion of triads in the network. Reciprocity measures
whether two nodes are mutually connected to each other. It is measured as (Csardi &
Nepusz, 2006):
R =
*.
*
,-
-
,
.
,-
-
,

where A-dot, indicated A., is the element-wise product of matrix A. A high proportion of
reciprocated ties indicates that group members are closely connected each other and may
suggest stronger ties (Valente, 2012). There are several generalizations of transitivity and
the igraph package uses a local vertex-level quantity assuming a weighted graphs (Barrat,
Barthelemy, Pastor-Satorras, & Vespignani, 2004):
/
0
1
=
2
3
,
(4
,
52)
1
,-
61
-7
8
9
0:
9
0;
9
:; :,;

where =
0
is the strength of vertex i, 9
0:
are elements of the adjacency matrix, >
0
is the
vertex degree, and ?
0
are the weights. Although the igraph package uses the formula
above, weights are not relevant as the social ties in the current study are not weighted.
Diameter and average path length (APL) indicate network cohesiveness (Valente,
2012). The diameter is defined as “the number of steps in the longest path in the
network”; and the APL is defined as “the average of the distances between all the nodes
in a network.” A large APL indicates greater overall distances between nodes (Valente,
2012, p. 134 - 135).” The APL is calculated as:
@ABC9DB E9Fℎ $B%DFℎ =
1
H
0:
%(%−1)

54
where d is the distance between nodes i and j and n is the number of nodes in the network.
A low diameter and small APLs indicate a cohesive network.
Analysis
To test hypotheses and answer the research questions proposed in Study I, chi-
squared tests, t-tests, and exponential random graph model (ERGM) analyses were
performed. ERGMs are a method of social network analysis that predicts the formation of
network links in terms of node attributes, edge attributes, and higher-order network
features (Valente, 2012). In the context of this dissertation, ERGMs allow us to test the
relationship between user behaviors and tie formation.
Results
In the 986 user accounts, the median number of followings were 886 (M =
16985.31; SD = 99,295.56), followers were 2,288.5 (M = 415,780.49; SD = 2,590,879.9),
and the tweets were 3,638 (M = 173316.86.49; SD = 44258.78). In the 530 users from the
pro-smoking seed accounts, the median number of followings were 666.5 (M = 6,804.04;
SD = 66,890.62), followers were 1368 (M = 198,523.50; SD = 1,542,452.45), and the
tweets were 2,679.5 (M = 14,630.92; SD = 37,215.50). In the 461 users from the anti-
smoking seed accounts, the median number of followings were 1,152 (M = 32,701.17; SD
= 141,748.25), followers were 6,143 (M = 668,288; SD = 3,395,072.82), and the tweets
were 4,938 (M = 20,497.68; SD = 50,913.61).
Among the 986 users, individual accounts were most prevalent (N = 558, 56.59%),
followed by non-profit or academic accounts (N = 174, 17.65%) and commercial entities
(N = 134, 13.59%). This distribution was similar in both pro-smoking and anti-smoking
follower networks. Fourteen users in the pro-smoking network did not have any
55
information about source characteristics in their profile descriptions, whereas all of the
users in the anti-smoking network provided this information.

Table 1. Frequency and Percentage of Source Characteristics in Social Networks from the
Pro-smoking and Anti-smoking Seed Accounts

Pro-
smoking
Anti-
smoking
Total
1
# of users obtained from seed accounts 530 461 986
Account Types
Individual 315(59.43) 247(53.58) 558(56.59)

Non-profit or academic 85(16.04) 93(20.17) 174(17.65)
Media 49(9.24) 57(12.36) 106(10.75)
Commercial 67(12.64) 64(13.88) 134(13.59)
Not identified 14(2.64) 0(0) 14(1.42)
Industry/Occupation of Users
Pressure groups/activists/foundations 149(28.11) 102(22.13) 250(25.35)
Academic/research/forums/conferences 26(4.91) 34(7.38) 58(5.88)
Medical (e.g. hospitals; doctors) 1(.19) 18(3.90) 19(1.93)
News/news agents/journalists/news
sharing platforms
84(15.85) 53(11.50) 137(13.89)
Tobacco/e-cigarette industries 47(8.87) 0(0) 47(4.77)
Political (e.g. politicians) 21(3.96) 4(.87) 25(2.54)
Marketing/consultants/life coaches 24(4.53) 103(22.34) 126(12.78)
Entertainment 36(6.79) 54(11.71) 89(9.03)
Technologies 8(1.51) 9(1.95) 17(1.72)
Artists/writers/travelers/other 84(15.85) 62(13.45) 146(14.81)
Not identified 50(9.43) 22(4.77) 72(7.30)
Note. 1) Five users are included in both the pro-smoking and anti-smoking following
networks. Totals represent unique users.

The distribution of user categories differed for pro-smoking and anti-smoking
networks. In the pro-smoking network, the most frequently observed user category was
pressure groups/activists/foundations (N = 149, 28.11%), followed by news/news
agents/journalists/news sharing platforms (N = 84, 15.85%) and
56
artists/writers/travelers/other (N = 84, 15.85%). In anti-smoking networks,
marketing/consultants/life coaches were most prevalent (N = 103, 22.34%), followed by
pressure groups/activists/foundations (N = 102, 22.13%) and
artists/writers/travelers/other (N = 62, 13.45%). Table 1 depicts the source characteristics
and user categories of all users, by stance and altogether.
Among 986 unique users selected, a total of 285 users explicitly expressed their
stance towards smoking behavior in their profile descriptions. Specifically, 136 pro-
smoking users and 149 anti-smoking users were identified. In the whole network with
285 users, the median number of followings were 655 (M = 2,758.15; SD = 14,913.66),
followers were 1,350.5 (M = 26,230.95; SD = 142,025.75), and the tweets were 2,679.5
(M = 8,280.65; SD = 17,490.18). In the pro-smoking network of 136 users, the median
number of followings were 698 (M = 1383.50; SD = 4647.86), followers were 1019 (M =
3929.12; SD = 16214.94), and the tweets were 2690 (M = 9496.84; SD = 20864.32). In
the anti-smoking network of 149 users, the median number of followings were 647 (M =
4056.95; SD = 20251.51), followers were 1,976 (M = 47,302.33; SD = 195,431.52), and
the tweets were 2,659 (M = 7,131.56; SD = 13,530.34).
While the characteristics of the sources did not appear to vary much between
users from pro-smoking and anti-smoking seed accounts when all users were included,
both account types and industry/occupation of users did appear to vary among users who
explicitly stated their stance (see Table 2).

57
Table 2. Frequencies and Percentages of Sources Characteristics in Pro-smoking and
Anti-smoking Networks, Users with Explicit Positions on Smoking

Pro-
smoking
Anti-
smoking
Total
# of users with explicit positions on smoking 136 149 285
Account Types
Individual 80(58.82) 46(30.87) 126(44.21)
Non-profit or academic 41(30.15) 93(62.42) 134(47.02)
Media 9(6.62) 5(3.36) 14(4.91)
Commercial 6(4.41) 5(3.36) 11(3.86)
Not identified 0(0) 0(0) 0(0)
Industry/Occupation of Users
Pressure groups/activists/foundations 93(68.38) 97(65.10) 190(66.67)
Academic/research/forums/conferences 3(2.21) 19(12.75) 22(7.72)
Medical (e.g. hospitals; doctors) 0(0) 16(10.74) 16(5.61)
News/news agents/journalists/news
sharing platforms
6(4.41) 10(6.71) 6(2.11)
Tobacco/e-cigarette industries 2(1.47) 0(0) 2(.70)
Political (e.g. politicians) 0(0) 0(0) 0(0)
Marketing/consultants/life coaches 2(1.47) 1(.67) 3(1.05)
Entertainment 4(2.94) 1(.67) 5(1.75)
Technologies 0(0) 0(0) 0(0)
Artists/writers/travelers/other 20(14.71) 4(2.68) 24(8.42)
Not identified 6(4.41) 1(.67) 7(2.46)

In the pro-smoking network, individual users were most prevalent (N = 80,
58.82%), followed by non-profit or academic (N = 41, 30.15%). On the other hand, in the
anti-smoking network, non-profit or academic users were most frequently observed (N =
93, 62.42%), followed by individual users (N = 46, 30.87%). Regarding user categories,
in the pro-smoking network pressure groups/activists/foundations were most frequently
observed (N = 93, 68.38%), followed by artists/writers/travelers/other (N = 20, 14.71%)
and news/news agents/journalists/news sharing platforms (N = 6, 4.41%). In the anti-
smoking network, on the other hand, pressure groups/activists/foundations were most
58
prevalent (N = 97, 65.10%), followed by academic/research/forums/conferences (N =19,
12.75%) and medical (N = 16, 10.74%).
To test H1 (homophily by stance), an ERGM was employed. As stated above,
bridging ties between all users except the seed accounts were included in the analysis.
The results showed that an explicit stance towards smoking behavior had a significant
impact on the formation of online social relationships, controlling for density and
reciprocity, (PE = .146, SE = .029, p < .05). Specifically, if two users expressed the same
stance towards smoking behaviors, the chance of one’s following the other was 1.20
times higher than among those with different stances. Thus, H1 was supported.
To test H2 (network cohesiveness), network-level analysis was performed
separately on the pro-smoking and anti-smoking networks. The result showed that pro-
smoking networks had higher cohesiveness than anti-tobacco users’ networks.
Specifically, the pro-smoking network showed significantly higher density than the anti-
smoking network (t(264.2) = 2.14, p = .033). The pro-smoking network also had
statistically higher reciprocity than the anti-smoking network (t(251.95) = 4.78, p < .001).
The difference of transitivity was marginally significant (t(254.49) = 1.77, p = .079). (see
Table 3). Also, the average path length and diameter were smaller in the pro-smoking
networks than the anti-smoking networks. However, the difference in the average path
lengths of the pro- and anti-networks was not statistically significant. It is important to
note that there is an inverse relationship between size and density: as size increases,
density decreases (Valente, 2012). Although considering the smaller size of the pro-
smoking network, it is reasonable to say that the pro-smoking network was somewhat
cohesive than the anti-smoking network. Thus, H2 was partially supported. Figure 3
59
represents the pro-smoking and anti-smoking social networks on Twitter. A post hoc
analysis compared networks from the pro-smoking and anti-smoking seed accounts. The
results showed that the two different following networks did not differ in terms of
network level measures. The implications of these mixed results are discussed later.

Table 3. Network-Level Measures of Pro-smoking, Anti-smoking, and Combined
Networks

Combined
network
Pro-smoking
network
Anti-smoking
network
Size 986 136 149
Density .02 .15 .12
Reciprocity .46 .65 .50
Transitivity .28 .58 .52
Average path length 3.11 2.14 2.33
Diameter 12 5 8

60

Figure 3. Visualization of the Network Using Gephi. Red dots represent pro-smoking
users while green dots represent anti-smoking users. Yellow links represent ‘pro-to-anti’
ties while blue links represent ‘anti-to-pro’ ties.

61
To test H3 (the exclusiveness of social relationships), a chi-squared test was
performed. The result was in the opposite directions to that expected. Specifically, 72.06%
(N = 98) of the all pro-smoking users followed anti-smoking users, whereas 21.48% (N =
32) of the all anti-smoking users followed pro-smoking users (c
8
= 15.33, p < .001) (see
Table 4). This constitutes 21.21% of total ties that pro-smoking users have and 6.13% of
total ties that anti-smoking users have. Thus, H3 was not supported.

Table 4. Bridging Ties between Pro-smoking and Anti-smoking Networks
Ties from “Pro” to “Anti” Ties from “Anti” to “Pro”
Number of users 98 (72.06%) 32 (21.48%)
Number of links 768 (21.28%) 169 (6.13%)
Chi-square tests c
8
(1) = 15.33, p < .001

A post hoc analysis showed that users were more likely to form social
relationships with those who have homophilous social ties in terms of stance (PE = .03,
SE = .01, p < .05). For example, pro-smoking users who were connected to anti-smoking
users were likely to have relationships with other pro-smoking users who were connected
to anti-smoking users as well.. Specifically, users having homophilous social ties had
1.03 times greater odds of forming a tie than users having heterogeneous social ties.
To explore RQ1 (source characteristics in pro- and anti-smoking networks), a chi-
squared test was performed. Individual users were most frequently observed in pro-
smoking networks while non-profit or academic group users were most frequently
observed in anti-smoking networks (c
8
(3) = 30.06, p < .001).
Tie formation in the anti-smoking network was also predicted by source
characteristics. Specifically, users sharing the same account types (i.e. individuals; non-
62
profit or academic groups) had 1.10 times greater odds of having a tie than users with
different account types (PE = .09, SE = .03, p < .05). However, account types were not a
significant factor in predicting tie formation in pro-smoking networks (PE = -.03, SE
= .04, p > .05). Thus, the analysis showed that the role of sources characteristics differed
in the two networks.

Table 5. Source Characteristics of Bridging Users
Pro-smoking users who
follow anti-smoking users
Anti-smoking users who
follow pro-smoking users
Account Types
Individual 62.24% 45.16%
Non-profit or Academic 29.59% 45.16%
Media 4.08% 3.23%
Commercial Entities 4.08% 9.68%
Industry/Occupation of Users
Pressure groups/Campaigns 73.47% 54.84%
Academic/Medical 5.10% 29.03%
Individual advocators/other 21.43% 19.35%

To explore RQ2 (characteristics of users who connect pro- and anti-smoking
networks), account types and industry/occupation of bridging users were identified.
Results showed that among pro-smoking users who followed anti-smoking users,
individual users were most prevalent (62.24%), followed by non-profit or academic users
(29.59%) (c
8
(3) = 8.29, p < .05). Regarding industry/occupation of users, among pro-
smoking users who followed anti-smoking users, pressure groups/campaigns were most
frequently observed (73.47%), followed by individual advocates/other (21.43%). Among
anti-smoking users who followed pro-smoking users, individual and non-profit or
academic users were observed with equal frequency (45.16%). Regarding
63
industry/occupation of users, among anti-smoking users who followed pro-smoking users,
pressure groups/campaigns (54.84%) and academic/medical (29.03%) were most
prevalent (see Table 5) (c
8
(2) = 19.54, p < .001).

Study II
Study II aims to examine the informational influence mechanisms among people
who promote risky health behavior, especially through testing their information sharing
patterns with the sharing patterns of people who promote healthy behavior. To achieve
this goal, the Twitter accounts of one pro-smoking organization and one anti-smoking
organization were selected and the dissemination pattern of Tweets was examined for
each.
This analysis tests whether users share Tweets with social support and an
affective tone more than other Tweets. It also examines whether Tweets with emotional
support are shared more by users in the pro-smoking group than by users in the anti-
smoking group. Analysis was also conducted to test whether Tweets with an affective
tone were more likely to be shared by users in the pro-smoking group or the anti-smoking
group.
Data Collection
Data were collected between April 12 and April 20, 2017 after receiving approval
from the University of Southern California Institutional Review Board (IRB). Target
advocacy groups were selected from the groups collected in Study I. Through the data
collection process in Study I, four pro-smoking groups and ten anti-smoking groups were
identified using search terms “pro”, “tobacco”, “groups”, “smoker”, and “rights” for pro-
64
smoking groups and “anti,” “tobacco”, “groups”, and “rights” for anti-smoking groups on
Google. After excluding groups that were not directly related to smoking, three pro-
smoking and four anti-smoking groups were identified. All of the groups had official
Twitter accounts except one pro-smoking group. Groups were selected that had the
highest number of tweets and the largest number of followers. The selected pro-smoking
group had 8,303 tweets and 1,808 followers. The selected anti-smoking group had 4,762
tweets and 3,325 followers.
Both groups were pressure groups representing the rights of smokers and
nonsmokers, respectively. The pro-smoking group formed in 1979 and campaigned
against tobacco control activities. The anti-smoking group formed in 1967. It had worked
to eliminate the harms caused by tobacco and to promote smoke-free regulations. Both
groups joined Twitter on February 2010.
The project used a Twitter API and a Python programing technique to crawl
timeline posts and other information in both groups’ Twitter accounts. Due to the rate
limits on Twitter API, the most recent 6,432 Tweets (3,216 Tweets for each group) were
obtained. The final data included both the Tweets and the number of times each Tweet
had been shared (retweets).
Data Cleaning
In order to examine message characteristics contributing to information sharing
behavior, approximately five percent of the most highly-disseminated Tweets and five
percent of non-disseminated Tweets were randomly selected from each Twitter account.
This yielded a total of 640 posts for analysis: 160 highly-disseminated and 160 non-
65
disseminated Tweets from the pro-smoking group and 160 highly-disseminated and 160
non-disseminated Tweets in the anti-smoking group.
Measurement
Information dissemination.
Dissemination was measured by the number of tweets that were shared with
others. Twitter has a function called ‘retweet,’ which allows users to redistribute a tweet
to their followers. Tweets with high number of retweets were regarded as highly-
disseminated Tweets. Tweets that were never retweeted were regarded as non-
disseminated Tweets.
Social support.
Cutrona and Suhr (1992)’s typology of social support was used to code the
support content of the Tweets. Five dimensions of social support were coded:
informational, emotional, esteem, network, and tangible supports. Two coders separately
evaluated the social support content of Tweets and decided which support best
corresponded to each Tweet. Coders focused on the nominal category of each social
support type in the Tweet, regardless of whether it was seeking or providing social
support. Coders made a note about Tweets that contained two or more types of social
support, but coded the most salient type for the analysis. Coding discrepancies were
discussed. If a consensus was reached, discrepancies were resolved; if a consensus was
not reached, they remained discrepant. Cohen’s kappa coefficient of agreement was .995,
suggesting a very high level of coding reliability. Per the hypotheses stated earlier, Study
II focused on two of the five categories of social support, informational and emotional
support.
66
Affective tone.
To measure the affective tone of Tweets, two researchers separately evaluated the
content of Tweets and decided if it was positive, neutral, or negative. Cohen’s kappa
was .950, suggesting a very high level of reliability.
Analysis
To test the hypotheses and research questions, chi-squared tests and Poisson
regression tests were employed.
Results
The average number of retweets was 57.38 in the pro-smoking group and 81.22 in
the anti-smoking group. In the pro-smoking group, Tweets about policy/social movement
were most frequently observed (N = 125, 39.06%). Tweets derogating anti-tobacco
groups (N = 81, 25.31%) and Tweets about analysis/research (N = 65, 20.31%) were
prevalent as well. Note that a Tweet may be counted multiple times if it contains two or
more types of content. In the anti-smoking group, the most frequently observed Tweets
were about analysis/research (N = 107, 33.44%). Tweets about policy/movement (N = 79,
24.69%), risks of smoking (N = 43, 13.44%), and encouraging smoking cessation (N =
43, 13.44%) were also prevalent (see Table 6).

67
Table 6. Content Analysis of Tweets in Pro-smoking and Anti-smoking Groups
Pro-smoking Group

Analysis/
Research
Policy/
Movement
Derogating
anti-
smoking
groups
Justifying Encouraging Other Pictures
Not
relevant
Total
Propagated
Tweets
45 64 48 7 15 10 40 5 160
Non-
propagated
Tweets
20 55 29 3 3 3 31 23 160
Total Tweets 65 125 81 10 18 55 71 28 320

Anti-smoking Group

Analysis/
Research
Policy/
Movement
Risks of
Smoking
Benefits
of
Quitting
Encouraging Other Pictures
Not
relevant
Total
Propagated
Tweets
39 53 36 9 19 7 19 4 160
Non-
propagated
Tweets
68 26 7 3 24 13 17 12 160
Total Tweets 107 79 43 12 43 20 36 16 320
Note. Counts by category are not mutually exclusive.

68
The number of Tweets related to social support was 406 (63.4%). Among them,
the number of Tweets containing informational support was 302 (74.38%), emotional
support 85 (20.94%), esteem support 23 (4.93%), network support 16 (3.94%), and
tangible support 6 (1.48%). As with content type, a Tweet may be counted multiple times
if it contains two or more types of support. Regarding the affective tone of Tweets, the
majority of Tweets were affectively neutral (58.6%). Tweets with positive (21.3%) and
negative (20.2%) valence appeared at similar rates.
A chi-squared test showed that Tweets related to social support were more
prominent among highly-disseminated posts than among non-disseminated posts (!
"
(1) =
70.09, p < .001) (see Figure 4).

Figure 4. Number of Tweets by social support and dissemination level
69
Table 7 shows that Tweets related to social support (N = 254, 39.7%) were
observed more than expected among highly-disseminated posts. On the other hand,
Tweets related to social support (N = 152, 23.8%) were observed less than expected
among non-disseminated Tweets. Thus, H4 was supported. In addition, a Poisson
regression model was used to examine the contributions of social support content in
explaining Tweet dissemination. The number of times each Tweet got repeatedly posted
on Twitter (frequency of retweets) was a dependent variable. The result showed that
social support was a significant predictor of Tweet dissemination frequency (Wald !
"
(df
=1) = 16.57, p < .001).

Table 7. Frequency of Social Support Relevance by Level of Dissemination

Social Support Total
Not Relevant Relevant
Disseminate
None
N
168 152 320
Expected N 117 203 320
% of Total 26.3% 23.8% 50.0%
High
N 66 254 320
Expected N 117 203 320
% of Total 10.3% 39.7% 50.0%
Total
N
234 406 640
Expected N 234 406 640
% of Total 36.6% 63.4% 100.0%

The same overall pattern was also manifest in each group, separately. Within the
pro-smoking group, Tweets relevant to social support were more prominent among
highly-disseminated posts than among non-disseminated posts (!
"
(1) = 17.69, p < .001).
Likewise, within the anti-smoking group, Tweets about social support were more
70
frequently observed among highly-disseminated posts than among non-disseminated
posts (!
"
(1) = 60.49, p < .001).
A chi-square test was performed to test whether users in the pro-smoking group
were more likely to disseminate Tweets relevant to emotional social support than users in
the anti-smoking group. As seen in Table 8, among highly-disseminated posts, Tweets
relevant to informational support were more prevalent than emotional support in both
groups. However, Tweets relevant to emotional support were observed more than
expected within the pro-smoking group, and less than expected within the anti-smoking
group (!
"
(1) = 21.54, p < .001).

Table 8. Observed and Expected Frequencies of Tweets by Social Support Type and
Group

Social Support Type Total
Informational Emotional
Group
Pro-smoking
N 64 47 111
Expected N 80.2 30.8 111.0
% of Total 26.1% 19.2% 45.3%
Anti-
smoking
N 113 21 134
Expected N 96.8 37.2 134.0
% of Total 46.1% 8.6% 54.7%
Total
N 177 68 245
Expected N 177.0 68.0 245.0
% of Total 72.2% 27.8% 100.0%

Figure 5 demonstrates that Tweets relevant to emotional support were more likely
to be retweeted in the pro-smoking group than the anti-smoking group. Thus, H5 was
partially supported.

71

Figure 5. Number of Tweets by social support type in pro-smoking and anti-smoking
groups
To test H6, the affective tone of highly-disseminated and non-disseminated
Tweets was compared. As seen in Table 9, Tweets with an affective tone were more
prominent among highly-disseminated posts than among non-disseminated posts (!
"
(1) =
79.35, p < .001). Each group showed the same pattern. Tweets having an affective tone
were more prominent among highly-disseminated Tweets than non-disseminated Tweets
within the pro-smoking group (!
"
(1) = 51.55, p < .001) and within the anti-smoking
group (!
"
(1) = 60.49, p < .001). A Poisson regression test also showed that Tweets with
72
an affective tone were more likely to be disseminated (Wald !
"
(df =1) = 15.67, p < .001)
in the smoking related groups.

Table 9. Observed and Expected Frequencies of Tweets by Affective Tone and
Dissemination Level

Affective Tone Total
Neutral Positive or Negative
Disseminate
Low
N 243 77 320
Expected N 187.5 132.5 320.0
% of Total 38.0% 12.0% 50.0%
High
N 132 188 320
Expected N 187.5 132.5 320.0
% of Total 20.6% 29.4% 50.0%
Total
N 375 265 640
Expected N 375.0 265.0 640.0
% of Total 58.6% 41.4% 100.0%

However, neither type of affective tone, positive or negative, was more prevalent
in highly-disseminated Tweets than in non-disseminated Tweets (!
"
(1) = .123, p > .05).
Figure 6 shows that messages having an affective tone were more prominent among
highly-disseminated Tweets than among non-disseminated Tweets. However, the
dissemination of Tweets having a negative tone was similar to that of Tweets having a
positive tone.
73

Figure 6. Number of Tweets by affective tone dissemination level

To explore RQ3, the affective tone of highly-disseminated Tweets was examined
in the pro-smoking and anti-smoking groups. As seen in Table 10, Tweets with a negative
tone were observed more often than expected in the pro-smoking group, while Tweets
with a positive tone were observed more often than expected in the anti-smoking group
(!
"
(1) = 27.57, p < .001). In other words, users in the pro-smoking group are more likely
to disseminate posts with a negative affective tone than users in the anti-smoking group
(see Figure 7).

74
Table 10. Observed and Expected Frequencies of Tweets by Affective Tone and Group

Affective Tone
Total Negative Positive
Group
Pro-
smoking
N 68 28 96
Expected N 50.0 46.0 96.0
% of Total 36.2% 14.9% 51.1%
Anti-
smoking
N 30 62 92
Expected N 48.0 44.0 92.0
% of Total 16.0% 33.0% 48.9%
Total
N 98 90 188
Expected N 98.0 90.0 188.0
% of Total 52.1% 47.9% 100.0%

Figure 7. Number of highly-disseminated Tweets by affective tone and group

75
Table 11 summarizes the chi-squared difference tests on social support (relevant
or not relevant) and dissemination (highly-disseminated or non-disseminated) and
affective tone (Tweets having an affective tone or Tweets having a neutral tone) and
dissemination.

Table 11. Chi-squared Difference Tests on Social Support, Affective Tone, and Tweets
Dissemination
Groups !
"
d.f. p
Social Support X
Tweets
Dissemination
All 70.09 1 < .001
Pro-smoking 17.69 1 < .001
Anti-smoking 60.49 1 < .001
Affective Tone X
Tweets
Dissemination
All 79.35 1 < .001
Pro-smoking 51.55 1 < .001
Anti-smoking 60.49 1 < .001

Table 12 summarizes the chi-squared difference tests on stance (pro-smoking or
anti-smoking) and social support type (Tweets having emotional support or Tweets
having informational support) and stance and affective tone type (Tweets with positive
affective tone or Tweets with negative affective tone).

Table 12. Chi-squared Difference Tests on Stance and Types of Social Support and
Affective Tone in Disseminated Tweets
!
"
d.f. p
Stance X
Social Support Type
21.54 1 < .001
Stance X
Affective Tone Type
27.57 1 < .001

76
Table 13 summarizes hypothesis test results.

Table 13. A Summary of the Results of Hypotheses Testing
Study I Results
H1
Individuals are more likely to form social ties with those
who share their stance towards risky health behavior
than those who have different stance on social media.
Supported
H2
Individuals who promote risky health behavior will form
more cohesive networks than those who promote healthy
behavior on social media.
Partially
supported
H3
Social ties to individuals holding the opposite stance will
be more frequently observed among individuals who
promote healthy behavior than among those who
promote risky health behavior on social media.
Not Supported
RQ1
Are source characteristics (i.e. individuals, non-profit or
academic groups, media, and commercial entities) in
networks promoting risky health behavior different from
those in networks promoting healthy behavior?
Yes
RQ2
What are the characteristics of individuals who have
social ties with those holding an opposite stance?
Pro-smoking
activists/Anti-
smoking NGO
Study II
H4
Information with social support will be more prominent
among highly disseminated posts than among non-
disseminated posts both for individuals who promote
risky health behaviors and those who promote healthy
behaviors.
Supported
H5
Information relevant to emotional support are more
likely to be disseminated among individuals who
promote risky health behavior than among those who
promote healthy behavior.
Partially
supported
H6
Posts with an affective tone will be more prominent
among highly disseminated posts than among non-
disseminated posts both for individuals who promote
risky health behavior and those who promote healthy
behavior.
Supported
RQ3
Which affective tones (positive or negative) contribute
more to post dissemination among individuals who
promote risky health behaviors compared to those who
promote healthy behaviors?
Negative
77
CHAPTER 6: DISCUSSION AND CONCLUSION
This dissertation project has two fundamental purposes. The first is gaining a
better understanding of how individuals who promote risky health behavior form social
relationships, exchange information, and share social support through social networks on
social media (Study I). The second is to understand what characteristics of highly-
disseminated posts influence information sharing among individuals who promote risky
health behavior (Study II).
Study I
The first study examined the social networks of pro-smoking users and anti-
smoking users on Twitter in order to understand the mechanisms of relational formation
and influence among advocates of risky health behavior. Specifically, this study looked at
the effect of attitudinal homophily about smoking on network cohesiveness and
relationship exclusiveness. I hypothesized that individuals were more likely to form
relationships with those who shared their stance towards smoking behavior (H1). It was
predicted that pro-smoking users were more likely to follow other pro-smoking users, and
that anti-smoking users were more likely to follow other anti-smoking users. It was also
predicted that, compared to anti-smoking users, pro-smoking users would be more
closely connected to each other and have a more cohesive network (H2). Regarding
relationship exclusiveness, I predicted that social ties between users of the opposite
stance would be observed more frequently among anti-smoking users than among pro-
smoking users. In other words, pro-smoking users would have more exclusive social
relationships than anti-smoking users (H3). In addition, source characteristics including
account types and industry/occupation of users in pro-smoking networks and anti-
78
smoking networks were explored (RQ1). Lastly, users connected to those holding an
opposite stance were explored. Specifically, characteristics of pro-smoking users who
follow anti-smoking users and those of anti-smoking users who follow pro-smoking users
were investigated (RQ2).
This study found that attitudinal homophily about smoking was as a significant
predictor of relationship formation. Individuals tended to create social ties with users who
shared their same stance. This result shows that a person’s taking an explicit stance
towards smoking behavior has substantial predictive power regarding his or her
connections on social media. For one, it provides empirical support for homophily on
social media. Several studies have observed that social relationships are, in general,
homogeneous with regard to many sociographic, behavioral, and interpersonal
characteristics (McPherson, Smith-Lovin, & Cook, 2001). Social ties formed between
similar individuals limit information variety, attitude formation, and communicative
actions. The principle of homophily, however, has not been consistently observed in
social media research, particularly with respect to attitudes. One study, Aiello et al.,
(2012) found that users who shared similar tags, tagged items, and groups had a high
possibility of becoming friends on social media. However a study by Bisgin, Agarwal, &
Xu (2012) found that similar interests (e.g. similar tags and tagged items) did not
significantly predict ties on social media. The current finding adds evidence to the
literature confirming the principle of homophily on social media. Individuals had a
greater chance of forming a tie with someone who shares their stance towards risky
health behavior.
79
Several network-level measures showed that pro-smoking users formed more
cohesive networks than anti-smoking users. First, the pro-smoking network had a higher
density than the anti-smoking network. That is, within their social network, pro-smoking
users were more likely to know each other than anti-smoking users. It should be noted,
however, that, the size of the pro-smoking network was smaller than the anti-smoking
network. Given the formula to calculate network density, networks with smaller size are
more likely to have higher density than larger networks. Thus, social cohesion of the two
networks should be interpreted using density along with other indicators. The pro-
smoking network showed higher reciprocity and transitivity than the anti-smoking
network. ‘Higher reciprocity’ means that mutual ties between two individuals were more
frequently observed in the pro-smoking network than in the anti-smoking network. In
other words, pro-smoking users had stronger ties than anti-smoking users had. This result
is in line with several studies that have showed a positive association between reciprocity
and the likelihood of people engaging in risky health behavior together (Valente &
Vlahov, 2001). As reciprocity implies a higher degree of trust, individuals are more likely
to engage in risky health behavior when they share a strong tie with someone, rather than
a weak one. Pro-smoking users may build stronger ties with other pro-smoking users to
compensate for a lack of offline support, and the interactions between them could
promote smoking behavior. The high proportion of triads in the pro-smoking network is
another indicator of a cohesive network. It means that pro-smoking users are more likely
to have homogenous opinions and facilitate pro-smoking information/behavior within
their groups, as well as inhibit anti-smoking information/behavior entering from outside
of group. Lastly, average path length and network diameter in the pro-smoking network
80
were shorter than in the anti-smoking network. This implies that pro-smoking users can
more easily reach other pro-smoking users within their networks, and therefore, pro-
smoking users are more closely connected to each other than are anti-smoking users.
It is important to note that network-level differences were only observed between
285 pro-smoking and anti-smoking users who explicitly expressed their stance. When
analyzing networks comprised of all 986 users, the two networks of users who were
followed by the pro-smoking or anti-smoking seed accounts exhibited similar measures
of cohesiveness. One possible explanation is that many users in these networks might not
be ego-involved in pro-smoking or anti-smoking issues. They may be followed by pro- or
anti-smoking seed accounts because of the seed accounts’ other interests. Those social
ties not formed on the basis of attitudes and interests about smoking behavior are not
likely to have been motivated out a desire for strong connection. Another possible
explanation is that users who explicitly describe their stance towards smoking behavior
are enthusiastic advocates and would have strong motivations to find similar peers. This
suggests that the cohesive pro-smoking network might not have a significant impact on
inactive pro-smoking users. Future research can further examine the level of a member’s
activity in a movement and the effect of this on relationship formation.
The study also found the anti-smoking network to be more exclusive than the pro-
smoking network. Specifically, more pro-smoking users followed anti-smoking users
than the reverse. Most anti-smoking users only followed other anti-smoking users. This
result contradicts what was predicted in the hypothesis. Yom-Tov, Fernandez-Luque,
Weber, & Crain (2012) found that pro-health behavior groups try to reach pro-risky
behavior groups thorough the use of similar tags so pro-risky behavior groups could be
81
exposed to information promoting healthy behavior. However, this study showed that
anti-smoking users had more exclusive social ties: they were less likely to reach pro-
smoking users on Twitter. A possible explanation for this is that pro-smoking users might
pay more attention to what anti-smoking users say, as the perceived status and legitimacy
of the anti-smoking movement are much higher than that of the pro-smoking movement.
Thus, pro-smoking users may make frequent reference to the anti-smoking discourse,
whereas anti-smoking users largely ignore pro-smoking users. Pro-smoking users can
benefit from following anti-smoking users and receiving information about smoking
regulations, the progress of smoking ban policies or actions taken against tobacco
companies. The efforts of pro-smoking users to avoid Twitter regulations could be
another possible explanation. As social media started being blamed for promoting risky
health behavior, social media administrators from Twitter, Tumblr, Instagram, and others
started to regulate search terms related to risky health behavior. Despite these regulations,
such content continued to thrive as users created clever jargon designed to avoid
censorship (Park, Sun, & McLaughlin, 2017). For example, pro-smoking users might use
jargon such as ‘juice’ to refer to the liquid in e-cigarettes, making it difficult for anti-
smoking users to find pro-smoking groups
Interestingly, the post hoc analysis showed that homophily in relationship
exclusiveness predicted social ties on Twitter, although the effect size was not large. Pro-
smoking users who followed anti-smoking users were more likely to have social ties with
other pro-smoking users who also followed anti-smoking users. Similarly, pro-smoking
users who did not follow anti-smoking users were more likely to have social ties with
pro-smoking users who also did not follow anti-smoking users.
82
The result with respect to account types and industry/occupation of users showed
that pro-smoking users were mostly individuals whereas anti-smoking users were mainly
non-profit or academic groups. The analysis also suggested that anti-smoking users
tended to form social ties with those of the same categories of account type, but pro-
smoking users did not. A possible explanation for this is that there might be offline
connections that contribute to online social ties among anti-smoking academic groups,
activists, and smoking cessation-related commercial entities. On the other hand, pro-
smoking users who engage in stigmatized health behavior may have fewer chances to
meet similar others offline. Thus, their online relationships may be solely based on online
interactions and such source characteristics do not play a significant role in online
network formation.
Finally, this study showed that pro-smoking users who followed anti-smoking
users were mostly individuals, not groups. Most of them were from pressure groups and
many of them were individual advocates. On the other hand, anti-smoking users who
followed pro-smoking users were individuals or non-profit or academic groups. This
included pressure groups and academic/medical groups. It is possible that anti-smoking
groups have more interest in what pro-smoking users say because these users are
involved in designing health campaigns and crafting smoking-related policy. This result
may be related to H3. Anti-smoking users who have better access to pro-smoking users
are groups that have more information about the pro-smoking movement, not individual
pro-smoking advocates. On the other hand, pro-smoking individuals have motivations to
follow anti-smoking users in order to gain information about smoking bans.
83
Overall, the findings have implications for designing health promotion campaigns
on social media for people engaging in risky health behavior. The first step should be to
examine explicit stances toward risky health behavior and identify users who promote
such behavior. Stance homophily plays an important role in network formation. The
examination of explicit stances would help identify social networks of targeted users on
social media. As individuals who promote risky health behaviors are more likely to
facilitate within-group interactions and inhibit outside ideas, social media campaigns
must be prepared to overcome this resistance in order to reach their intended targets.
Also, identifying source characteristics such as account types or industry/occupation of
users would help to disseminate health campaign messages. As account types of
individual users are more likely to have connections with others who have different
stances, health campaigns should be designed to target individual users through these
connections. Bridging connections may play an important role in facilitating discussion
about the campaigns between users.
Study II
The second study examined the information sharing behavior of pro-smoking
users and anti-smoking users in order to understand the informational influence
mechanisms among people engaging in risky health behavior. Specifically, this study
looked at the relationship between Tweet dissemination and Tweet characteristics,
namely social support and affective tone. It was predicted that both pro-smoking and anti-
smoking users would share Tweets with social support more than Tweets without social
support (H4). Also, Tweets with emotional support were expected to be shared more by
pro-smoking users than by anti-smoking users (H5). It was also hypothesized that both
84
pro-smoking and anti-smoking users would share Tweets having an affective tone more
than Tweets having a neutral tone (H6). Lastly, it explored the quality of the affective
tone in Tweets that were shared more by pro-smoking users than by anti-smoking users.
Affective tone could be positive or negative. This study examined which affective tone
would contribute more to information sharing by pro-smoking users (RQ3).
The study found that content containing social support was more frequently
observed in highly disseminated Tweets than non-disseminated Tweets. In addition,
social support was a significant predictor of Tweet propagation frequency. In line with
studies on social media and health (Chung, 2013; M. McLaughlin et al., 2012), this result
can be interpreted to mean that individuals who promoted in risky health behaviors were
more motivated to exchange social support, and this influenced their information sharing
behavior on social media. This insight may be useful when designing health messages
intended for social media users. Health messages containing social support are more
likely to be disseminated and receive more exposure than messages not containing social
support.
This study also found that Tweets expressing emotional support were observed
more than expected among the pro-smoking group and less than expected among the anti-
smoking group. This can be interpreted to mean that pro-smoking users paid more
attention to Tweets expressing emotional support than anti-smoking users did. This result
reveals a significant difference in the informational influence mechanisms of pro-
smoking users from anti-smoking users. When information containing emotional support
is propagated, pro-smoking users will more likely be exposed to it. It may well be that
this content provides emotional support which, as stigmatized individuals, they do not get
85
enough of from the offline world. In this way, individuals engaging in risky health
behavior encouraged each other to maintain and promote their behavior.
Affective tone was also found to be an important factor in Tweet dissemination
both in pro- and anti-smoking groups. Affective tone was observed more than expected
among highly-disseminated Tweets and less than expected among non-disseminated
Tweets. It was also found that Tweets having an affective tone were likely to be
retweeted with higher frequency. This result confirms existing literature claiming that
affective tone provides significant clues to understanding information sharing behavior
among social media users who are interested in risky health behavior (McLaughlin et al.,
2016; Rime, 2009; Shoemaker, 1996). As with other types of information, information
about risky health behavior is more likely to be disseminated when it has an affective
tone.
Interestingly, in the pro-smoking group, Tweets having a negative tone were
shared more than those having a positive tone. On the other hand, in the anti-smoking
group, highly disseminated Tweets having a positive tone were observed more than those
having a negative tone. There were competing hypotheses based on the motivations of
individuals who experienced stigma: 1) They would share information with a positive
tone in order to compensate for a lack of offline social support; 2) They would share
information with a negative tone when defending their self-concept from outside attacks.
The findings imply that in this case, the latter is a stronger motivation for information
sharing. The content analysis of Tweets in the pro-smoking group also supports this result.
Tweets derogating anti-smoking groups were the second largest category among
86
disseminated Tweets. Future studies may wish to analyze Tweet content and affect
together in order to uncover possible interactions between the two..
Overall, Study II contributes to the literature on information sharing behavior
among people who promote risky health behavior as it is the first line of research to
differentiate mechanisms of how different forms of social support and affective tone
influence dissemination of risky health-related information on social media. In particular,
for individuals engaging in risky health behavior, information containing emotional
support or a negative affective tone predicted higher rates of post dissemination. This
indicates that user information sharing behavior is motivated by a need to compensate for
a lack of external support and to defend their self-concept from external attacks. By
exchanging information and support with risk-taking peers, stigmatized individuals can
protect their self-esteem and self-efficacy, and defend themselves against health
campaigns or other negative social influences.
These findings have implications for designing social media interventions aimed
at advocates of risky health behavior. First, researchers or health practitioners must
incorporate social support content when designing health messages. Health promotion
messages with social support may naturally attract attention from social media users who
engage in risky health behavior. In particular, emotional support should be considered in
addition to informational support as a means of boosting message dissemination. Also,
social media intervention must convey an affective tone as this easily captures the
attention of social media users. More specifically, a negative affective tone such as a fear
appeal (Witte, 1994) should be considered. A meta-analysis shows that fear appeal is
effective when combined with self-efficacy (Peters, Ruiter, & Kok, 2013). Thus, social
87
media interventions may be more effective if their designs combine emotional support
with a negative affective tone.
Limitations and Future Research
The present project was limited in a number of ways. One major limitation is the
variation in users’ health status. This study compared individuals who promote risky
health behavior and those who promote healthy behavior. Although the current
dissertation focused on one’s stance towards health behavior, an individual’s health status
could play an important role in relationship formation and information sharing. For
example, in case of individuals who promote healthy behavior, a person who is healthy
and promotes healthy eating may have different motivations for finding peers than
another person who has lung cancer and promotes smoking cessation. I assume that the
healthy person would experience fewer negative social reactions than the other person
with lung cancer. The person with lung cancer may even experience stigmatization that
healthy people would not do. Future research should consider health status in order to
better understand individuals who promote risky health behavior compared to others who
promote healthy behavior. In the procedure of examinations of one’s health condition,
additional methods, such as content analysis or survey methods, would be needed.
A second limitation is the basic assumption of the association between explicit
stance and actual behaviors. In these studies, pro-smoking users were assumed to be
engaging in smoking behaviors, whereas anti-smoking users were assumed to be
engaging in smoking cessation or healthy behavior. The association may not be always
true. For example, non-smokers may promote smoking behaviors out of their
commitment to smokers’ rights. Inconsistency between behavior and stance is an
88
interesting area for further examination. Nevertheless, I would argue that this assumption
should not have biased the findings in this project, as the hypotheses were drawn from a
social identity perspective. Pro-smoking users may be stigmatized and fail to receive
offline social support for their ideas, regardless of whether they actually engage in risky
health behavior. Offline, they have fewer chances of meeting those who share their
opinions and therefore may be equally motivated to use social media to exchange
information and seek support as actual smokers. Thus, from the social identity
perspective, the mechanisms of relationship formation and information sharing may
apply equally for people who only promote risky health behavior as for those who both
promote and engage.
A third limitation of the two studies relates to the data sampled. As Twitter is a
popular place for bots, I had to check the offline activities of user accounts in order to
confirm that they were authentic. Although I selected representative pro-smoking and
anti-smoking groups in terms of offline activities and Twitter popularity for the initial
seeds, there may have been other social media users or online groups that did not have
any connection with offline groups, but still impact pro-smoking users. Although I
included online media organizations in the seeds to capture them, the media groups were
also selected from among following networks of the offline groups. Future research
should identify users who are only influential online and combine their influence in pro-
risky health behavior networks. The groups in Study II have the same limitation as they
were also chosen from the initial seed accounts.
Additional limitations of this dissertation are the sample sizes and sampling bias
in Study I and Study II. The data cleaning process excluded a portion of network
89
members who did not explicitly express their stance towards smoking behavior. As the
purpose of this dissertation research was to examine the relationship formation and
information sharing patterns of pro-smoking and anti-smoking users, users with an
unknown stance could not be included in the final analysis. As a result, the final network
sizes of each stance were less than 200. According to Valente (2012), in a network of less
than two hundred nodes, density can be affected by only a few links. Other network-level
measures may also be considerably influenced by the addition or removal of a few links.
Thus, future research on these processes should collect larger sample sizes. In Study II,
although the number of sampled Tweets was sufficient for statistical analyses, they were
only sampled from the Twitter accounts of two groups. These two accounts, one pro-
smoking and one anti-smoking, were in terms of their popularity on the Twitter platform;
however, sampling from more groups and a greater variety of group types would help to
generalize the results.
Another limitation is that Study I was unable to test whether the difference of
network cohesiveness is meaningful. Most indicators of network cohesiveness showed
that the pro-smoking network was more cohesive than the anti-smoking network.
However, this result is descriptive and it could not be statistically tested given the
statistical incomparability of the two networks. Replicating these analyses with different
samples would help to generalize findings on the patterns of network cohesiveness
among people who engage in risky health behavior relative to those who engage in
healthy behavior.
In spite of these limitations, this study suggests a number of valuable avenues for
future research. This project compared advocacy of risky and healthy behavior on Twitter.
90
Twitter is a useful place for capture network structures as most social ties are visible. For
this same reason, however, risky health behavior advocates may prefer other social media
platforms that ensure more closed networks. Thus, future research should replicate this
study in order to see how relationship formation and information sharing behavior vary
across different types of social media platforms.
In addition, affective tone plays a significant role in sharing information about
risky health behavior. The properties of tone affect could be elaborated by considering
the level of arousal and the strength of the tone. Given that sharing information requires
action, information resulting in a higher level of physiological arousal or activation (Chen
& Berger, 2012) may be shared more often. By contrast, a message that results in lower
arousal, deactivation, or inaction, may be shared with lower frequency. In the case of
risky health behavior, a negative affective tone resulting in high arousal, such as anger
rather than sadness, may increase information dissemination. In addition, on social media
extreme messages tend to be more popular (Heath, 1996). According to Heath (1996),
individuals are more willing to pass along information that matches the emotional
valence of the conversational topic. For example, individuals share exaggeratedly bad
information when the topic is emotionally negative, but share exaggeratedly good news
when the topic is emotionally positive. Examination of the strength of affective tone and
information content may be useful for understanding the dissemination of information
promoting risky health behavior.
Sharing content promoting risky health behavior may be more problematic than
sharing information about illegal behaviors such as black market activity (Hanson et al.,
2013; Mackey & Liang, 2013) because no laws can restrict individual users’ information
91
sharing behavior. Some harmful content or search terms have been banned by social
media administrators, but in spite of the efforts to regulate such content, information
promoting risky health behavior continues to thrive. Users can disseminate information
that encourages smoking, alcohol consumption, or self-harm without any restrictions. For
these reasons, social media research on risky health behavior deserves more attention and
discussion.
Conclusion
This dissertation examined relationship formation and information sharing on
risky health behavior-related social networks. It investigated the theoretical mechanisms
that drive the formation of online social networks among individuals who promote risky
health behavior, and analyzed those network structures. Drawing on a social identity
framework, the first study found risk-promoting networks to manifest greater stance
homophily, higher levels of network cohesion, and less social exclusivity than health-
promoting networks. This study also found that advocates of risky health behavior were
mostly individual users whereas advocates against risky health behavior were primarily
non-profits or academic groups. Relationships in health-promoting networks manifested
greater homophily by account types, whether individuals, non-profit or academic groups,
media, and commercial entities. Members of risk-promoting networks who had social ties
with users of the opposite stance were mostly individuals and were connected to pressure
groups. Members of health-promoting networks who had social ties with users of the
opposite stance were mostly individuals or non-profit/academic groups who were related
to pressure groups. The second study drew from literature on social support, emotion, and
information sharing. It found that information with social support and affective tone
92
contributed to the dissemination of information in both risk-promoting and health-
promoting networks. Information conveying emotional support and a negative affective
tone significantly enhanced dissemination of information among members of risk-
promoting networks.

93
REFERENCES
Ahn, Y.-Y., Han, S., Kwak, H., Moon, S., & Jeong, H. (2007). Analysis of topological
characteristics of huge online social networking services (p. 835). ACM Press.
https://doi.org/10.1145/1242572.1242685
Aiello, L. M., Barrat, A., Schifanella, R., Cattuto, C., Markines, B., & Menczer, F.
(2012). Friendship prediction and homophily in social media. ACM Transactions
on the Web, 6(2), 1–33. https://doi.org/10.1145/2180861.2180866
Albrecht, T. L., Goldsmith, D. J., & Thompson, T. (2003). Social support, social
networks, and health. Handbook of Health Communication, 263–284.
Alexandrov, A., Lilly, B., & Babakus, E. (2013). The effects of social- and self-motives
on the intentions to share positive and negative word of mouth. Journal of the
Academy of Marketing Science, 41(5), 531–546. https://doi.org/10.1007/s11747-
012-0323-4
Arseniev-Koehler, A., Lee, H., McCormick, T., & Moreno, M. A. (2016). #Proana: Pro-
Eating Disorder Socialization on Twitter. Journal of Adolescent Health, 58(6),
659–664. https://doi.org/10.1016/j.jadohealth.2016.02.012
Bandura, A. (1995). Exercise of personal and collective efficacy in changing societies. In
A. Bandura (Ed.), Self-efficacy in changing societies. Cambridge University
Press.
Bandura, A. (2001). Social Cognitive Theory: An Agentic Perspective. Annual Review of
Psychology, 52(1), 1–26. https://doi.org/10.1146/annurev.psych.52.1.1
94
Barak, A., Boniel-Nissim, M., & Suler, J. (2008). Fostering empowerment in online
support groups. Computers in Human Behavior, 24(5), 1867–1883.
https://doi.org/10.1016/j.chb.2008.02.004
Barrat, A., Barthelemy, M., Pastor-Satorras, R., & Vespignani, A. (2004). The
architecture of complex weighted networks. Proceedings of the National Academy
of Sciences, 101(11), 3747–3752. https://doi.org/10.1073/pnas.0400087101
Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger
than good. Review of General Psychology, 5(4), 323–370.
https://doi.org/10.1037/1089-2680.5.4.323
Baumeister, R. F., Dale, K., & Sommer, K. L. (1998). Freudian Defense Mechanisms and
Empirical Findings in Modern SocialPsychology: Reaction Formation, Projection,
Displacement, Undoing, Isolation, Sublimation, and Denial. Journal of
Personality, 66(6), 1081–1124. https://doi.org/10.1111/1467-6494.00043
Bayer, M., Sommer, W., & Schacht, A. (2012). Font Size Matters—Emotion and
Attention in Cortical Responses to Written Words. PLoS ONE, 7(5), e36042.
https://doi.org/10.1371/journal.pone.0036042
Bell, V. (2007). Online information, extreme communities and internet therapy: Is the
internet good for our mental health? Journal of Mental Health, 16(4), 445–457.
https://doi.org/10.1080/09638230701482378
Berger, J. (2011). Arousal Increases Social Transmission of Information. Psychological
Science, 22(7), 891–893. https://doi.org/10.1177/0956797611413294
Berger, J., & Milkman, K. L. (2012). What Makes Online Content Viral? Journal of
Marketing Research, 49(2), 192–205. https://doi.org/10.1509/jmr.10.0353
95
Berger, J., & Rand, L. (2008). Shifting Signals to Help Health: Using Identity Signaling
to Reduce Risky Health Behaviors. Journal of Consumer Research, 35(3), 509–
518. https://doi.org/10.1086/587632
Bert, F., Gualano, M. R., Camussi, E., & Siliquini, R. (2016). Risks and Threats of Social
Media Websites: Twitter and the Proana Movement. Cyberpsychology, Behavior,
and Social Networking, 19(4), 233–238. https://doi.org/10.1089/cyber.2015.0553
Biddle, L., Donovan, J., Hawton, K., Kapur, N., & Gunnell, D. (2008). Suicide and the
internet. BMJ, 336(7648), 800–802.
https://doi.org/10.1136/bmj.39525.442674.AD
Bisgin, H., Agarwal, N., & Xu, X. (2012). A study of homophily on social media. World
Wide Web, 15(2), 213–232. https://doi.org/10.1007/s11280-011-0143-3
Borawski, E. A., Ievers-Landis, C. E., Lovegreen, L. D., & Trapl, E. S. (2003). Parental
monitoring, negotiated unsupervised time, and parental trust: the role of perceived
parenting practices in adolescent health risk behaviors. Journal of Adolescent
Health, 33(2), 60–70. https://doi.org/10.1016/S1054-139X(03)00100-9
Bradley, M. M. (2014). Emotional memory: A dimensional analysis. In S. H. M. Van
Goozen, N. E. Van de Poll, & J. A. Sergeant (Eds.), Emotions Essays on Emotion
Theory. (pp. 97–134). Hoboken: Taylor and Francis. Retrieved from
http://public.eblib.com/choice/publicfullrecord.aspx?p=1639302
Braithwaite, D. O., Waldron, V. R., & Finn, J. (1999). Communication of Social Support
in Computer-Mediated Groups for People With Disabilities. Health
Communication, 11(2), 123–151. https://doi.org/10.1207/s15327027hc1102_2
96
Branscombe, N. R., Schmitt, M. T., & Harvey, R. D. (1999). Perceiving pervasive
discrimination among African Americans: Implications for group identification
and well-being. Journal of Personality and Social Psychology, 77(1), 135–149.
https://doi.org/10.1037/0022-3514.77.1.135
Brashers, D. E., Neidig, J. L., & Goldsmith, D. J. (2004). Social Support and the
Management of Uncertainty for People Living With HIV or AIDS. Health
Communication, 16(3), 305–331. https://doi.org/10.1207/S15327027HC1603_3
Brown, R. (2000). Social identity theory: past achievements, current problems and future
challenges. European Journal of Social Psychology, 30(6), 745–778.
https://doi.org/10.1002/1099-0992(200011/12)30:6<745::AID-EJSP24>3.0.CO;2-
O
Bull, S. S., Levine, D. K., Black, S. R., Schmiege, S. J., & Santelli, J. (2012). Social
Media?Delivered Sexual Health Intervention. American Journal of Preventive
Medicine, 43(5), 467–474. https://doi.org/10.1016/j.amepre.2012.07.022
Chancellor, S., Mitra, T., & De Choudhury, M. (2016). Recovery Amid Pro-Anorexia:
Analysis of Recovery in Social Media (pp. 2111–2123). ACM Press.
https://doi.org/10.1145/2858036.2858246
Chen, Z., & Berger, J. A. (2012). When, Why, and How Controversy Causes
Conversation (SSRN Scholarly Paper No. ID 2050275). Rochester, NY: Social
Science Research Network. Retrieved from
http://papers.ssrn.com/abstract=2050275
97
Chhabra, N., & Bryant, S. M. (2016). Snapchat Toxicology: Social Media and Suicide.
Annals of Emergency Medicine, 68(4), 527.
https://doi.org/10.1016/j.annemergmed.2016.05.031
Chu, Z., Gianvecchio, S., Wang, H., & Jajodia, S. (2012). Detecting Automation of
Twitter Accounts: Are You a Human, Bot, or Cyborg? IEEE Transactions on
Dependable and Secure Computing, 9(6), 811–824.
https://doi.org/10.1109/TDSC.2012.75
Chung, J. E. (2013). Social interaction in online support groups: Preference for online
social interaction over offline social interaction. Computers in Human Behavior,
29(4), 1408–1414. https://doi.org/10.1016/j.chb.2013.01.019
Chung, J. E. (2016). A Smoking Cessation Campaign on Twitter: Understanding the Use
of Twitter and Identifying Major Players in a Health Campaign. Journal of Health
Communication, 21(5), 517–526. https://doi.org/10.1080/10810730.2015.1103332
Cohen, S., Underwood, L., & Gottlieb, B. H. (2000). Social Support Measurement and
Intervention: A Guide for Health and Social Scientists. Oxford University Press.
Cohen, S., & Wills, T. A. (1985). Stress, social support, and the buffering hypothesis.
Psychological Bulletin, 98(2), 310–357. https://doi.org/10.1037/0033-
2909.98.2.310
Csardi, G., & Nepusz, T. (2006). The igraph software package for complex network
research. InterJournal, Complex Systems, 1695.
Cutrona, C. E., & Suhr, J. A. (1992). Controllability of Stressful Events and Satisfaction
With Spouse Support Behaviors. Communication Research, 19(2), 154–174.
https://doi.org/10.1177/009365092019002002
98
Davies, P., & Lipsey, Z. (2003). Ana’s gone surfing. Psychologist, 16(8), 424–425.
Dunlop, S. M., More, E., & Romer, D. (2011). Where do youth learn about suicides on
the Internet, and what influence does this have on suicidal ideation?: Influence of
the Internet on suicidal ideation. Journal of Child Psychology and Psychiatry,
52(10), 1073–1080. https://doi.org/10.1111/j.1469-7610.2011.02416.x
Ellison, N. B., & boyd, d. (2013). Sociality through Social Network Sites. In W. H.
Dutton (Ed.), The Oxford Handbook of Internet Studies (pp. 151–172). Oxford:
Oxford University Press.
Ferrara, E., Varol, O., Davis, C., Menczer, F., & Flammini, A. (2016). The rise of social
bots. Communications of the ACM, 59(7), 96–104.
https://doi.org/10.1145/2818717
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2),
117–140.
Forsyth, S. R., & Malone, R. E. (2010). “I’ll be your cigarette—Light me up and get on
with it”: Examining smoking imagery on YouTube. Nicotine & Tobacco
Research, 12(8), 810–816. https://doi.org/10.1093/ntr/ntq101
Fox, S., & Duggan, M. (2013). Health online 2013. Pew Research Internet Project.
Retrieved from http://www.pewinternet.org/2013/01/15/health-online-2013/
Freud, S. (1961). Instincts and their vicissitudes. In J. Strachey (Ed.), The standard
edition of the complete works of Sigmund Freud (Vol. 14, pp. 111–142). London:
Hogarth Press.
99
Gerbner, G., Gross, L., Morgan, M., Signorielli, N., & Shanahan, J. (2002). Growing up
with television: Cultivation processes. In A. Bandura, J. Bryant, & D. Zillman
(Eds.), Media effects: Advances in theory and research.
Glanz, K., Rimer, B. K., & Viswanath, K. (2008). Health behavior and health education:
theory, research, and practice. John Wiley & Sons.
Gochman, D. S. (1998). Health Behavior. In Health behavior (pp. 3–17). Springer.
Retrieved from http://link.springer.com/chapter/10.1007/978-1-4899-0833-9_1
Goffman, E. (1986). Stigma: Notes on the Management of Spoiled Identity (Reissue
edition). New York: Touchstone.
Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology,
78(6), 1360–1380.
Greenwood, S., Perrin, A., & Duggan, M. (2016). Social media update 2016. Pew
Research Internet Project. Retrieved from
http://www.pewinternet.org/2016/11/11/social-media-update-2016/
Hanson, C. L., Cannon, B., Burton, S., & Giraud-Carrier, C. (2013). An Exploration of
Social Circles and Prescription Drug Abuse Through Twitter. Journal of Medical
Internet Research, 15(9), e189. https://doi.org/10.2196/jmir.2741
Harris, J. K., Snider, D., & Mueller, N. (2013). Social Media Adoption in Health
Departments Nationwide. College of Public Health, University of Kentucky.
https://doi.org/10.13023/FPHSSR.0201.05
Heath, C. (1996). Do People Prefer to Pass Along Good or Bad News? Valence and
Relevance of News as Predictors of Transmission Propensity. Organizational
100
Behavior and Human Decision Processes, 68(2), 79–94.
https://doi.org/10.1006/obhd.1996.0091
Herring, S. C. (1996). Two variants of an electronic message schema. In S. C. Herring
(Ed.), Computer-mediated communication: Linguistic, social, and cross-cultural
perspectives (Vol. 39, pp. 81–106). Philadelphia: John Benjamins Publishing.
Hether, H. J., Murphy, S. T., & Valente, T. W. (2014). It’s Better to Give Than to
Receive: The Role of Social Support, Trust, and Participation on Health-Related
Social Networking Sites. Journal of Health Communication, 19(12), 1424–1439.
https://doi.org/10.1080/10810730.2014.894596
Hether, H. J., Murphy, S. T., & Valente, T. W. (2016). A social network analysis of
supportive interactions on prenatal sites. Digital Health, 2(0).
https://doi.org/10.1177/2055207616628700
Hiltz, S. R., & Wellman, B. (1997). Asynchronous Learning Networks As a Virtual
Classroom. Communications of the ACM, 40(9), 44–49.
https://doi.org/10.1145/260750.260764
Hogg, M. A., Abrams, D., Otten, S., & Hinkle, S. (2004). The social identity perspective
intergroup relations, self-conception, and small groups. Small Group Research,
35(3), 246–276. https://doi.org/10.1177/1046496404263424
Hogg, M. A., & Blaylock, D. L. (2011). Extremism and the Psychology of Uncertainty
(Vol. 8). John Wiley & Sons.
Hogg, M. A., & Reid, S. A. (2006). Social Identity, Self-Categorization, and the
Communication of Group Norms. Communication Theory, 16(1), 7–30.
https://doi.org/10.1111/j.1468-2885.2006.00003.x
101
Hoyt, L. T., Chase-Lansdale, P. L., McDade, T. W., & Adam, E. K. (2012). Positive
Youth, Healthy Adults: Does Positive Well-being in Adolescence Predict Better
Perceived Health and Fewer Risky Health Behaviors in Young Adulthood?
Journal of Adolescent Health, 50(1), 66–73.
https://doi.org/10.1016/j.jadohealth.2011.05.002
Huang, G. C., Unger, J. B., Soto, D., Fujimoto, K., Pentz, M. A., Jordan-Marsh, M., &
Valente, T. W. (2014). Peer Influences: The Impact of Online and Offline
Friendship Networks on Adolescent Smoking and Alcohol Use. Journal of
Adolescent Health, 54(5), 508–514.
https://doi.org/10.1016/j.jadohealth.2013.07.001
Huffaker, D. (2010). Dimensions of Leadership and Social Influence in Online
Communities. Human Communication Research, 36(4), 593–617.
https://doi.org/10.1111/j.1468-2958.2010.01390.x
Jonas, S., Goldsteen, R. L., Goldsteen, K., & Jonas, S. (2013). Jonas’ introduction to the
U.S. health care system (7th ed). New York: Springer Pub. Co.
Joseph, A. J., Tandon, N., Yang, L. H., Duckworth, K., Torous, J., Seidman, L. J., &
Keshavan, M. S. (2015). #Schizophrenia: Use and misuse on Twitter.
Schizophrenia Research, 165(2–3), 111–115.
https://doi.org/10.1016/j.schres.2015.04.009
Kasl, S. V., & Cobb, S. (1966). Health Behavior, Illness Behavior and Sick Role
behavior: I. Health and Illness Behavior. Archives of Environmental Health: An
International Journal, 12(2), 246–266.
https://doi.org/10.1080/00039896.1966.10664365
102
Keelan, J., Pavri-Garcia, V., Tomlinson, G., & Wilson, K. (2007). YouTube as a Source
of Information on Immunization: A Content Analysis. JAMA, 298(21), 2481.
https://doi.org/10.1001/jama.298.21.2482
Kim, K., Paek, H.-J., & Lynn, J. (2010). A Content Analysis of Smoking Fetish Videos
on YouTube: Regulatory Implications for Tobacco Control. Health
Communication, 25(2), 97–106. https://doi.org/10.1080/10410230903544415
Klemm, P., Bunnell, D., Cullen, M., Soneji, R., Gibbons, P., & Holecek, A. (2003).
Online Cancer Support Groups: A Review of the Research Literture. Computers,
Informatics, Nursing, 21(3), 136–142.
Klosky, J. L., Howell, C. R., Li, Z., Foster, R. H., Mertens, A. C., Robison, L. L., & Ness,
K. K. (2012). Risky Health Behavior Among Adolescents in the Childhood
Cancer Survivor Study Cohort. Journal of Pediatric Psychology, 37(6), 634–646.
https://doi.org/10.1093/jpepsy/jss046
Korda, H., & Itani, Z. (2013). Harnessing social media for health promotion and behavior
change. Health Promotion Practice, 14(1), 15–23.
Kwak, H., Lee, C., Park, H., & Moon, S. (2010). What is Twitter, a Social Network or a
News Media? In Proceedings of the 19th International Conference on World
Wide Web (pp. 591–600). New York, NY, USA: ACM.
https://doi.org/10.1145/1772690.1772751
Lapinski, M. K. (2006). StarvingforPerfect.com: A Theoretically Based Content Analysis
of Pro-Eating Disorder Web Sites. Health Communication, 20(3), 243–253.
https://doi.org/10.1207/s15327027hc2003_4
103
Lewis, S. P., Heath, N. L., Sornberger, M. J., & Arbuthnott, A. E. (2012). Helpful or
Harmful? An Examination of Viewers’ Responses to Nonsuicidal Self-Injury
Videos on YouTube. Journal of Adolescent Health, 51(4), 380–385.
https://doi.org/10.1016/j.jadohealth.2012.01.013
Lewis, S. P., Heath, N. L., St Denis, J. M., & Noble, R. (2011). The Scope of Nonsuicidal
Self-Injury on YouTube. Pediatrics, 127(3), e552–e557.
https://doi.org/10.1542/peds.2010-2317
Lienemann, B. A., Unger, J. B., Cruz, T. B., & Chu, K.-H. (2017). Methods for Coding
Tobacco-Related Twitter Data: A Systematic Review. Journal of Medical Internet
Research, 19(3), e91. https://doi.org/10.2196/jmir.7022
Luxton, D. D., June, J. D., & Fairall, J. M. (2012). Social Media and Suicide: A Public
Health Perspective. American Journal of Public Health, 102(S2), S195–S200.
https://doi.org/10.2105/AJPH.2011.300608
Mackey, T. K., & Liang, B. A. (2013). Global Reach of Direct-to-Consumer Advertising
Using Social Media for Illicit Online Drug Sales. Journal of Medical Internet
Research, 15(5), e105. https://doi.org/10.2196/jmir.2610
Mackie, D. M. (1986). Social identification effects in group polarization. Journal of
Personality and Social Psychology, 50(4), 720–728. https://doi.org/10.1037/0022-
3514.50.4.720
Maher, C. A., Lewis, L. K., Ferrar, K., Marshall, S., De Bourdeaudhuij, I., &
Vandelanotte, C. (2014). Are Health Behavior Change Interventions That Use
Online Social Networks Effective? A Systematic Review. Journal of Medical
Internet Research, 16(2), e40. https://doi.org/10.2196/jmir.2952
104
McKenna, K. Y., Bargh, J. A., Levine, J. M., & Moreland, R. L. (2004). Coming out in
the age of the Internet: Identity “demarginalization”through virtual group
participation. In J. M. Levine & R. L. Moreland (Eds.), Small groups: Key
readings. Psychology Press.
McLaughlin, M. L., Hou, J., Meng, J., Hu, C.-W., An, Z., Park, M., & Nam, Y. (2016).
Propagation of information about Preexposure Prophylaxis (PrEP) for HIV
prevention through Twitter. Health Communication, 1–10.
https://doi.org/10.1080/10410236.2015.1027033
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(2), 631–641.
https://doi.org/10.1016/j.chb.2011.11.009
Mclaughlin, M., Park, M., & Sun, Y. (2015). Sharing Information Promoting Unhealthy
Behavior through Social Media: Examination of the Users and the Content Using
the Case of Pro-Anorexia in Tumblr (pp. 800–803). IEEE.
https://doi.org/10.1109/CSCI.2015.122
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a Feather: Homophily
in Social Networks. Annual Review of Sociology, 27(1), 415–444.
https://doi.org/10.1146/annurev.soc.27.1.415
Montano, D. E., & Kasprzyk, D. (2008). Theory of reasoned action, theory of planned
behavior, and the integrated behavioral model. In K. Glanz, B. K. Rimer, & K.
Viswanath (Eds.), Health behavior and health education: theory, research, and
practice (pp. 67–96). John Wiley & Sons.
105
Moreno, M. A., Grant, A., Kacvinsky, L., Egan, K. G., & Fleming, M. F. (2012). College
students’ alcohol displays on facebook: Intervention considerations. Journal of
American College Health, 60(5), 388–394.
https://doi.org/10.1080/07448481.2012.663841
Napolitano, M. A., Hayes, S., Bennett, G. G., Ives, A. K., & Foster, G. D. (2013). Using
facebook and text messaging to deliver a weight loss program to college students.
Obesity, 21(1), 25–31. https://doi.org/10.1002/oby.20232
Noar, S. M., Myrick, J. G., Morales-Pico, B., & Thomas, N. E. (2014). Development and
Validation of the Comprehensive Indoor Tanning Expectations Scale. JAMA
Dermatology, 150(5), 512. https://doi.org/10.1001/jamadermatol.2013.9086
Norris, P. (2004). The bridging and bonding role of online communities. In P. N. Howard
& S. Jones (Eds.), Society Online. Thousand Oaks, California: Sage Publications.
Oksanen, A., Garcia, D., Sirola, A., Näsi, M., Kaakinen, M., Keipi, T., & Räsänen, P.
(2015). Pro-Anorexia and Anti-Pro-Anorexia Videos on YouTube: Sentiment
Analysis of User Responses. Journal of Medical Internet Research, 17(11), e256.
https://doi.org/10.2196/jmir.5007
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 behaviour: a systematic review and meta-
analysis. International Journal of Behavioral Nutrition and Physical Activity,
10(1), 29. https://doi.org/10.1186/1479-5868-10-29
Paek, H.-J., Kim, S., Hove, T., & Huh, J. Y. (2014). Reduced Harm or Another Gateway
to Smoking? Source, Message, and Information Characteristics of E-Cigarette
106
Videos on YouTube. Journal of Health Communication, 19(5), 545–560.
https://doi.org/10.1080/10810730.2013.821560
Pang, B., & Lee, L. (2008). Opinion Mining and Sentiment Analysis. Found. Trends Inf.
Retr., 2(1–2), 1–135. https://doi.org/10.1561/1500000011
Park, M., Sun, Y., & McLaughlin, M. (2017). Social Media Propagation of Content
Promoting Risky Health Behavior. Cyberpsychology, Behavior, and Social
Networking, 20(5), 278–285. https://doi.org/10.1089/cyber.2016.0698
Park, N., Kee, K. F., & Valenzuela, S. (2009). Being Immersed in Social Networking
Environment: Facebook Groups, Uses and Gratifications, and Social Outcomes.
CyberPsychology & Behavior, 12(6), 729–733.
https://doi.org/10.1089/cpb.2009.0003
Parks, M. R., & Floyd, K. (1996). Making Friends in Cyberspace. Journal of Computer-
Mediated Communication, 1(4), 0–0. https://doi.org/10.1111/j.1083-
6101.1996.tb00176.x
Peters, G.-J. Y., Ruiter, R. A. C., & Kok, G. (2013). Threatening communication: a
critical re-analysis and a revised meta-analytic test of fear appeal theory. Health
Psychology Review, 7(sup1), S8–S31.
https://doi.org/10.1080/17437199.2012.703527
Putnam, R. D. (2000). Bowling alone: The collapse and revival of American community.
Simon and Schuster.
Rains, S. A., & Tukachinsky, R. (2015). Information Seeking in Uncertainty
Management Theory: Exposure to Information About Medical Uncertainty and
Information-Processing Orientation as Predictors of Uncertainty Management
107
Success. Journal of Health Communication, 20(11), 1275–1286.
https://doi.org/10.1080/10810730.2015.1018641
Reavley, N. J., & Pilkington, P. D. (2014). Use of Twitter to monitor attitudes toward
depression and schizophrenia: an exploratory study. PeerJ, 2, e647.
https://doi.org/10.7717/peerj.647
Ribisl, K. M. (2003). The potential of the internet as a medium to encourage and
discourage youth tobacco use. Tobacco Control, 12(suppl 1), i48–i59.
https://doi.org/10.1136/tc.12.suppl_1.i48
Ridings, C., Gefen, D., & Arinze, B. (2006). Psychological Barriers: Lurker and Poster
Motivation and Behavior in Online Communities. Communications of the
Association for Information Systems, 18, 329–354.
Ridings, C. M., & Gefen, D. (2004). Virtual Community Attraction: Why People Hang
Out Online. Journal of Computer-Mediated Communication, 10(1), 00–00.
https://doi.org/10.1111/j.1083-6101.2004.tb00229.x
Rime, B. (2009). Emotion Elicits the Social Sharing of Emotion: Theory and Empirical
Review. Emotion Review, 1(1), 60–85.
https://doi.org/10.1177/1754073908097189
Rouleau, C. R., & von Ranson, K. M. (2011). Potential risks of pro-eating disorder
websites. Clinical Psychology Review, 31(4), 525–531.
https://doi.org/10.1016/j.cpr.2010.12.005
Rubin, A. M. (2009). Uses-and-gratifications perspective on media effects. In J. Bryant &
M. B. Oliver (Eds.), Media Effects: Advances in Theory and Research. Routledge.
108
Ruder, T. D., Hatch, G. M., Ampanozi, G., Thali, M. J., & Fischer, N. (2011). Suicide
Announcement on Facebook. Crisis: The Journal of Crisis Intervention and
Suicide Prevention, 32(5), 280–282. https://doi.org/10.1027/0227-5910/a000086
Seidenberg, A. B., Rodgers, E. J., Rees, V. W., & Connolly, G. N. (2012). Youth Access,
Creation, and Content of Smokeless Tobacco (“Dip”) Videos in Social Media.
Journal of Adolescent Health, 50(4), 334–338.
https://doi.org/10.1016/j.jadohealth.2011.09.003
Shaw, J. M., Mitchell, C. A., Welch, A. J., & Williamson, M. J. (2015). Social media
used as a health intervention in adolescent health: A systematic review of the
literature. Digital Health, 1, 2055207615588395.
https://doi.org/10.1177/2055207615588395
Shoemaker, P. J. (1996). Hardwired for News: Using Biological and Cultural Evolution
to Explain the Surveillance Function. Journal of Communication, 46(3), 32–47.
https://doi.org/10.1111/j.1460-2466.1996.tb01487.x
Stieglitz, S., & Dang-Xuan, L. (2013). Emotions and Information Diffusion in Social
Media—Sentiment of Microblogs and Sharing Behavior. Journal of Management
Information Systems, 29(4), 217–248. https://doi.org/10.2753/MIS0742-
1222290408
Tajfel, H. (1982). Social Psychology of Intergroup Relations. Annual Review of
Psychology, 33(1), 1–39. https://doi.org/10.1146/annurev.ps.33.020182.000245
Thackeray, R., Neiger, B. L., Smith, A. K., & Van Wagenen, S. B. (2012). Adoption and
use of social media among public health departments. BMC Public Health, 12(1).
https://doi.org/10.1186/1471-2458-12-242
109
Thelwall, M., Buckley, K., & Paltoglou, G. (2011). Sentiment in Twitter events. Journal
of the American Society for Information Science and Technology, 62(2), 406–418.
https://doi.org/10.1002/asi.21462
Tong, S. T., Heinemann-LaFave, D., Jeon, J., Kolodziej-Smith, R., & Warshay, N.
(2013). The Use of Pro-Ana Blogs for Online Social Support. Eating Disorders,
21(5), 408–422. https://doi.org/10.1080/10640266.2013.827538
Valente, T. W. (2012). Network Interventions. Science, 337(6090), 49–53.
https://doi.org/10.1126/science.1217330
Valente, T. W., Fujimoto, K., Unger, J. B., Soto, D. W., & Meeker, D. (2013). Variations
in network boundary and type: A study of adolescent peer influences. Social
Networks, 35(3), 309–316. https://doi.org/10.1016/j.socnet.2013.02.008
Valente, T. W., Hoffman, B. R., Ritt-Olson, A., Lichtman, K., & Johnson, C. A. (2003).
Effects of a Social-Network Method for Group Assignment Strategies on Peer-
Led Tobacco Prevention Programs in Schools. American Journal of Public
Health, 93(11), 1837–1843. https://doi.org/10.2105/AJPH.93.11.1837
Valente, T. W., & Vlahov, D. (2001). Selective risk taking among needle exchange
participants: implications for supplemental interventions. American Journal of
Public Health, 91(3), 406–411.
van der Tempel, J., Noormohamed, A., Schwartz, R., Norman, C., Malas, M., &
Zawertailo, L. (2016). Vape, quit, tweet? Electronic cigarettes and smoking
cessation on Twitter. International Journal of Public Health, 61(2), 249–256.
https://doi.org/10.1007/s00038-016-0791-2
110
Vilhauer, R. P., McClintock, M. K., & Matthews, A. K. (2010). Online Support Groups
for Women with Metastatic Breast Cancer: A Feasibility Pilot Study. Journal of
Psychosocial Oncology, 28(5), 560–586.
https://doi.org/10.1080/07347332.2010.504504
Walther, J. B., Heide, B. V. D., Hamel, L. M., & Shulman, 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. https://doi.org/10.1177/0093650208330251
Wang, L., Zhan, Y., Li, Q., Zeng, D., Leischow, S., & Okamoto, J. (2015). An
Examination of Electronic Cigarette Content on Social Media: Analysis of E-
Cigarette Flavor Content on Reddit. International Journal of Environmental
Research and Public Health, 12(12), 14916–14935.
https://doi.org/10.3390/ijerph121114916
Wasko, M., & Faraj, S. (2000). “It is what one does”: why people participate and help
others in electronic communities of practice. The Journal of Strategic Information
Systems, 9(2–3), 155–173. https://doi.org/10.1016/S0963-8687(00)00045-7
Weinhardt, L., Forsyth, A., Carey, M., Jaworski, B., & Durant, L. (1998). Reliability and
Validity of Self-Report Measures of HIV-Related Sexual Behavior: Progress
Since 1990 and Recommendations for Research and Practice. Archives of Sexual
Behavior, 27(2), 155–180.
Williams, D. (2006). Virtual Cultivation: Online Worlds, Offline Perceptions. Journal of
Communication, 56(1), 69–87. https://doi.org/10.1111/j.1460-2466.2006.00004.x
111
Williams, G. C., Hedberg, V. A., Cox, E. M., & Deci, E. L. (2000). Extrinsic Life Goals
and Health-Risk Behaviors in Adolescents1. Journal of Applied Social
Psychology, 30(8), 1756–1771. https://doi.org/10.1111/j.1559-
1816.2000.tb02466.x
Witte, K. (1994). Fear control and danger control: A test of the extended parallel process
model (EPPM). Communication Monographs, 61(2), 113–134.
https://doi.org/10.1080/03637759409376328
Wright, K. B. (2009). Increasing computer-mediated social support. In J. C. Parker & E.
Thorson (Eds.), Health communication in the new media landscape (pp. 243–
265). Springer New York, NY.
Yeshua-Katz, D., & Martins, N. (2013). Communicating Stigma: The Pro-Ana Paradox.
Health Communication, 28(5), 499–508.
https://doi.org/10.1080/10410236.2012.699889
Yom-Tov, E., Fernandez-Luque, L., Weber, I., & Crain, S. P. (2012). Pro-Anorexia and
Pro-Recovery Photo Sharing: A Tale of Two Warring Tribes. Journal of Medical
Internet Research, 14(6), e151. https://doi.org/10.2196/jmir.2239
Yoo, W., Yang, J., & Cho, E. (2016). How social media influence college students’
smoking attitudes and intentions. Computers in Human Behavior, 64, 173–182.
https://doi.org/10.1016/j.chb.2016.06.061

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

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

PDF

Democrats vs. Republicans: how Twitter users share and react to political fact-checking messages

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

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

School-based Health Centers 2.0: Using social media to increase utilization of UMMA Wellness Center Services

PDF

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

PDF

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

PDF

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

PDF

Benefits of social networking in online social support groups

PDF

The social meaning of sharing and geocoding: features and social processes in online communities

PDF

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

PDF

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

PDF

Emotion culture through the lens of PUA (pickup artist): emotions and the social media narratives of structual gendered violence

PDF

Communication and social support of parents of children treated at Childrens Hospital Los Angeles

PDF

Evaluating social networks and impact of micro-influencers who promote e-cigarettes on social media

PDF

Adolescent social networks, smoking, and loneliness

PDF

Developing and testing a heuristic-systematic model of health decision making: the role of affect, trust, confidence and media influence

PDF

The role of depression symptoms on social information processing and tobacco use among adolescents

PDF

Social support in Cambodia: the role of peer educators in behavior change

Asset Metadata

Creator Park, Mina (author)

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

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Publication Date 03/12/2019

Defense Date 07/20/2017

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

Tag information sharing,OAI-PMH Harvest,relationship formation,risky health behavior,smoking,social identity,social media,social support,stigma

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor McLaughlin, Margaret (committee chair), Cody, Michael (committee member), Valente, Tom (committee member)

Creator Email minapark@usc.edu,minapark47@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-427810

Unique identifier UC11265339

Identifier etd-ParkMina-5715.pdf (filename),usctheses-c40-427810 (legacy record id)

Legacy Identifier etd-ParkMina-5715.pdf

Dmrecord 427810

Document Type Dissertation

Rights Park, Mina

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