University of Southern California Dissertations and Theses

Friendship network position on adolescent behaviors: an examination of a broker position and the likelihood of alcohol and cigarette use

(USC Thesis Other)

Friendship network position on adolescent behaviors: an examination of a broker position and the likelihood of alcohol and cigarette use

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Friendship Network Position on Adolescent Behaviors:
An Examination of a Broker Position and The Likelihood of Alcohol and Cigarette Use

by

Jihye Yoo Lee

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

August 2023

ii
ACKNOWLEDGEMENTS
I would like to thank all the people who helped me make it through the dissertation process.
First, I would like to express my sincere gratitude to my adviser and dissertation chair, Dr. Ann
Owens for continuous support and guidance during my dissertation journey. I am really thankful
that I could share my frustrations and challenges with her during this journey and she was there
for me especially when I felt lost and needed guidance. Without her support and mentorship, I
cannot imagine how this journey could have come to this end. I truly appreciate her patience (in
coping with my frustrations and this “long” journey) and endless support. My son, Noah, also
wants to express special thanks for your holiday gifts including nut snacks and hot chocolates.
I would also like to express my appreciation to Dr. Lynne Casper. Although she is no
longer a member of my dissertation committee, she provided me with support and advice even
prior to my USC journey. She is the person who inspired and taught me about the importance of
family sociology and demography. When I worked as a teaching assistant for her classes like the
sociology, demography and health course, she taught me interactive pedagogy and showed her
intelligence and experience on social demography with her charisma. Very special thanks to Dr.
Thomas Valente for his practical advice and knowledge on social network analyses. It is
fortunate that I had his social network course at the beginning of my doctoral program and have
him as my committee member. Dr. Valente not only gave me advice on social network analyses
and theories on my dissertation but also connected me with other social network researchers and
doctoral students. I would like to thank Dr. Timothy Biblarz for your mentorship and support. I
have learned from your family sociology course and developed the idea for my dissertation about
brokers and their roles in adolescent behaviors. This dissertation would have been almost
impossible without Dr. James Polk. He is my friend, ALI writing editor, and intelligent kitty-

iii
lover. I truly appreciate his support and comments on my endless drafts. I hope he enjoys his
“brighter” life after this semester.
I want to express to my appreciation to my best friend, Hyojin Song. Since high school,
she has always been there for me. She is more than a friend; Hyojin is my family, colleague,
unofficial counselor, and fellow USC alumnae. Without her, I cannot imagine how I could have
started and finished my PhD journey. In Korea, we prepared together and supported each other in
pursuing graduate studies in the U.S. When I was planning to apply to USC, she showed me
what life at USC and in LA would be like. When I started my PhD program at USC, she helped
me move, took me out to explore the city, and introduced me to her friends and colleagues. She
cried, laughed, and celebrated with me as the course of my PhD unfolded, and whenever I
wanted to quit or felt inadequate, she always said, “just hang in there” and encouraged me that I
would eventually achieve my goals. Hyojin, I am finally done!
Thank you to my family for their many sacrifices. My four-year-old son is my inspiration
and motivation for improving adolescent life. My husband, Josh, is a patient and good listener
because he listened and responded when I talked about my research, teaching, conference
presentation even though he has no interest in my research. His feedback and commentary often
confused me and raised more questions, but it also ended up giving me good ideas for improving
my research, although I did not tell my husband. Thank you, Josh.
Finally, I would like to give a huge thanks to my parents and sister. My parents are not
college graduates, but they always encouraged me and my sister to pursue higher education and
to dream big. Both of my parents worked hard, in different ways, to help foster and nurture these
ideals. Growing up, my dad worked from 4 am in the morning until late at night to support us.
He never complained; his support was unequivocal. My mother always pushed and encouraged

iv
me to pursue academic opportunities. Her support of my graduate studies in America was no
different. From 2019 on, my mom regularly came to the U.S. to help take care of my son.
Although she does not speak English and is unfamiliar with American life, she was and is always
willing to come support me and to help take care of her grandson. And a special thank you to my
sister, Soyoung, for listening to my random stories and watching Netflix shows with me to
understand what I have been through. I never say this to my parents and sister, but I want to
express that “I really thank you for everything and I love you.”

v
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ......................................................................................................... ii
LIST OF TABLES ...................................................................................................................... vii
LIST OF FIGURES ..................................................................................................................... ix
ABSTRACT ....................................................................................................................................x
CHAPTER 1 INTRODUCION .....................................................................................................1
Background ..........................................................................................................................1
Overview of Dissertation .....................................................................................................5
CHAPTER 2 WHO ARE THE BROKERS? ..............................................................................8
Introduction ..........................................................................................................................8
The Concepts and Measurements of a Broker in Social Networks ...................................11
Individual Status and Brokers ............................................................................................13
Extracurricular Activities and Brokers ..............................................................................16
Data and Measures .............................................................................................................18
Analytic Strategy ...............................................................................................................23
Results ................................................................................................................................24
Discussion ..........................................................................................................................33
CHAPTER 3 WHEN BROKERS SEEK MORE FRIENDSHIPS ..........................................38
Introduction ........................................................................................................................38
Previous Studies on Broker Position and Substance Use ..................................................41
Numbers and Directionality of a Broker’s Ties and Substance Use Behaviors ................43
Data, Measures, and Methods ............................................................................................46
Results ................................................................................................................................55
Discussion ..........................................................................................................................72
CHAPTER 4 STRUCTURAL AND RELATIONAL EMBEDDEDNESS ON
SUBSTANCE USE .......................................................................................................................80
Introduction ........................................................................................................................80
Structural and Relational Dimensions of Brokers and the Likelihood of Substance
Use .....................................................................................................................................82
Cross-gender Friendships and Substance Use ...................................................................85
Data, Measures, and Methods ............................................................................................89
Results ................................................................................................................................96
Conclusion .......................................................................................................................105

vi
CHAPTER 5 CONCLUSION ...................................................................................................110
Summary ..........................................................................................................................110
Limitations .......................................................................................................................113
REFERENCES ...........................................................................................................................117
APPENDICES ............................................................................................................................128
Appendix Chapter 2 .........................................................................................................128
Appendix Chapter 3 .........................................................................................................131
Appendix Chapter 4 .........................................................................................................139

vii
LIST OF TABLES
Table 2.1. Descriptive Statistics for Students and Schools ............................................................25

Table 2.2. Relationship between Heterogeneous Friendships and Broker Students
(N=64,412) .....................................................................................................................................26

Table 2.3. Results of Multilevel Models Predicting EV-Brokerage Scores from ECA
Participation and Race/Ethnicity and SES of Students ..................................................................27

Table 2.4. Interaction Effects between Race/Ethnicity of Students and ECA Participation .........30

Table 3.1. Descriptive Sample Statistics Among 12th-grade students in Five Southern
California Schools: Social Network Study, Spring 2013 (n = 1,265) ............................................55

Table 3.2. Logistic Regression Models of Alcohol Use: Broker vs. Broker’s Friend Ties
(N = 1,265) .....................................................................................................................................58

Table 3.3. Logistic Regression Models of Cigarette Use: Broker vs. Broker’s Friend Ties
(N= 1,265) ......................................................................................................................................60

Table 3.4. Logistic Regression Models of Alcohol Use: Broker’s User Friends Who
Reported Alcohol and Cigarettes (N=1,265) .................................................................................62

Table 3.5. Logistic Regression Models of Cigarette Use: Broker’s User Friends Who
Reported Alcohol and Cigarettes (N=1,265) .................................................................................65

Table 3.6. Logistic Regression Models of Alcohol Use: Broker’s Indegree vs. Broker’s
Outdegree (N= 1,265) ....................................................................................................................66

Table 3.7. Logistic Regression Models of Cigarette Use: Indegree Brokers vs.
Outdegree Brokers (N= 1,265) ......................................................................................................70

Table 4.1. A Summary of Previous Studies on a Broker’s Substance Use ....................................83

Table 4.2. A Summary of Previous Studies on Cross-gender Friendships ....................................88

Table 4.3. Descriptive Sample Statistics of Wave 1 in Add Health Data, 1994-95
(n = 61,608 of students, N = 100 of schools) .................................................................................96

Table 4.4. Odds Ratios on the Likelihood of Alcohol Consumption by Gender
(N = 61,608) ...................................................................................................................................98

Table 4.5. Odds Ratios on the Likelihood of Cigarette Smoking by Gender
(N = 61,608) .................................................................................................................................100

viii
Table 4.6. Odds Ratios on the Likelihood of Drunkenness by Gender (N = 61,608)..................102

Appendix 2 Table A. Missing/Imputed Values ...........................................................................128

Appendix 2 Table B. Multilevel Models Predicting EV Brokerage Scores from
Key Predictors Including Each of the Student-level Control Variables ......................................130

Appendix 3 Table A. Brokerage Measures Classified by Centrality vs. Bridging and
by Local vs. Global Network .......................................................................................................131

Appendix 3 Table B. Bivariate Analysis Table of Substance Use (Alcohol and Cigarette) .......132

Appendix 3 Table C. Logistic Regression Models of Control Variables on Substance Use .......135

Appendix 3 Table D. Descriptive Statistics of 14 Participants Who Have Missingness
in Network Data, comparing 1,265 Respondents Who Have Network Information ..................136

Appendix 3 Table E. Missing/Imputed Values ............................................................................137

Appendix 3 Table F. Logistic Regression Results between Brokers with Multiple Ties
and High-central Individuals with Low Brokerage .....................................................................138

Appendix 4 Table A. Missing/Imputed Values (N = 61,608)......................................................139

Appendix 4 Table B. Descriptive Sample Statistics Differing by Gender ..................................140

Appendix 4 Table C. Unconditional and Basic Models for Alcohol and Cigarette Use and
Drunkenness by Gender ...............................................................................................................141

Appendix 4 Table D. Odds Ratios on the Likelihood of Alcohol Consumption by Gender
Excluding the Interaction Term ...................................................................................................142

Appendix 4 Table E. Odds Ratios on the Likelihood of Cigarette Smoking by Gender
Excluding the Interaction Term ...................................................................................................143

Appendix 4 Table F. Odds Ratios on the Likelihood of Drunkenness by Gender
Excluding the Interaction Term ...................................................................................................144

ix
LIST OF FIGURES
Figure 2.1. Hypothetical Network with a Broker ................................................................................. 13
Figure 3.1. Comparison of Probability of Drinking Depending on Numbers of Ties Between
Low- and High-brokerage Groups .......................................................................................................... 59

Figure 3.2. Comparison of Probability of Drinking Depending on Numbers of Peers Who Drink
Alcohol Between Low- and High-brokerage Groups .......................................................................... 63

Figure 3.3. Comparison of Probability of Drinking Depending on Numbers of Peers Who Smoke
Cigarettes Between Low- and High-brokerage Groups ....................................................................... 64

Figure 3.4. Comparison of Probability of Drinking Depending on Numbers of Indegrees Between
Low- and High-brokerage Groups .......................................................................................................... 68

Figure 3.5. Comparison of Probability of Drinking Depending on Numbers of Outdegrees
Between Low- and High-brokerage Groups.......................................................................................... 71

Figure 3.6. Comparison of Probability of Smoking Depending on Numbers of Indegrees Between
Low- and High-brokerage Groups .......................................................................................................... 72

Figure 3.7. Comparison of Probability of Smoking Depending on Numbers of Outdegrees
Between Low- and High-brokerage Groups.......................................................................................... 24

Appendix 2 Figure 2.2. Percentage of Cross-Racial Friendships by Race/Ethnicity
(N=64,412)................................................................................................................................................. 129

x
ABSTRACT
Extensive literature has proven that brokers—those who bridge distant individuals or
cliques—are more likely to adopt new or innovative behaviors. However, brokers’ propensity to
also engage in risky behaviors such as substance use is yet to fully be investigated. And, if
brokers engage in risky behaviors, it is necessary to know what conditions and mechanisms they
do. To explore these main questions, this dissertation examines 1) the potential characteristics of
adolescent brokers and environments that help the emergence of brokers in friendships networks,
2) the number and directionality of brokers’ ties in understanding different levels of alcohol and
cigarette use, and 3) the structural and relational aspects of brokers that help explain the
likelihood of risky behavior such as alcohol consumption and cigarette smoking.
Focusing on the adolescent friendship networks, the second chapter of this dissertation
starts with the examination of characteristics that conceptualizes a broker. Drawing on multilevel
models using Wave 1 (64,412 students in 110 schools) of the National Longitudinal Study of
Adolescent to Adult Health, I find that adolescent brokers are more likely to be (1) racial/ethnic
minority (except Asian) students, (2) students from low socioeconomic status backgrounds, and
(3) art club members. Accordingly, I argue that minority and disadvantaged-status students often
play a prominent role as brokers in integrating classmates within schools. The results also
identify that the effects of extracurricular activities (ECAs) are dependent upon the types of the
activities pursued and an adolescent’s race/ethnicity. Stated differently, each race/ethnicity group
shows different levels of being brokers depending on whether they participate in sports, arts,
and/or academic activities.
In the third chapter, this dissertation examines whether and how brokers engage in
substance use with respect to the number and directionality of their ties. Based on the contagion

xi
theory and social acceptance hypothesis, this study examines two main questions: 1) does the
number of friendship ties impact substance use? And 2) are brokers who nominate more
friendships at a higher risk of substance use than brokers who do not? Using the 2013 Social
Networking Study data (1,265 students in 5 schools) administered by Dr. Valente and his
colleagues from the University of Southern California, I conducted logistic regression analyses
that explore whether and how brokers’ friendships explain the likelihood of their alcohol
consumption and cigarette smoking. The results do not provide the support of contagion theory
that there is no positive association between the number of a brokers’ friends the likelihood of
their substance use. Instead, the results support the social acceptance hypothesis that brokers who
nominate more friendships seem to smoke cigarettes more compared to those who do not.
In the fourth chapter, I take a closer look at both the structural and relational dimensions
of a broker’s position in explaining the likelihood of alcohol and cigarette use. Because brokers
occupy advantageous social positions via long ties that can reach out to distant individuals and
their heterogeneous ties including substance users, this study hypothesizes that brokers are more
likely to engage in substance use. At the same time, it is also expected that brokers can be
accessible to non-substance users and their ties with substance users are too weak to activate
their risky behaviors. To investigate these conflicting assumptions, the present study examines
three hypotheses: 1) brokers use substances more than non-brokers due to their long ties that
reach out to distant others, 2) brokers use substances more than non-brokers due to their
heterogeneous ties such as cross-gender friendships that immediately connect with diverse
individuals, and 3) these associations differ by gender. Using Wave 1 (61,191 students in 100
schools) of the National Longitudinal Study of Adolescent to Adult Health, I conducted
multilevel logistic regression models that explore whether and how brokers’ long and

xii
heterogeneous ties explain the likelihood of their alcohol and cigarette use. Findings show that
the long ties of brokers do not necessarily increase the likelihood of substance use, but having
heterogeneous ties—cross-gender friendships in this study—is positively associated with the
likelihood of substance use for both male and female students.
This dissertation provides practical and theoretical contributions. While most prior
studies primarily focus on the benefits of a broker position, this dissertation identifies potential
characteristics of adolescent brokers and surrounding environments that foster brokers in
friendship networks. Considering the importance of social status in adolescent friendships, this
dissertation investigates whether and how an adolescent’s “status” matters in occupying a broker
position. Additionally, this study examines whether and what conditions an adolescent broker
can engage in “risky” behaviors, expanding the discussion on the contagion theory and social
acceptance explanation in understanding a broker’s behaviors. Exploring a broker’s structural
and relational aspect on substance use, this research provides a potential mechanism of whether
and how a broker’s long and heterogeneous ties help engage in substance use behaviors.

1
CHAPTER 1
INTRODUCTION

Background
What is a broker? This question usually conjures images of a real estate agent or financial
planner helping to facilitate and connect buyers and seller of assets. Considering the role of a
broker that connects two or more different parties, these images of a broker are not completely
far from the concept of a broker in social network literature. Similarly, in social networks, a
broker is defined as a unique intermediary who connects distant or separate individuals and
groups (Burt 1992; Gould and Fernandez 1989; Stovel and Shaw 2012). This concept of a broker
has its origins in the work of Granovetter (1973). Although Granovetter did not directly use or
define the term "broker" in his weak ties theory, he observed that people who loosely connect
acquaintances and dissimilar others serve important roles in bridging gaps and reducing the
social distance between individuals. Accordingly, brokers tend to receive and transfer different
resources, information, goods, and opportunities from each of the groups, and therefore a broker
is often viewed as an advantageous position in social networks.
Five decades after Granovetter’s theory of weak ties (1973), social scientists have
examined how beneficial a broker position is, particularly in adopting new and innovative
information and behaviors (Brown and Konrad 2001; Crowell 2004; Fleming, Mingo, and Chen
2007; Gould and Fernandez 1989; Mangino 2009; Granovetter 1977; Burt 2004; 2002b).
Because a broker has not only long ties that connect distant individuals as far as possible but also
heterogeneous ties that connect diverse people, a broker serves a crucial role in reducing social
distances and integrating different individuals or groups in social networks. Although the broker
position has attracted the attention of researchers, research rarely addresses the characteristics of

2
individuals acting as brokers or the environment that fosters development of brokers.
Considering the importance of the broker position, it is not surprising to believe that high-status
individuals easily occupy these advantageous positions because they have more capital to expend
(Briggs 2002; McDonald 2011; Mouw 2009; Stanton-Salazar and Dornbusch 1995). On the
contrary, high-status individuals may not serve broker positions; instead, they prefer to form
social closure with individuals of similar backgrounds, because they get little or no reward from
social interaction with dissimilar people, particularly with low-status actors (Coleman 1988;
Molm and Cook 1995; Molm, Schaefer, and Collett 2007; Schaefer 2012). Due to the lack of
attention on potential characteristics of brokers, we have an opaque picture of who brokers are
and what environments foment the emergence of brokers in friendship networks.
In addition to these questions, recent scholarship also casts doubt on whether a broker
position is always beneficial (Burt 2002a; Stovel and Shaw 2012; Xiao and Tsui 2007; Aral
2016). Although a broker’s weak ties with others can bring diverse opportunities, information, or
resources to a broker, these benefits may not last long. Additionally, the diversity of
opportunities, information, and resources can have both positive and negative impacts. Because a
broker’s connections with other people are often weak and precarious, the volatility and decay
rates of a broker’s ties are quite high (Burt 2002a; Burt, Kilduff, and Tasselli 2013). This
indicates that brokerage is dynamic: brokers at time 1 may not be brokers at time 2 and vice
versa, and that a broker’s social capital may dissipate or disappear over time. Furthermore,
brokers may experience role strains derived from multiple group memberships or face conflicts
between diverse peer groups. Given the unique nature of brokerage, some seek refuge in using
substances to deal with the tremendous stress and expectations of their social status or to
promote social acceptance from their peer groups (Henry and Kobus 2007). In fact, several

3
studies have found a positive association between being a broker and likelihood of substance use
(Henry and Kobus 2007; Osgood et al. 2014; Kreager and Haynie 2011; Kreager, Haynie, and
Hopfer 2013). In addition, researchers also find that some brokers may facilitate diffusion of
risky behaviors across social networks (Centola and Macy 2007; Fink et al. 2016; Ghasemiesfeh,
Ebrahimi, and Gao 2013; Vasconcelos, Levin, and Pinheiro 2019). These findings raise the
question: “does a broker also engage in deviant or risky behaviors?”
The initial idea—supporting benefits and advantages of a broker position—elucidates that
a broker’s long and heterogeneous ties help a broker to receive diverse resources and good
information or behaviors. However, it is also possible that brokers can engage in or facilitate
risky behaviors such as substance use because of their long and heterogeneous ties. Their long
and heterogeneous ties may provide brokers greater exposure and access to both good
information or behaviors and also risky behaviors and delinquent peers.
Centola and Macy (2007) suggest the idea that a broker can engage in or promote
behaviors across networks subject to certain conditions. Most social behaviors follow the
complex contagion process that requires higher thresholds for activation and multiple exposures
to sources. That is, brokers may engage in adoption or diffusion of risky behaviors when they
have multiple exposures to risky behaviors or other people who help activate risky behaviors
regardless of their advantageous position in social networks. Accordingly, brokers need to have
not only “long” ties—connecting distant individuals as far as possible—but also “wide” ties—
having multiple connections with as many ties as possible—to activate such contagions. In
addition, the social acceptance explanation also supports this notion considering the unstable and
fragile status of a broker position in peer networks. Because brokers have weak and precarious
peer relations, some brokers who desperately seek status and social acceptance from peers may

4
willing engage in risky behaviors to gain acceptance and status from peer groups as discussed
above.
Moreover, it is uncertain if a broker’s long ties or heterogeneous ties lead to such
consequences. Although many brokerage studies rely on the assumption that a broker’s
immediate ties are diverse (heterogeneous ties) and a broker is connected with many others
across networks (long ties), these two types of ties are often treated as identical or without
distinction. However, a broker’s long ties and heterogeneous ties are different in many ways.
Most importantly, the long ties indicate a broker’s structural aspect, that is, where a broker is
positioned in an entire network. On the other hand, the heterogeneous ties describe a broker’s
relational aspect such as whom a broker directly befriends. In this context, it is possible that
some individuals can have long ties but no heterogeneous ties and vice versa. Additionally,
measurement approaches are also different. Long ties are evaluated with a macro- or global-level
measure such as the shortest path calculation (Freeman 1977), whereas heterogeneous ties are
measured by a micro-level method such as a number of a broker’s immediate friends. Regardless
of these differences, prior studies often fail to distinguish between a broker’s long ties and
heterogeneous ties in examining brokerage.
To fill the gaps in broker literature, I will use adolescent friendship networks and to focus
on their alcohol and cigarette use behaviors as risky behaviors in this dissertation. Above all,
alcohol and cigarette use are the most commonly used illicit substances used by adolescents
(Fagbule et al. 2021; Shih et al. 2017; Olds and Thombs 2001). For example, 68.2% and 39.9%
of 12th graders from 1994-95 National Longitudinal Study of Adolescent to Adult Health (Add
Health) data reported that they had used alcohol and cigarettes respectively during the past year.
Even when adolescents were asked about their alcohol and cigarette use during the past month,

5
the prevalence rates were still substantial. In 2013 Social Networking Study data, 34.1% and
13.1% of 12th graders reported alcohol consumption and cigarette smoking respectively (Valente
et al. 2013; Huang et al. 2014). Considering that adolescence is a time of seeking more advice
and support from peers, it is not surprising that friendship relations are critical in influencing
alcohol and cigarette use among adolescents (Hussong 2002; Lundborg 2006; Shih et al. 2017;
Reed and Rountree 1997; Olds and Thombs 2001). Thus, this dissertation uses the Social
Networking Study data and Add Health data that include adolescent socio-demographic
information and substance use behaviors and their friendship networks.

Overview of Dissertation
In Chapter 2, I probe potential characteristics of adolescent brokers and school
environments where brokers may emerge in friendship networks. Considering benefits and costs
of a broker position, I hypothesize that social status of adolescents—student’s race/ethnicity,
socio-economic status, and types of extracurricular activities (ECAs) the student participates
in—is highly associated with holding a broker position. Moreover, I pay extra attention on the
effects of ECAs on the emergence of brokers. Because adolescents participate in different types
of ECAs depending on race or ethnicity, adolescents have different chances of occupying broker
positions depending on the type of ECAs they participate in and their races/ethnicities. So, I test
the second question whether and how extracurricular activities differently promote the
emergence of brokers depending on types of ECAs and student’s race or ethnicity. Drawing on
multilevel models from the Wave 1 datasets (64,412 students in 110 schools) of the Add Health,
I identify key characteristics of a broker and the findings suggest that minority or disadvantaged
students often serve crucial roles as brokers who connect classmates in schools. I also provide

6
evidence that ECAs help increase the probability of the emergence of brokers, but some
racial/ethnic groups tend to have a greater likelihood of holding broker positions depending on
the types of ECAs. These results indicate that ECAs help promote “status-leveling effects” that
mitigate racial/ethnic biases and stereotypes toward a particular race or ethnicity, and therefore
students from the group can gain status to hold a broker position.
In Chapter 3, I examine whether and how a broker position is associated with the
likelihood of alcohol and cigarette use. Because scholars claim that brokers can easily access
friends who use or have substances via their long ties that can reach out to many distant others, I
hypothesize that brokers have a high risk for alcohol and cigarette use. At the same time, their
long ties can also reach out to both substance using and non-using peers, and therefore I also test
the null hypothesis that there is no association between a broker position and likelihood of
alcohol and cigarette use. To examine “how” a broker shows high (or, no) risk for substance use,
I examine the number and directionality of a broker’s ties using contagion theory and the social
acceptance explanation. Because brokers need not only long ties that connect distant individuals
as far as possible but also wide ties that have multiple ties for adoption of risky behaviors
(Centola and Macy 2007), I test if the likelihood of alcohol and cigarette use changes depending
on the number of a broker’s ties. Considering instability and precarity of brokers’ ties with peers
(Burt 2002a; Granovetter 1973), some brokers who desperately seek social acceptance from
peers are willing to take risky behaviors, and therefore I test the directionality of a broker’s ties
on substance use to see if a broker who nominates many friends shows more alcohol
consumption and cigarette smoking. Using the Social Networking Study data (1,265 students in
five Southern California high schools) with the EV-brokerage measure (Valente et al. 2013;
Everett and Valente 2016; Huang et al. 2014), I find no association between the number and

7
directionality of a broker’s ties with the likelihood of alcohol consumption. However, the results
provide conditional support of social acceptance explanation in explaining the likelihood of
cigarette smoking.
In Chapter 4, I examine both structural and relational aspects of brokers on substance use
behaviors. Because brokers are connected to many distant others via their long ties, these indirect
ties allow greater chances for brokers to access substances or those who have access. On the
other hand, brokers’ immediate connections are diverse, and the diversity of their close
friendships may lead to greater exposure or access to substances or those who have or use
substances. To examine structural and relational aspect of brokers on substance use behaviors, I
ask three questions: 1) do adolescents show a high risk for drinking alcohol and smoking
cigarettes when they have long ties? 2) do adolescents show a high risk for drinking alcohol and
smoking cigarettes when they have heterogeneous ties such as cross-gender friendships? 3) are
there any different patterns of brokers’ substance use by gender? Using Wave 1 of the National
Longitudinal Study of Adolescent to Adult Health (61,608 students in 100 schools), I find
differences in the likelihood of alcohol and cigarette use for brokers’ long ties and heterogeneous
ties respectively. Although brokers’ heterogeneous ties—cross-gender friendships in this
dissertation—have stronger impact on their substance use behaviors than their long ties, the long
ties help explain the likelihood of heavy use of alcohol for both gender and male adolescents’
smoking. The findings also indicate that these associations of brokers’ long and heterogeneous
ties with substance use differ by gender.
In Chapter 5, I review the major findings and takeaways from this dissertation. The
summary helps describe the theoretical frames and findings of each chapter. I present the
contributions of this dissertation and limitations for future studies.

8
CHAPTER 2
WHO ARE THE BROKERS?
INDIVIDUAL STATUS AND EXTRACURRICULAR ACTIVITIES ON BROKERAGE
FOR ADOLESCENT FRIENDSHIP

Introduction
Scholars have long been interested in how heterogeneous friendships such as interracial
friendships are better formed as part of school desegregation efforts. For example, prior studies
have found that student’s racial or ethnic minority status (McPherson and Smith-Lovin 1987;
Quillian and Campbell 2003; Quillian and Redd 2009) and participation in extracurricular
activities (Moody 2001; Schaefer et al. 2011; Schaefer, Simpkins, and Ettekal 2018) are
associated with high levels of interracial friendship formations. However, the studies have
mainly examined micro-level friendships (e.g., friendships between dyads), leaving questions
about whether or to what extent such micro-level connections can lead to group- or macro-level
integration. Because a friendship network can remain segregated even if a number of students
report cross-race friendships with no connection between separate groups, examination of macro-
level integration in friendship networks is important.
In promoting connectivity between individuals or groups, sociologists have emphasized
the role of a broker—an intermediary who connects segregated individuals and groups (Burt
1992; 2002b; 2004; Granovetter 1973; Stovel and Shaw 2012). Because a broker has the shortest
path connecting many of the individuals, particularly those who used to be separated, in a
network, a broker serves the important role that reduces social distances between people or
groups in social networks. In spite of the importance of brokers to social integration, it is still
uncertain who serves in such important positions and what environments help individuals hold a
broker position (Stovel and Shaw 2012). Instead, previous studies have predominately paid

9
attention on how brokerage is advantageous to the brokers themselves (Brown and Konrad 2001;
Crowell 2004; Fleming, Mingo, and Chen 2007; Gould and Fernandez 1989; Mangino 2009).
Although no study has directly answered those questions, sociological studies suggest
that an individual’s status may be a pivotal characteristic for brokers. According to social capital
theory, high-status individuals tend to have more other types of capital, such as financial and
human capital, compared to low-status individuals, and therefore high-status individuals have
easier entry to advantageous social network positions, including that of broker (Briggs 2002;
McDonald 2011; Mouw 2009; Stanton-Salazar and Dornbusch 1995). On the contrary, high-
status individuals may prefer to form social closure with individuals of similar backgrounds,
because they get little or no reward from social interactions with dissimilar people, particularly
with low-status actors (Coleman 1988; Molm and Cook 1995; Molm, Schaefer, and Collett 2007;
Schaefer 2012). Instead, low-status actors, who perceive greater rewards from heterogeneous
ties, are more likely to strategically occupy broker positions. Although there are a few studies
that have addressed the characteristics of brokers (Barnes-Mauthe et al. 2015; Briggs 2002;
Cornwell 2009), their descriptive results fail to explain whether and how individual’s status
matters to the likelihood of holding a broker position.
Social contexts also matter to an emergence of a broker along with individual’s status.
For adolescents, participation in extracurricular activities (ECAs) can be important for promoting
brokerage because ECAs help develop micro-level integration such as interracial friendships
(Moody 2001; Schaefer et al. 2011; Schaefer, Simpkins, and Ettekal 2018). However, no
empirical study has tested the role of ECAs on brokerage or macro-level integration. More
importantly, it is also necessary to examine the effects of ECAs on brokerage depending on types
of ECAs (e.g. sports, arts, academics) or race/ethnicity of students, considering the

10
disproportionate demographic participation in particular ECAs, as well as the varied statuses
attached to different ECAs (B. B. Brown and Dietz 2009; Eder and Kinney 1995; Mahoney et al.
2009; McNeal Jr 1998; Quiroz 2000; Schaefer et al. 2011). For example, participation in sports
may help Black or Asian students have more brokerage power compared to other type of ECAs
because sport activities help mitigate racial biases or stereotypes toward these groups of students
and helps them gain status that helps access a network position connecting diverse and distant
individuals (Holland 2012; Ispa-Landa 2013). To expand our understanding of brokers, the
present study tests the hypotheses to examine how an individual’s status and ECAs matter to the
likelihood of holding a broker position in a social network.
This paper focuses on the role of an individual’s status and participation in ECAs on the
likelihood of holding a broker position among adolescents. Based on social capital theory, I
assume that there is a significant association between an individual’s status and the likelihood of
holding a broker position. Further, I hypothesize that participation in ECAs help promote
brokerage, but its effects may differ by types of ECAs and a student’s race/ethnicity. Drawing on
multilevel models from Wave 1 datasets (64,412 students in 110 schools) of the National
Longitudinal Study of Adolescent to Adult Health (Add Health), the present study identifies
adolescents’ friendship networks including brokerage power of each individual and related
socio-demographic characteristics. I expect that the findings on characteristics of potential
brokers and types of activities they participate in lead us to see if high-status is the key feature of
brokers. Moreover, the results indicate how participation in ECAs help adolescents hold broker
positions better. In the end of this paper, I discuss the theoretical and policy implications of the
results.

11
The Concepts and Measurements of a Broker in Social Networks
The concept of a broker has its origins in the work of Granovetter (1973). In “The
Strength of Weak Ties,” Granovetter did not directly use the term "broker," but he did engage
with the key idea that people who loosely connect acquaintances or dissimilar others tend to
receive different resources and information from each of the groups. In more recent studies, the
term "a broker" refers to a unique intermediary who connects separate individuals or groups in
the social structure (Burt 1992; Gould and Fernandez 1989; Stovel and Shaw 2012). This
conceptualization of a broker does not formally address that a broker’s ties also
connect dissimilar others, but there is an underlying assumption that two separate actors or
groups connected via a broker tend to be dissimilar based on the network homophily theory.
According to homophily theory, people are more likely to befriend or connect to similar others
and these connections are strong due to high frequency of contacts or interactions; on the other
hand, weak connections are mostly interactions with acquaintances or dissimilar others (Burt
1992; 2002b; 2004; Gould and Fernandez 1989; McPherson, Smith-Lovin, and Cook 2001b).
Borrowing the concept of brokers, a broker has two key dimensions: (1) a relational
dimension: a broker connects dissimilar individuals, and (2) a structural dimension: a broker is a
unique intermediary who links separate individuals or groups as far as possible, and therefore
reduces the distances between most of the individuals in networks or the entire network. Because
brokers connect diverse and separate people (or, groups) in social networks, individuals can
interact with different people and have direct and indirect access to others via brokers. In spite of
the importance of both dimensions of a broker on social integration, previous studies on brokers
have measured either the relational dimension or micro-level connections, but not both
simultaneously. For example, studies that use Granvotter’s concept of weak ties have often relied

12
on the tie strength between two individuals such as the emotional intensity or intimacy between
two individuals as their key measures without knowing if the weak ties can help reduce social
distances between people (D. W. Brown and Konrad 2001; Crowell 2004; Montgomery 1992;
Ryan 2000). A proportion of cross-race friendships between dyads is another measure that
identifies potential brokers who foster interactions between different others (Kao and Joyner
2004; Moody 2002; Briggs 2007). Both tie strength and cross-race friendships are good measures
of a broker’s relational dimension between two individuals (i.e., dyads), but these methods do
not necessarily detect the structural dimension or group-level connectivity. Because a network
may be still segmented by several separate groups if there is no group-level connection, it is
salient to examine the structural dimension in using the concept of a broker that can tell us
whether a broker connects not only neighboring individuals but also all or most of individuals in
a network.
For a structural aspect of a broker, Freeman introduced the formal, common measurement
of a broker’s structural dimension, betweenness centrality, that calculates the shortest path
connecting with all of individuals in a network (Freeman 1977). Although betweenness centrality
is one of the essential measures that identify the structural dimension of a broker, this measure is
sensitive to a size of a network and therefore difficult to compare betweenness centrality values
across different networks. Further, it also causes difficulties to distinguish between a broker and
a popular individual. A popular individual who has many connections has a high probability of
having a bridging tie among one of many connections, but it does not necessarily indicate that
the popular individual promote integration. Instead, a popular individual tends to maintain strong
relationships with similar others and often to interact with similar others via strong ties (Berger
and Dijkstra 2013; Eder 1985), and therefore it is necessary to utilize a metric that can

13
differentiate a broker from a popular individual.
Figure 2.1. Hypothetical Network with a Broker

To improve on the betweenness centrality metric, a recent study introduced the EV-
brokerage measure (Everett and Valente 2016). Although this measure does not directly examine
the relational dimension of a broker, it successfully identifies individuals who have the shortest
path lengths with most or all other individuals and controls for an individual’s number of
connections and network size. Thus, this method can distinguish between a broker and popular
individuals with a bridge. In Figure 2.1, nodes 2 and 3 will receive high EV-brokerage scores
compared to other nodes in the network, but node 2 will have a higher score because node 3’s
popularity is penalized in this calculation. The EV-brokerage measure also allows us to compare
broker scores across different networks, because this measure standardizes the effect of network
size on broker scores. Yet, this measure still relies on the assumption that a broker connects
dissimilar individuals without an examination of the relational dimension of a broker. Thus, I test
whether potential brokers identified by the EV-brokerage measure have heterogeneous
friendships prior to the main analysis.

Individual Status and Brokers
Prior studies have established the importance of brokers, especially the benefits for the

14
brokers themselves (Brown and Konrad 2001; Burt 2004; Crowell 2004; Fleming, Mingo, and
Chen 2007; Gould and Fernandez 1989; Mangino 2009), but little is known about who serves in
such important positions. Despite lack of discussion on this question, sociological theories hint
how an individual’s status is associated with the likelihood of holding a broker position. Social
capital theorists believed that theoretically low-status individuals can receive benefits and
compensate for low human or financial capital from social capital such as their relationships
(Coleman 1988; Coleman, Hoffer, and Kilgore 1982). However, the reality is that high-status
individuals who have more capital of other types than low-status individuals easily access
advantageous social relations and positions while low-status individuals are often left in
disadvantaged positions such as isolates or periphery members (DiMaggio and Garip 2012; Lin
1999; 2002; McDonald 2011; Mouw 2009). From this perspective, high-status individuals may
also be more likely to engage in broker positions than low-status individuals, considering
significant benefits of being a broker such as high accessibility of diverse information and
resources.
On the other hand, social exchange theory offers insight into why high-status actors
might not occupy in broker positions. The theory explains that individuals look for valued
rewards, which can be material goods or reputation, from their social interactions (Schaefer
2009). Although many people are eager to build social relations with high-status individuals,
high-status individuals tend to form social closure with other high-status individuals due to no or
little reward from interactions with dissimilar people, particularly with low-status individuals
(Molm and Cook 1995; Molm, Schaefer, and Collett 2007; Schaefer 2012). Contrarily, low-
status individuals perceive more gains from interactions with dissimilar people, particularly with
high-status individuals, than social closure with similar others. Thus, low-status individuals are

15
likely to engage in broker positions that often connect with dissimilar others.
Although there are many different indicators of an individual’s social status,
socioeconomic status (SES) is commonly used as a primary indicator of status in adolescent
studies (Finkelstein, Kubzansky, and Goodman 2006; Ensminger et al. 2000; Soteriades and
DiFranza 2003; Goodman et al. 2001). For adolescents, SES of their parents and/or families
basically serves an indicator of SES of adolescents, and it is often measured by parents’
educational attainment, income, or occupation. These measurements are easily constructed as a
hierarchical order; for example, parental education can be categorized from no high school
diploma, high school graduate, some college education, to college degree or higher. One of the
common indicators of social status is race/ethnicity. It is obvious that race/ethnicity is more a
perceived and subjective concept rather than objective, and a particular race or ethnicity does not
always connote high-status or low-status. However, it is also undeniable that status of
race/ethnicity is hierarchically placed even in adolescent peer relations (Branch, Tayal, and
Triplett 2000; Ghavami and Mistry 2019). In Ghavami and Mistry’s study, middle school
students viewed profiles of Whites as the highest, African American and Latino profiles as the
lowest regarding social class of their peers. In addition to SES and race/ethnicity, participation in
particular activities can be an indicator of high status among adolescents. For example,
adolescent studies reveal that athletic activities as high-status activities compared to academic or
art activities (B. B. Brown and Dietz 2009; Eder and Kinney 1995; Mahoney et al. 2009; McNeal
Jr 1998; Quiroz 2000; Schaefer et al. 2011).
In fact, a few studies also demonstrated characteristics of potential brokers, using SES
and race/ethnicity. For example, scholars have identified a significant relationship between
educational attainment (Barnes-Mauthe et al. 2015; Briggs 2002; Cornwell 2009) and broker’s

16
characteristics. And racial/ethnic minority status (Mouw 2002; 2009; Smith 2005) are likely to
associate with heterogeneous friendships as a number of studies suggested. Although these
studies have described characteristics of brokers, their descriptive findings failed to test above
conflicting hypotheses on whether and how high-status (or, low-status) is associated with the
likelihood of engaging in a broker position. To test the conflicting arguments, I examine how
adolescents engage in broker positions differently depending on their status in terms of a
student’s race/ethnicity, SES, and extracurricular activities.

Extracurricular Activities and Brokers
In the studies of friendship integration in schools, extracurricular activities (ECAs) are
often cited as the social spaces where diverse students can interact with each other due to
propinquity—physical, social and psychological proximity—and shared interest (McPherson and
Smith-Lovin 1987; McPherson, Smith-Lovin, and Cook 2001b; Moody 2001; Schaefer et al.
2011; Schaefer, Simpkins, and Ettekal 2018). While academic tracking is the practice of
grouping students based on academic abilities that often leads to maintain or exacerbate
segregation, ECAs attract students for many different reasons beyond academic performance
(Mickelson 2001; Tyson 2011). Thus, the demographic characteristics of ECA participants may
be more heterogeneous compared to those in an academic tracking program. In addition, ECAs
encourage students to achieve shared goals and cooperation across different groups of races and
SES, and these shared activities and goals help create and maintain heterogeneous friendships
among participants (Allport 1954). Although studies have revealed the salient role of ECAs on
friendship integration at a micro-level, (e.g., connections between dyads), it is not certain
whether ECAs also help lead to macro- or group-level integration due to no study conducted yet.

17
In terms of the role of ECAs on friendship integration, the impact of ECAs may differ by
each type of ECAs due to the different statuses attained by each type of activity. Schaefer et al.
(2011) argued that students who participated in sports tend to have more diverse friendships than
those who join performing arts or academics because of its high status of sports among
adolescents. Put differently, student athletes attract greater opportunities to interact with not only
participants but also non-participants compared to arts or academic ECAs. This can indicate
sport participants are likely to have more diverse friendships and therefore to hold broker
positions than those who participate in art or academic activities. However, this argument is a
conflicting claim that high-status activities may lead participants to secure their positions through
social closure with similar others as discussed previously.
More importantly, the effects of each type of ECAs may further differ by race/ethnicity
of students because each type of ECAs may help mitigate or exacerbate racial biases and
stereotypes differently depending on student’s race or ethnicity. It is called “status-leveling
effects”, explaining that certain types of ECAs helps reduce racial/ethnic stereotypes or biases
and therefore also reduce differential statuses between racial/ethnic groups (Hallinan and
Teixeira 1987). Recent studies have supported this idea of how sport involvement may create the
status-leveling effects especially for Black participants and how it helps integrate Black and
White male students in suburban schools (Holland 2012; Ispa-Landa 2013). Since playing sports
reinforces collective images of Black masculinity as cool and athletic, Black boys who are
involved in sports are more likely to gain status and to access interracial contacts. On the other
hand, participation in academic activities may not help promote the status-leveling effects for
Asian students, who are often viewed as non-athlete or academic overachiever, whereas
participation in sports may help (Lee 2015).

18
The impacts of ECAs on friendship integration may also differ by student’s race/ethnicity
because participation rates in each type of ECAs would be different by race/ethnicity of students.
In other words, each type of ECA provides different levels of propinquity opportunities
differently depending on student’s race/ethnicity. Although previous studies have often believed
that ECAs promote interracial/interethnic contacts due to diverse socio-demographic
backgrounds of participants unlike academic tracking, Schaefer and colleagues found that
demographic characteristics of ECA participants may not be diverse in certain types of ECAs
(Schaefer, Simpkins, and Ettekal 2018). For example, participation in sports, but not art or
academic activities, may give more propinquity opportunities to Asian students because Asian
students tend to join art or academic clubs more than sport activities. In addition, Hispanic
students are generally less likely to participate in ECAs (Elpus and Abril 2011; Ingels, Dalton,
and LoGerfo 2008; McNeal 1995; Meier, Hartmann, and Larson 2018). Thus, participation in
ECAs can be generally helpful to the formation of diverse friendships for Hispanic students.
Considering that each type of ECA may reduce or reinforce stereotypes or cultural biases toward
certain groups of race or ethnicity differently and different participation rates, the present study
examines the hypothesis that the effects of ECAs on brokerage may differ by each type of
activity and the effects also differ by students’ races/ethnicities.

Data and Measures
Data and Sample
I used Wave 1 of Add Health, a nationally representative study of U.S. adolescents in
grades 7 through 12 conducted in 1994-1995. Leveraging a stratified sampling design, Add
Health surveyed 90,118 students across 144 middle and high schools for the initial wave of in-

19
school data collection. Although Wave 1 was collected more than two decades ago, Add Health
is best suited for this study because it collects information on not only friendship networks but
also individual characteristics and school environments that still remain as key predictors in
friendship formation among middle and high school students.
I analyzed four datasets: (1) two individual-level datasets: in-school survey and
friendship nomination data, and (2) two school-level datasets: school information and school
administrator survey. Both the in-school survey and the friendship nomination data include all
90,118 students. After cleaning invalid IDs
1
, the cleaned datasets include 80,026 students within
141 schools. The school-level datasets provide a wide-ranging description of school
environments such as school racial composition, school size, and region. To avoid potential
biases when a small number of students represents the school population (Haas, Schaefer, and
Kornienko 2010; Schaefer et al. 2011), I included only schools with a response rate of 75 percent
or higher on the in-school survey. For the final sample after merging the four datasets, I have
information from 64,412 students within 110 schools.
Dependent Variable
The dependent variable is the EV-brokerage measure, which indicates the degree to
which a student holds a broker position in a friendship network (Everett and Valente 2016). The
EV-brokerage measure score is calculated in three main steps: 1) calculate standard node
betweenness,
2
2) double each score and add n-1 to all non-zero nodes where n is network size,
and 3) divide non-zero nodes by their degrees. Lastly, all nodes are divided by the possible

1
Among three IDs (student ID, survey questionnaire ID, school ID) in the datasets, I confirmed invalid values in
student IDs and survey questionnaire IDs that hindered connecting the datasets.
2
The standard node betweenness is calculated as 𝐶 𝑏 (𝑖 ) = ∑ (
𝑡 𝑗𝑖𝑘 𝑡 𝑗𝑘
)
𝑗 =𝑘 where 𝑡 𝑗𝑘
denotes the total number of shortest
paths in a (network) graph and 𝑡 𝑗𝑖𝑘 is the number of shortest paths connecting j to k that pass through vertex i.

20
maximum EV-brokerage score to normalize EV-brokerage by network size.
3
Thus, the EV-
brokerage score can be interpreted as the percentage chance of being the best broker within a
network, and the scores are comparable across different networks. For example, an EV-
brokerage score of 15 refers to a 15 percent chance of being a perfect broker who has the
possible maximum EV-brokerage score in the network. For the calculation the brokerage scores,
I used the R function calculate.EV.brokerage()
4
. Because the present study does not require
examining reciprocity or strength of ties, a tie direction is not considered in the analysis.
The Add Health sample in this study shows a range of 0 to 6.45 EV-brokerage scores.
Considering that friendship networks in the real world rarely happen to have a perfect broker,
and the Add Health counts only up to 10 friend nominations, this small range of the EV-
brokerage values is not surprising. To ease presentation of coefficients and standard errors, I
multiplied the EV-brokerage scores by 100, but this multiplication does not change any statistical
significance or leading outcomes.
Regardless of several advantages provided by the EV-brokerage measure, the measure
does not necessarily prove whether a broker identified from the measure has “dissimilar”
connection(s) as discussed above. Thus, I first examined the relationship between friendship
heterogeneity and brokerage identified by EV-brokerage measure. Since the present study
focuses on friendship integration across race or ethnicity, dissimilar connections or
heterogeneous friendships refer to friendships with different races or ethnicities. Once I
confirmed the significant relationship between the EV-brokerage scores and
interracial/interethnic friendships, I continued to use the EV-brokerage measure in multilevel

3
Each node is divided by (n
2
- 1)/4 if n is odd and (n
2
- 2)/4 if n is even number, where n refers to a network size.
4
R codes for the function are available upon a request with approval from the code’s author, Babak Mahdavi
Ardestani.

21
modeling.
Independent Variables
The key independent variables are 1) race/ethnicity, 2) socio-economic status (SES), and
3) participation in ECAs. First, I constructed a six-category the race/ethnicity variable, from a
self-reported questionnaire of Add Health as follows: 1) Non-Hispanic White, 2) Non-Hispanic
Black, 3) Hispanic, 4) Non-Hispanic Asian, 5) Non-Hispanic other-race and Native Americans,
and 6) Non-Hispanic multiracial.
SES is a composite measure based on two categories: 1) parents’ educational attainment
and 2) parents’ occupation, following McFarland et al.'s (2014) definition. Educational
attainment is segmented into five categories: 1) less than high school, 2) high school degree, 3)
some college, 4) college degree, and 5) graduate/professional degree. Occupation is segmented
into six categories: 0) not in labor force, 1) unskilled laborer, 2) skilled laborer, 3) white-collar-2,
4) white-collar-1, and 5) professional. Because mothers’ information is more readily available
than fathers’ information
5
, I primarily rely upon the mother’s SES, and only use the father’s SES
when the mother’s SES is unknown. Both education and occupation are ordinally categorized,
and therefore I summed these two categories to create the SES variable.
Third, Add Health asked adolescents whether they participated in any of 30 school-based
clubs, organizations, or activities during the school year. Based on prior studies (Schaefer et al.
2011; Schaefer, Simpkins, and Ettekal 2018), ECAs are sorted by three types of activities—
sports, arts, academics—and participation in each is coded as a separate binary variable
6
.

5
More than 40% of the sample does not have father’s SES, while more than 70% of the respondents report their
mothers’ SES.
6
If an adolescent participated in at least one of 12 sports clubs: cheerleading/dance team, baseball/softball,
basketball, field hockey, football, ice hockey, soccer, swimming, tennis, track, volleyball, and wrestling, I coded 1
for the sport binary variable. The art binary variable indicates the following four clubs: drama club, band,
chorus/choir, and orchestra. The academic binary variable is coded as 1 if an adolescent participated in at least one
of 14 academic clubs: French club, German club, Latin club, Spanish club, book club, computer club, debate team,

22
Control Variables and Multiple Imputation
Other student and school characteristics are included as control variables: 1) student-level
characteristics: gender, grade (grades 7 to 12), grade-point-average (GPA), network degree, and
2) school characteristics: racial/ethnic heterogeneity, average SES, type of school (public vs.
private), urbanicity, region, network density, and network size. I used gender and grade variables
from the original datasets with no alteration; I decided to use grade rather than age because grade
is more associated with friendship formations than age (Moody 2001). Student GPA is calculated
by computing the average of the student’s grades in four subjects: English/language arts, math,
history/social studies, and science (1=D, 4=A). The network degree refers to the sum of being
nominated and nominating others, calculated by degree() function in igraph R package. The
proportion of heterogeneous friends represents the number of interracial/interethnic friends
divided by the total number of friends.
I conducted multiple imputation to handle missing values on SES, GPA, race/ethnicity,
grade, and gender. Although I substituted fathers’ SES for mothers’ in 4,844 cases, about 20%
of the sample has no data in both parents’ SES (Appendix 2 Table A). About 13% of the sample
is missing GPA, and smaller percentages of students are missing data on race (2%), grade (<1%),
and gender (<1%). I also imputed three of the interaction terms (2% missing)—race/ethnicity
interacting with participation in sports, arts, and academics—prior to multilevel analysis. This
transform-then-impute
7
approach is recommended for interaction terms imputation to avoid
potential biases (Von Hippel 2009). In the imputation process, three variables are used as
auxiliary variables: number of female friends, number of male friends, and number of degrees

history club, math club, science club, honor society, newspaper, student council, and yearbook.
7
I transformed the interaction terms into a single variable. For example, the interaction term between race and sports
ECA is the product of the race and sports ECA variables. Then, I imputed these transformed variables.

23
because these auxiliary variables have no missing observations and a relatively high correlation
with imputed variables. I utilized multiple imputation by chained equations (MICE) with 10
imputation datasets. MICE is a better fit, than multivariate normal distribution (MVN), to impute
the variables because the variables consist of binary, categorical, and continuous variables and
MICE allowed me to use separate conditional distribution for each imputed variable (UCLA,
Institute for Digital Research & Education, Statistical Consulting 2020).
School race heterogeneity is calculated using Blau’s heterogeneity index (1−∑𝑝 i
2
),
where 𝑝 i is the proportion of group members in each of the 𝑖 categories (Blau 1977). A high
heterogeneity index indicates diverse racial/ethnic groups in a school. The school-level average
SES variable is the average of SES values of all students in each school. Type of school (public
vs. private), urbanicity (urban: reference, suburban, rural), region (South: reference, West,
Midwest, and Northeast) are from the Add Health questionnaires. Clustering coefficients indicate
the extent of network segregation, measured by transitivity() function from igraph R package.
Network density is the proportion of actual connections divided by the proportion of potential
maximum connections between students, measured by edge_density() function from the igraph R
package.

Analytic Strategy
Main Models
In the first stage of the analysis, I used correlation and bivariate regression to examine
whether EV brokerage scores are positively associated with heterogeneous friendships as
hypothesized. In the second stage of the analysis, I used hierarchical linear modeling (HLM)
where the student is level 1 and the school is level 2 (Raudenbush and Bryk 2002). Because

24
students are nested in schools and residuals of level 1 factors are likely to be clustered by school,
multilevel models are appropriate for this nested data. First, I constructed four models using
equation (1) that includes the key independent variables of individual’s status and individual-
and school-level control variables. To see any significant changes in effects of variables on
dependent variables, the first model is an unconditional, null model without covariates; the
second model includes only key independent variables; the third model adds individual-level
control variables; and the fourth model is the full model that adds school-level control variables.
𝐸𝑉 𝑏𝑟𝑜𝑘𝑒𝑟𝑎𝑔𝑒 𝑖𝑗
= 𝛾 00
+ 𝛾 10
𝑆 𝑝 𝑜𝑟𝑡 𝑖𝑗
+ 𝛾 20
𝐴𝑟𝑡 𝑖𝑗
+ 𝛾 30
𝐴𝑐𝑎𝑑𝑒𝑚𝑖𝑐 𝑖𝑗
+ 𝛾 40
𝐵𝑙𝑎𝑐𝑘 𝑖𝑗
+
𝛾 50
𝐻𝑖𝑠𝑝𝑎𝑛𝑖𝑐 𝑖𝑗
+ 𝛾 60
𝐴𝑠𝑖𝑎𝑛 𝑖𝑗
+ 𝛾 70
𝑂𝑡 ℎ𝑒𝑟 𝑟𝑎𝑐𝑒 𝑖𝑗
+ 𝛾 80
𝑀𝑢𝑙𝑡𝑖𝑟𝑎𝑐𝑖𝑎𝑙 𝑖𝑗
+ 𝛾 90
𝑆𝐸𝑆 𝑖𝑗
+
𝛾 𝑘 0
𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑖𝑗
+ 𝑢 0𝑗 + 𝑟 𝑖𝑗
(1)
To investigate differential effects of ECA by race and ethnicity of students, I added
interaction terms between extracurricular activities and race/ethnicity into the full models. For
this analysis, the three models rely on equation (2) and each model includes interaction terms
between race/ethnicity with sport, art, and academic activities, respectively. I conducted the
multiple imputation procedure and multilevel analyses using the STATA 14.1.
𝐸𝑉 𝑏𝑟𝑜𝑘𝑒𝑟𝑎𝑔𝑒 𝑖𝑗
= 𝛾 00
+ 𝛾 10
𝑆𝑝𝑜𝑟𝑡 𝑖𝑗
+ 𝛾 20
𝐴 𝑟 𝑡 𝑖𝑗
+ 𝛾 30
𝐴𝑐𝑎𝑑𝑒𝑚𝑖𝑐 𝑖𝑗
+ 𝛾 40
𝐵𝑙𝑎𝑐𝑘 𝑖𝑗
+
𝛾 50
𝐻𝑖𝑠𝑝𝑎𝑛𝑖𝑐 𝑖𝑗
+ 𝛾 60
𝐴𝑠𝑖𝑎𝑛 𝑖𝑗
+ 𝛾 70
𝑂𝑡 ℎ𝑒𝑟 𝑟𝑎𝑐𝑒 𝑖𝑗
+ 𝛾 80
𝑀𝑢𝑙𝑡𝑖𝑟𝑎𝑐𝑖𝑎𝑙 𝑖𝑗
+ 𝛾 90
𝑆𝐸𝑆 𝑖𝑗
+
𝛾 10_0
𝐵𝑙𝑎𝑐𝑘 𝑖𝑗
∗ 𝐸𝐶𝐴 (𝑆𝑝𝑜𝑟𝑡 , 𝐴𝑟𝑡 , 𝐴𝑐𝑎𝑑𝑒𝑚𝑖𝑐 )
𝑖𝑗
+ 𝛾 11_0
𝐻𝑖𝑠𝑝𝑎𝑛 𝑖 𝑐 𝑖𝑗
∗
𝐸𝐶𝐴 (𝑆𝑝𝑜𝑟𝑡 , 𝐴𝑟𝑡 , 𝐴𝑐𝑎𝑑𝑒𝑚𝑖𝑐 )
𝑖𝑗
+ 𝛾 12_0
𝐴𝑠𝑖𝑎𝑛 𝑖𝑗
∗ 𝐸𝐶𝐴 (𝑆𝑝𝑜𝑟𝑡 , 𝐴𝑟𝑡 , 𝐴𝑐𝑎𝑑𝑒𝑚𝑖𝑐 )
𝑖𝑗
+
𝛾 13_0
𝑂𝑡 ℎ𝑒𝑟 − 𝑟𝑎𝑐𝑒 𝑖𝑗
∗ 𝐸𝐶𝐴 (𝑆𝑝𝑜𝑟𝑡 , 𝐴𝑟𝑡 , 𝐴𝑐𝑎𝑑𝑒𝑚𝑖𝑐 )
𝑖𝑗
+ 𝛾 14_0
𝑀𝑢𝑙𝑡𝑖𝑟𝑎𝑐𝑖𝑎𝑙 𝑖𝑗
∗
𝐸𝐶𝐴 (𝑆𝑝𝑜𝑟𝑡 , 𝐴𝑟𝑡 , 𝐴 𝑐𝑎𝑑𝑒𝑚𝑖𝑐 )
𝑖𝑗
+ 𝛾 𝑘 0
𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑖𝑗
+ 𝑢 0𝑗 + 𝑟 𝑖𝑗
(2)

25
Results
Table 2.1. Descriptive Statistics for Students and Schools

Descriptive Results
As shown in Table 2.1, the mean value of EV brokerage is 26.14 (SD=29.45) with a
range of 0 to 644.58, which indicates a raw value of the range of 0 to 6.45. Approximately half
Variable
Dependent Variables Mean/% SD Min Max
EV-Brokerage 26.14 29.45 0 644.58
Independent Variables
Race/Ethnicity
Non-Hispanic White 55.66%
Non-Hispanic Black 16.09%
Hispanic 15.09%
Non-Hispanic Asian 4.96%
Non-Hispanic Other 2.40%
Non-Hispanic Multiracial 5.80%
Participation in Sports .55 .50 0 1
Participation in Arts .27 .44 0 1
Participation in Academics .32 .47 0 1
SES 5.44 2.65 1 10
Control Variables – Individual-level (n=64,412)
Total GPA 2.81 .80 1 4
Grade 9.53 1.63 7 12
Male .50 .50 0 1
Network Degree 8.60 5.60 0 43
% of Heterogeneous Friendships 28.8 34.9 0 100
Control Variables – School-level (N=110)
Race heterogeneity .46 .83 .14 .83
Average SES 5.31 .83 3.41 8.09
Private school (school type) .10 .30 0 1
Urbanicity
Urban 25.45%
Suburban 57.27%
Rural 17.27%
Region
West 14.55%
Midwest 23.64%
South 45.45%
Northeast 16.36%
Network Density 0.02 0.04 .001 .35
Network Size 585.56 455.27 25 2,550

26
the students are White (56%) and half are 16% Black, 15% Hispanic, 5% Asian, 2% other race
and Native Americans, and 6% Multiracial students. More than half of students participated in at
least one sport activity (55%), whereas roughly one third of the total students joined at least one
art-related (27%) or academic-related activity (32%). The average SES level is 5.44 (SD =2.65)
on a scale of 1 to 10, and the average total GPA is 2.81 (SD =.80). The mean grade is 9.53 and
half of the students are male. With regard to friendship networks, students have 8.6 friendship
ties with other students on average, either who nominated them or whom participant students
nominated, and roughly 30% of their friendships are cross-race/ethnic friendships. As for school
characteristics, a large proportion of the schools are public schools (90%) and are located in
suburban (57%) or in Southern regions (45%). School racial heterogeneity ranges widely from
.14 to .83, with average racial heterogeneity of .46. Network density and size vary by schools.
For example, some schools have very small network sizes or sparse density, the smallest being
25 students, whereas other schools have large networks and high density, the largest being 2,550
students.
Table 2.2. Relationship between Heterogeneous Friendships and Broker Students
(N=64,412)

I first tested the correlation between heterogeneous friendships and the EV-brokerage
measure as presented in Table 2.2. Although the proportion of heterogeneous friends does not
seem to have high correlation with EV-brokerage scores, p values indicate statistical significance
(p <.001). Results from a simple regression between heterogeneous friendships and EV-
brokerage measure also show a positive and statistically significant relationship (p <.001). Thus,
Pearson’s r ( ) p value
Coefficients
(SE)
z-value
(p value)
Percent of Cross-
race Friends***
.11 p <.001 9.12 (.34) 26.9 (p <.001)

27
I confirm that students with higher EV-brokerage scores are modestly more likely to have
heterogeneous friendships than those with lower EV-brokerage scores.

Multilevel Model Results: Students’ Status
The analyses provide several random-effects parameters that indicate the source of
variances, and a large portion of the variance in EV-brokerage values is due to differences within
schools. First, all the student- and school-level covariates accounts for 69.4% of variance at
student-level but only for 1.1% of variance at school-level, and 57.1% of total variance,
explained by the covariates, in EV-brokerage scores
8
. Intraclass correlation coefficients (ICCs)
also indicate that a large proportion of total variance is within schools; for example, the ICC of
the unconditional model is 0.82 whereas the ICC of the full model (Model 4) is 0.58. The
associations between EV-brokerage measure and primary predictors are presented in Table 2.3.
Model 1 (null model) shows the magnitude of variation in EV-brokerage values across different
schools. The result of the variance component (p<.001) reveals great variation in EV-brokerage
values, supporting the use of hierarchical linear modeling.
Table 2.3. Results of Multilevel Models Predicting EV-Brokerage Scores from ECA
Participation and Race/Ethnicity and SES of Students

8
I compare the variances of unconditional model (Model 1) and full model (Model 4); that is, the variance is
calculated by 1 – (variance in a full model/variance in unconditional model). First is 1-(25.19
2
/45.57
2
), the second
one is from 1-(21.26
2
/21.38
2
), and the last overall variance is from 1-(25.19
2
+ 21.26
2
)/(45.57
2
+ 21.38
2
). The
unconditional model is presented in Model 1.1.
Model 1 Model 2 Model 3 Model 4
Primary Predictors
Participation in sports - 0.45 (0.18)* -0.53 (0.18)** -0.54 (0.18)**
Participation in arts - 1.04 (0.20)*** 0.84 (0.20)*** 0.84 (0.20)***
Participation in academics - -0.17 (0.19) -0.31 (0.19) -0.31 (0.19)
Race (ref: White)
Black - 1.11 (0.31)*** 1.65 (0.32)*** 1.67 (0.32)***
Hispanic - 1.75 (0.30)*** 2.05 (0.30)*** 2.05 (0.30)***
Asian - -0.61 (0.45) -0.41 (0.45) -0.42 (0.45)
Other - 1.30 (0.57)* 1.54 (0.57)** 1.55 (0.57)**
Multiracial - 2.08 (0.38)*** 2.33 (0.38)*** 2.33 (0.38)***

28
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)
Model 2 includes only primary predictors, ECA participation, race/ethnicity, and SES.
Controlling for race/ethnicity and SES, participating in sports (p < 0.05) or arts (p < 0.001)
increases EV-brokerage scores, whereas participation in academics does not show any significant
impact on EV-brokerage scores. Except Asian students, racial/ethnic minority students tend to
have higher EV-brokerage scores than those of their White counterparts when controlling for
ECA participation and SES. Among racial/ethnic minority students, Hispanic and multiracial
students (p < 0.001) tend to have higher EV-brokerage values compared to those of Black (p <
0.001) or other-race (p < 0.05) students. On the other hand, SES is negatively associated with the
brokerage values (p < 0.001).
Model 3 adds student-level control variables and many of these control variables are
SES - -0.15 (0.03)*** -0.20 (0.03)*** -0.20 (0.03)***
Control Variables – level 1
Friend degree - - 0.38 (0.02)*** 0.38 (0.02)***
Male - - 1.41 (0.18)*** 1.40 (0.18)***
Grade - - -0.44 (0.08)*** -0.44 (0.08)***
Total GPA - - 0.19 (0.11)+ 0.19 (0.11)+
Control Variables – level 2
Race heterogeneity - - - 2.36 (16.17)
Average SES - - - -3.09 (3.33)
Private school - - - 17.71 (9.53)+
Region (ref: south)
Midwest - - - -7.72 (8.93)
West - - - 0.03 (8.28)
Northeast - - - -1.18 (9.23)
Urbanicity (ref: urban)
Suburban - - - -11.75 (6.16)+
Rural - - - -3.75 (8.66)
Network density - - - 367.59 (112.82)**
Network cluster - - - 241.90 (60.22)***
Intercept 46.98 (4.35)*** 46.70 (4.35)*** 46.88 (4.40)*** 8.79 (26.23)
√

11
45.57 45.53 45.30 25.19
√𝜃
21.38 21.36 21.26 21.26
ρ 0.82 0.82 0.82 0.58
Number of imputations 10 10 10 10
Sample size – students 64,412 64,412 64,412 64,412
Sample size - schools 110 110 110 110

29
statistically significant. As one level of friend degree increases, a student’s EV-brokerage score
increases 0.38 points (p < 0.001). Compared to female students, male students tend to have
higher chances of holding broker positions (p < 0.001). On the other hand, senior students are
less likely to have higher EV-brokerage scores compared to freshmen or students in lower grades
(p < 0.001). Total GPA is positively associated with EV brokerage scores, but only marginally
significant. Compared to Model 2, the relationships of all key predictors with EV-brokerage are
the same except participation in sports. The association between participation in sports and EV
brokerage becomes negative after controlling other student-level variables. This indicates that
students who join sport activities are likely to report lower EV brokerage scores than those of
non-sport-participants when their friend degree levels are identical according to Model 13 in
Table B of Appendix 2. Put differently, sport-participants tend to report higher EV-brokerage
scores due to numerous connections with other students, but participation in sports itself may
decrease the likelihood of holding a broker position when controlling for friend degree.
Model 4 contains both school-level and student-level control variables in addition to
primary predictors, rigorously examining attributes of potential brokers. As discussed in Model
3, the results propose that 1) art club members, 2) racial/ethnic minority students except Asian
students, and 3) those from low SES backgrounds are likely to hold broker positions. For
example, students who join art activities seem to have almost 10% more EV-brokerage points
compared to those who do not. Racial/ethnic minority students except Asian students tend to
report 18% to 27% more EV-brokerage points than those of their White counterparts
9
.
Additionally, EV-brokerage points decrease by 2% as one level of SES increases. Thus, this
result rebuts the hypothesis high-status is associated with attributes of potential brokers. Instead,

9
Other-race students tend to 18% points higher in EV-brokerage; Multiracial students tend to 27% points higher.

30
high-status individuals may not aim to be brokers and instead prefer to build social closure with
similar others. Minority- or disadvantaged-status students tend to have higher chances of holding
broker positions because they may seek to befriend dissimilar others more often than White or
high-status individuals who prefer to social closure.
In terms of school-level variables, network density and clustering have positive
associations with EV-brokerage scores. This finding indicates that students’ chances of holding
broker positions would increase in the friendship network that has sufficient connections
between students but fewer connections between groups. Although many of the school-level
factors are not significant, private schools and suburban schools have marginal significance on
the likelihood of holding broker positions (p < 0.10). For example, students in private schools
tend to have higher EV-brokerage points than those in public schools; and students in suburban
areas tend to have lower points compared to those in urban areas.
Table 2.4. Interaction Effects between Race/Ethnicity of Students and ECA Participation
Model 5 Model 6 Model 7
Primary Predictors – level 1
Participation in sports -1.42 (0.12)*** -0.54 (0.10)*** -0.53 (0.10)***
Participation in arts 0.42 (0.11)*** 0.10 (0.14) 0.44 (0.11)***
Participation in academics -0.45 (0.11)*** -0.46 (0.11)*** -0.65 (0.14)***
Race (ref: White)
Black 0.07 (0.19) 1.16 (0.16)*** 1.32 (0.16)***
Hispanic 0.65 (0.18)*** 0.88 (0.16)*** 0.98 (0.16)***
Asian -1.67 (0.27)*** -0.58 (0.24)* -0.88 (0.26)***
Other -0.66 (0.38)+ -0.31 (0.31) 0.03 (0.31)
Multiracial 1.64 (0.27)*** 1.96 (.23)*** 1.75 (0.23)***
SES -0.11 (0.02)*** -0.11 (0.02)*** -0.11 (0.02)***
Interaction terms
Race*Sport (ref: White*Sport)
Black in sport 2.38 (0.22)*** - -
Hispanic in sport 1.52 (0.23)*** - -
Asian in sport 1.70 (0.37)*** - -
Other-race in sport 1.73(0.52)*** - -
Multiracial in sport 0.93 (0.36)** - -
Race*Art (White*art)
Black in art - 0.24 (0.27) -
Hispanic in art - 2.47 (0.31)* -

31
Note: + p<0.1 * p<0.05 ** p<0.01 *** p<0.001 (two-tailed tests)

Multilevel Model Results: Interaction Effects
Table 2.4 presents the interaction effects between race/ethnicity and ECA participation to
reveal how each type of ECA differently associates with EV-brokerage values depending on the
race/ethnicity of students. Model 5 shows the different effects of participation in sports,
depending on students’ races and ethnicities. While participating in sports leads to a general
decrease in the EV brokerage scores, the result identifies benefits of joining sports for
Asian in art - -1.57 (0.45)*** -
Other-race in art - 2.12 (0.64)*** -
Multiracial in art - 0.39 (0.40) -
Race*Academic
(White*academic)

Black in academic - - -0.42 (0.26)
Hispanic in academic - - 1.27 (0.26)***
Asian in academic - - -0.06 (0.39)
Other-race in academic - - 0.58 (0.64)
Multiracial in academic - - 1.14 (0.40)**
Control Variables – level 1
Friend degree 0.74 (0.01)*** 0.57 (0.01)*** 0.57 (0.01)***
Male 1.31 (0.09)*** 1.32 (0.09)*** 1.33 (0.09)***
Grade -0.32 (0.04)*** -.33 (0.04)*** -0.32 (0.04)***
Total GPA 0.04 (0.05) 0.03 (0.05) 0.03 (0.05)
Control Variables – level 2
Race heterogeneity 1.43 (16.20) 1.73 (16.19) 1.68 (16.20)
Average SES -3.95 (3.34) -3.96 (3.34) -3.95 (3.34)
Private school 18.48 (9.55)+ 18.48 (9.54)+ 18.42 (9.55)+
Region (ref: south)
Midwest -6.93 (8.96) -6.97 (8.95) -6.93 (8.95)
West 0.15 (8.30) 0.16 (8.30) 0.20 (8.30)
Northeast -0.03 (9.25) -0.08 (9.24) -0.06 (9.24)
Urbanicity (ref: urban)
Suburban -11.07 (6.18)+ -11.05 (6.17)+ -11.05 (6.17)+
Rural -5.62 (8.68) -5.58 (8.67) -5.58 (8.67)
Network density 381.86 (112.79)*** 381.97 (112.74)*** 381.89 (112.76)***
Network cluster 243.79 (60.32)*** 243.36 (60.30)*** 243.46 (60.30)***
Intercept 10.95 (26.28) 10.67 (26.27) 10.42 (26.27)
√

11
25.27 25.26 25.26
√𝜃
22.04 22.04 22.04
ρ 0.57 0.57 0.57
Number of imputations 10 10 10
Sample size – students 64,412 64,412 64,412
Sample size - schools 110 110 110

32
racial/ethnic minority students in terms of the likelihood of holding broker positions. For all
racial/ethnic minority students who participate in sports, their mean brokerage values are
significantly higher than that of White counterparts. However, when they do not participate in
sports, some groups of minority students (e.g. Black, Asian, and Other-race students) have the
same or lower EV-brokerage scores than White students.
Model 6 presents the different effects of participation in art activities by race/ethnicity.
First, additional benefits of participation in arts are found among Hispanic and Other-race
groups. For example, the mean values of Hispanic art club participants are 2.47 points higher
than that of White counterparts, although Hispanic students who do not participate in arts
reported 0.88 higher brokerage points than White counterparts. Additionally, Other-race art club
participants reported 2.12 higher brokerage points, whereas Other-race students who do not
participate in arts do not differ significantly from White students. For Black, multiracial, and
Asian students, however, participation in arts seems to reduce the likelihood of holding broker
positions. The mean brokerage values for Black and multiracial students who do not join art
activities are 1.16 and 1.96 higher respectively than White counterparts, whereas Black and
multiracial art club participants’ mean brokerage score do not differ significantly from the White
peers. For Asian students, non-art participants have 0.58 lower brokerage points than White
students, and art participants’ brokerage scores are even lower: 1.57 points lower than their
White counterparts.
Model 7 identifies the different effects of academic activities on the likelihood of holding
broker positions, particularly the additional benefits of academic activities for Hispanic and
multiracial students. For example, the average brokerage values of academic participants among
Hispanic and multiracial students are, respectively, 1.27 and 1.14 higher than scores for White

33
academic activity participants. However, for Black students, participation in academic activities
does not help increase chances of holding broker positions; Black students who do not participate
in academics have 1.32 points higher than White counterparts, although there is no statistical
difference in brokerage points between Black and White academic participants. For Asian and
Other-race students, participation in academics does not lead to any significant impact on
chances of holding broker positions.

Discussion
Using the concept of a broker, the present study adds to the literature on how an
individual’s status explains a potential mechanism in adolescent friendship networks, the
emergence of brokers. The results demonstrate that high-status is not associated with attributes of
potential brokers. This finding challenges the “winners take all” argument in social networks
explaining that high-status individuals tend to hold advantageous positions such as brokers
(DiMaggio and Garip 2011). Instead, racial/ethnic minority students (except Asian students),
students from low SES family backgrounds, and students who do not participate in sports, but do
belong to arts clubs, are associated with characteristics of potential brokers. This may indicate
that high-status individuals do not aim to be brokers and instead prefer to build social closure
with other high-status individuals. As social exchange theory implies, high-status individuals
may perceive little benefits from a broker position, whereas minority or disadvantaged status
individuals may be willing to participate in the brokerage process due to potential benefits from
connecting with dissimilar others, such as high-status individuals. In spite of prior studies
suggesting that high status adolescents often possess beneficial network positions such as being a
center or leader (DiMaggio and Garip 2012; Lin 1999; 2002; McDonald 2011; Mouw 2009), the

34
current study suggests that an individual’s status may differently affect a mechanism for an
emergence of a broker.
Although gender and grades were not primary predictors in the analysis, the results also
show that male students tend to have higher brokerage scores than female ones, and freshmen are
likely to have higher brokerage scores compared to seniors. Considering the significant
associations between heterogeneous friendships and brokerage, it is not surprising that boys and
students in lower grades, who often reported higher levels of heterogeneous friendships, show
higher brokerage scores compared to girls and students in higher grades in the present study
(Hallinan and Teixeira 1987; McPherson, Smith-Lovin, and Cook 2001b; Morgan and
McPartland 1981). Furthermore, this association between heterogeneous friendships and
brokerage help explain why racial/ethnic minority students tend to have higher brokerage scores
than White ones.
However, it is puzzling that this pattern does not extend to Asian students who also
reported higher levels of heterogeneous friendships than White counterparts (Quillian and
Campbell 2003). In other words, Asian students are more likely than White students and as likely
as other students to have friends of other races/ethnicities, and yet it seems that they still may fail
to connect segregated groups or isolates. Though further examination is required, it seems that
Asian students may have different heterogeneous friendships than other racial/ethnic minority
groups. Figure 2.2 (Appendix 2) supports this speculation that Asian students’ heterogeneous
friendships have different characteristics. Specifically, Asian students in the present study
reported the lowest level of friendships with Black students compared to other groups.
Considering the long history of black-white segregation in American schools, friendships
connecting between black and white students, particularly befriending Black students, can allow

35
students to possess the key structural positions as brokers (Mickelson 2001; Mouw and Entwisle
2006; Quillian 2002; Quillian and Campbell 2003; Stearns, Buchmann, and Bonneau 2009).
The present study also supports that ECAs function as a social place for the emergence of
brokers through two potential mechanisms—a status-leveling effect and a propinquity process,
but these mechanisms may be differently shown by types of ECAs and by student’s race or
ethnicity. Put differently, the status associated with each type of activity differ by race/ethnicity
and students participated in sport, art, and academic activities at different rates. Even though
Black students tend to participate in sports more than other activities, sport activities are
beneficial to Black students to increase the probability of holding broker positions maybe
because of status-leveling effects. For Hispanic students, all of ECAs are helpful to increase their
probabilities of holding broker positions because their participate rates at all types of ECAs are
lower than other groups. This potential mechanism helps explain Asian students’ chances for
being brokers. Sport activities are helpful because the activities may give Asian students more
chances to interact with non-Asian students through sport activities because they are less likely
to join sports (McNeal 1995; Meier, Hartmann, and Larson 2018) and may also help mitigate
racial biases toward Asian students who are often viewed as non-athletes (S. J. Lee 2015). On the
other hand, art or academic clubs are not much helpful for Asian students on chances of
brokerage because Asian students tend to participate in these activities than sports and
participation in art or academic clubs may not help mitigate racial stereotypes toward Asian
students. Thus, the findings reinforce the idea that schools could still support integration in
adolescent friendship networks through ECA programs, but the programs should be carefully
designed considering that ECAs differently help students increase their chances of brokerage by
types of activities and student’s race/ethnicity.

36
The findings of this study suggest areas for future research. Although this study attempts
to provide two potential mechanisms in explaining different effects of ECAs by types of ECAs
and by student’s race/ethnicity, these mechanisms should be empirically and thoroughly
assessed. Using interview empirical evidence in previous studies, the explanations in this study
mostly address sport activities with Black and Asian students, but they still lack discussion of
why art and academic activities show different impacts to other racial/ethnic groups including
Other-race or multiracial students. For example, it is uncertain why Other-race students receive
an additional increase on their brokerage scores from participation in arts, and why multiracial
students show an increase, in their brokerage when they participate in academic activities under
two primary mechanisms. Further, it is also unclear whether and how academic activities differ
from academic tracking in terms of effects on brokerage, considering some academic activities
such as Honor society or student council are still selective rather than voluntary (Clotfelter
2002). Thus, future studies should examine to provide strong evidence on whether and how
participation in ECAs serves a social place through a propinquity process and status-leveling
effects differently depending on race/ethnicity.
Second, this study relies on cross-sectional data that shows adolescent friendship
networks at a single point in time. Because a broker’s ties are often weak and their decay rates
are quite high (Burt, Kilduff, and Tasselli 2013), brokers at time 1 may not be brokers at time 2
and vice versa. Fortunately, Add Health data provides the Wave 2 datasets that collected almost
15,000 of the same students one year after Wave 1. The availability of data at two different times
allows future studies to examine why and how brokers emerge and disappear in terms of
individual- and school-factors in more details and to allow further examinations such as different
types of brokers (e.g. strong and weak brokers) as Granovetter explained.

37
Lastly, future studies need to take a close look at school contexts or other individual-level
factors to better conceptualize the meaning of a high status among middle and high school
students. In the present study, high-status is defined as a racial/ethnic majority position (being
White), having a socio-economically advanced background (from high SES family
backgrounds), or engaging in sport activities. However, this definition of a high status may not
be the same in all schools. For example, being Black might be a high-status position in
predominately Black schools, or art/academic activities may be high-status activities in schools
where the school culture or practices emphasize such activities. Holland (2012) and Ispa-Landa
(2013) also show that sports may be a social place for gaining status only for boys, but not for
girls. Although only few school-level factors—private schools and suburban schools—show a
marginal significance on holding broker positions in this study, future studies may further
identify important school-level factors on brokerage. Thus, future studies may want to examine
other school contexts such as school-specific cultures and practices or the racial/ethnic diversity
of a student body and other individual-level characteristics that may affect the concept of a high
status among adolescents.

38
CHAPTER 3
WHEN BROKERS SEEK MORE FRIENDSHIPS:
EXAMINATION OF THE NUMBERS AND DIRECTIONALITY OF AN ADOLESCENT
BROKER’S FRIENDSHIP IN SOCIAL NETWORKS ON ALCOHOL AND
CIGARETTE USE

Introduction
Friendship relations are important in research on adolescent substance use. Several studies on
adolescent substance use behaviors have identified a significant association between adolescents’
network positions, particularly a broker position, and their risk for substance use (Henry and
Kobus 2007; Kreager and Haynie 2011; Kreager, Haynie, and Hopfer 2013; Osgood et al. 2014).
Osgood and colleagues (2014) found that liaisons—those who connect separate individuals or
groups as conceptually the same as brokers—were more likely to use marijuana than were non-
broker adolescents. Additionally, Henry and Kobus (2007) showed that brokers are at higher risk
for drinking and smoking compared to non-brokers, whereas liaisons in the study of Kreager and
colleagues reported high rates of drinking behaviors (Kreager and Haynie 2011; Kreager,
Haynie, and Hopfer 2013).
Although brokers are often referred to by different names such as liaisons, scholars
highlight the structural advantages of a broker position in explaining the association between a
broker and substance use. Because brokers connect distant individuals or groups, they have long
ties that can reach out to many distant others, including substance-using friends. Through their
long ties, brokers can easily access friends who use or have access to substances and this greater
exposure may lead to a higher chance that brokers use substances compared to isolates or non-
brokers. However, this rationale that brokers have a high risk for substance use remains unclear,
considering that their long ties can also reach out to both substance using and non-using peers. In

39
the same vein, other studies have also found that isolates, but not brokers, have a high risk for
cigarette smoking because of social isolation and fragile relations with peers among isolates
(Ennett and Bauman 1993; Fang et al. 2003).
Centola and Macy’s contagion theory (2007) supports a similar idea: that brokers may
not always engage in or transmit risky behaviors unless a special condition is met. Based on the
contagion theory, brokers need not only long ties that connect distant individuals as far as
possible but also wide ties that have many friendship ties for adoption and diffusion of risky
behaviors. Stated differently, brokers are more likely to participate in adoption or diffusion of
risky behaviors when they have “multiple ties with peers.” Because adoption or diffusion of
risky behaviors requires higher thresholds to activate the behaviors, brokers need multiple
affirmations from their peers that promote credibility and legitimacy of the behaviors and
eventually activate adoption and diffusion.
Along with the number of a broker’s ties, scholars have also identified significant
associations between the directionality of adolescents’ ties and risk for substance use. For
adolescents, the desire or pursuit of friendship formation may lead to an increased willingness to
try a risky behavior and therefore the person “who initiates or maintains friendships” would
impact the degree of peer influence. Previous studies have also provided results that adolescents
showed different levels of likelihood of adolescent substance use depending on directionality of
their friendship ties (Mathys, Burk, and Cillessen 2013; Pearson et al. 2006; Valente, Unger, and
Johnson 2005; Copeland et al. 2018). For example, those who reported high numbers of
outdegree—nominate others as friends—are likely to use substances because they tend to seek
status stability and peers’ acceptance. Considering that brokers’ ties with peers are often
precarious or weak (Burt 2002a; Granovetter 1973), some brokers may desperately seek social

40
acceptance and status from peers and these brokers may also be willing to take risky behaviors to
gain acceptance or status from peer groups.
Despite the importance of the number and directionality of a broker’s ties, these network
properties have rarely been examined in studies on broker position and the likelihood of
substance use. Although brokers serve a role of connecting distant people and reducing distances
between individuals in networks, their social ties are not identical. For example, some brokers
have very few connections with others, whereas other brokers have numerous friendship ties. In
addition, some brokers may receive lots of friendship nominations (i.e., high indegree), whereas
other brokers may nominate others as their friends (i.e., high outdegree) more than their
friendship nominations they received. Considering the possibility that brokers’ ties vary by
quantity and directionality, this study investigates whether and how the number and
directionality of a broker’s ties help explain the likelihood of a broker’s alcohol and cigarette
use.
To investigate the research questions, the present study uses the EV-brokerage measure
based on comparison with other measures. Although the EV-brokerage measure is the best tool
for this study, there are several brokerage measures available. For example, Granovetter (1977)
and Burt (1992, 2004) suggest the tie strength and constraint respectively as for brokerage
measures. Freeman (1978) also introduced a brokerage measure using the shortest path
calculation, referred it as “betweenness centrality.” More recently, Valente and Fujimoto (2010)
developed a VF-bridging measure that calculates the average changes in cohesiveness when each
of one’s ties is removed. Although each method has different pros and cons, I leveraged the EV-
brokerage measure, introduced by Everett and Valente (2016), as the best tool for this paper
because it calculates the shortest path with all others and controls for the potential effects of

41
personal network sizes.
Using the contagion and social acceptance perspectives with the EV-brokerage measure,
the present study examines whether and how an adolescent broker engages in alcohol and
cigarette use behaviors. Using the Social Networking Study data (Valente et al. 2013), I
construct logistic regression models to examine the likelihood of alcohol and cigarette use
among 12th-graders in five Southern California high schools. To test the contagion theory for a
broker’s substance use, I hypothesize that the likelihood of a broker’s substance use increases
when the number of a broker’s friends, including substance-using friends, increases. In addition
to the number of a broker’s friendship ties, I also explore the directionality of a broker’s ties.
Because risks and motives for alcohol and cigarette use are perceived differently among
adolescents, I examine alcohol consumption and cigarette smoking separately.

Previous Studies on Broker Position and Substance Use
An understanding of the relationship between friendships and substance use is crucial in
criminology and delinquency research. As network theories and data have become more widely
available, scholars have closely examined structural characteristics of adolescent friendship
networks on substance use. Several studies demonstrate that adolescents’ network positions,
particularly a broker position, in friendship networks are linked to likelihood of substance use
(Henry and Kobus 2007; Kreager and Haynie 2011; Kreager, Haynie, and Hopfer 2013; Osgood
et al. 2013). For example, Henry and Kobus (2007) found that brokers were more likely to use
tobacco and alcohol than isolates were. Brokers have long ties that enable them to reach out to
different and distant peers across networks including substance-using friends and therefore
engage in substance use due to greater exposure to substance-using friends than isolates or non-

42
brokers do. Osgood et al. (2014) also found that brokers’ connections with diverse groups
increase their chances of accessing illicit drugs such as marijuana.
Regardless of the positive association between a broker position and the likelihood of
substance use in previous studies, this long-tie explanation still seems to be insufficient for
understanding why brokers should show high risk for substance use. Theoretically, brokers’ long
ties may lead them to have greater exposure or access to substance-using friends as the previous
studies explained. However, these long ties can also allow brokers to connect with non-
substance-users who can reduce one’s risk for substance use. In this context, other scholars have
identified that isolates were at greater risk for smoking compared to brokers (Ennett and Bauman
1993; Fang et al. 2003), explaining that an isolate’s fragile and insecure status in peer groups
may result in a high risk for substance use. If lack of connections with peers indicates a fragile
and precarious status in peer groups, some brokers who have very few ties would also show a
high risk for substance use. On the other hand, popular brokers who have many ties, particularly
those who “receive” many friendship nominations from peers, are less likely to engage in
substance use due to their stable and high social standings in peer groups.
Although the number and direction of a broker’s ties may serve an important role in
understanding a potential mechanism in a broker’s substance use, the prior studies rarely
discussed these network properties. As briefly stated above, features of brokers may not be
identical regarding the number and directionality of their ties; instead, some brokers have only
two or very few ties, whereas other brokers have lots of ties with peers. Even if brokers have
several friendship ties, some of those ties can be indegree ties (i.e., being nominated by others)
and/or outdegree ties (i.e., nominating others). Considering the diversity of degrees and
directionality of brokers’ ties, it is necessary to investigate these network properties on a broker’s

43
substance use.
Theoretical arguments help support this notion that we need to consider the number of a
broker’s ties for understanding whether and how brokers show risky behaviors. Regarding the
number of a broker’s ties, Centola and Macy (2007) suggest that brokers are more likely to
engage in risky behaviors only when they have multiple ties with others. Because risky behaviors
require higher thresholds to be activated, brokers need many social ties that promote credibility
and legitimacy of the behaviors. Additionally, social network scholars have already identified the
significant relationship between the tie directionality and adolescent substance use (Almaatouq et
al. 2016; Copeland et al. 2018; C.-T. Lee et al. 2022; Mathys, Burk, and Cillessen 2013; Pearson
et al. 2006; Valente, Unger, and Johnson 2005; Vaquera and Kao 2008). For example, Copeland
et al. (2018) identified three different types of isolates depending on directionality of their ties
and these three types showed different levels of substance use. Thus, an inclusion of the number
and directionality of brokers’ ties in an examination of substance use may help better understand
a potential mechanism of whether and how a broker engages in substance use behaviors.

Numbers and Directionality of a Broker’s Ties and Substance Use Behaviors

Centola and Macy (2007) claim that brokers do not always engage in or pass on risky
behaviors across network. Instead, they suggest the contagion theory, explaining that a broker
needs a special condition to engage in or transmit risky behaviors. According to this theory, most
social behaviors follow the complex contagion process that requires higher thresholds for
activation and multiple exposures to sources. In other words, brokers may not engage in adoption
or diffusion of risky behaviors unless they have multiple exposures to risky behaviors or other
people who help activate risky behaviors regardless of their advantageous position in social
networks. Accordingly, brokers need to have not only “long” ties—connecting distant

44
individuals as far as possible—but also “wide” ties—having multiple connections with as many
ties as possible—to activate such contagions.
Although Centola and Macy focused mainly on diffusion of collective actions like high-
risk social movements, they provided several underlying mechanisms that help us understand
how adolescents may engage in substance use behavior. The mechanisms are 1) strategic
complementarity, 2) credibility, 3) legitimacy, and 4) emotional contagion. For substance use,
credibility and legitimacy mechanisms may help explain how adoption of substance use
behaviors involves complex contagion. Although substance use is often viewed as risky behavior
among adolescents, having several peers who use substances may lead adolescents to change
their view on substance use as being less riskier but more acceptable. Moreover, hearing about or
observing substance behaviors from different friends including socially distant contacts increases
credibility and legitimacy of the behaviors, leading brokers to adopt substance use behaviors.
Accordingly, the current study examines the potential association between the number of a
broker’s ties and the likelihood of substance use.
Considering the broker position’s multiple group memberships and fragile connections
with others, it is also important to consider “who want to initiate or maintain friendships” in
explaining a broker’s behavior changes. Based on this argument, brokers who want to befriend or
to maintain friendship ties are more willing to engage in risky behaviors because of their fragile
and precarious friendship relations. In fact, one of Burt’s studies, proposed the structural holes
and good ideas theory that highlights advantages of a broker position, also supports the idea of a
broker’s unstable and precarious relations with other individuals. Burt found high decay rates of
a broker’s ties over time (Burt 2002a); for example, his evidence shows that nine in ten of a
broker’s ties disappear the next year because of their unstable nature. Similarly, Granovetter

45
(1977) posited that a broker’s ties are “weak” ties, although he underscored beneficial aspects of
the tie weakness. These scholars did not argue that such tie weakness or instability necessarily
causes negative impacts for brokers, but Burt pointed out that brokers’ social capital eventually
dwindled as their ties become weaker or dissolved.
Although no study directly examines whether and how this unstable and precarious
nature of a broker position causes any negative consequences, scholars support the idea that
fragile connections with others may help explain a broker’s substance use behaviors. Henry and
Kobus (2007) suggested that brokers engage in substance use because they often experience role
strains derived from multiple group memberships and therefore they use substances to promote
social acceptance from their peer groups. In fact, studies on adolescents have also confirmed that
status seeking and social acceptance are one of the key drives. Allen, Porter, and McFarland
(2006) explained that when adolescents do not clearly know their social status or understand
their relationships with peers, “seeking validation” actions, including substance use behaviors,
helps them maintain their social status and relations. Other scholars also found that an
adolescent’s status in peer groups is highly associated with the likelihood of substance use
(Killeya-Jones, Nakajima, and Costanzo 2007). Particularly, adolescents who received relatively
few friendship nominations are more likely to smoke than are popular adolescents or those who
occupy central positions in the network (Ennett and Bauman 1993).
However, these results do not indicate that all brokers pursue social acceptance from
peers or an elevated status in peer groups. Although brokers’ ties are more likely to be unstable
and precarious than other friendship ties, some brokers may not need to seek status or social
acceptance. For example, for brokers who receive many friendship nominations, peer
nominations help validate their status and social acceptance and they therefore do not need to

46
engage in substance use to seek status or acceptance. On the other hand, when brokers nominate
others as friends but no or few peers nominate brokers, these brokers may be willing to seek
validation by drinking or smoking to bolster their status and peer acceptance. Assuming that
brokers’ ties have directionality and that such directionality indicates a level of demand for social
status and acceptance for adolescents, the current study tests whether indegree and outdegree ties
of brokers are significantly associated with the likelihood of their substance use.

Data, Measures, and Methods
Data and Sample
I used the fourth wave of the Social Network Study, a longitudinal network study of five
Southern California high schools (Huang et al. 2014; Valente et al. 2013). Dr. Valente and his
colleagues at the University of Southern California administered surveys during English class
(for 3 schools) or History class (2 schools). A total of four waves of Social Network Survey data
have been collected at approximately one-year intervals: 1) in wave 1, 2290 of 10th-grade
students participated in the survey in October 2010
10
; 2) in wave 3, 2016 of 11th-grade students
returned in December 2011; and 3) in wave 4, 1279 of 12th-grade students completed the survey
in April 2013. Because 12th-graders are likely to have more stable friendships and to report
higher rates of substance use than 10th-graders (Johnston et al. 2021; Poulin and Chan 2010), I
chose the fourth wave (12th-graders) to examine the likelihoods of substance use depending on a
number of a broker’s ties with peers and the directionality of their ties.
Social Network Study datasets contain not only friendship networks and respondents’
own substance use behaviors but also adolescents’ socio-demographic information and self-

10
Among 2290 students, 88.0% (2016 students) of the student returned valid parental consent forms and 1795
students completed surveys at time 1 and 1620 at time 2.

47
reported substance use information of their peers and family members. As for friendship
networks, adolescents in the survey nominated up to 19 best friends, although respondents were
generally asked up to five best friends in most network surveys. Considering a criticism that the
five-best-friends-nomination method may underestimate actual friendships, this option of
nominating 19 best friends can more accurately capture their actual friendship networks. As for
substance use behaviors, Social Network Study provides not only adolescents’ own substance
use experiences but also substance use behaviors of their peers and family members; for
example, students were asked about substance use behaviors of their own experiences and of
their family members such as siblings and parents. Researchers can also assess substance use
behaviors of their friends linking substance use information for each of an adolescent’s friends.
These details allow researchers to examine how friendship networks are associated with
adolescents’ substance use, controlling other factors including adolescent’s socio-demographic
characteristics and their peers and family substance use. Last but not least, the questions are also
comparable with national survey data on adolescent’s substance use such as the Monitoring the
Future (Johnston et al. 2021).
Dependent Variables: Alcohol Consumption and Cigarette Smoking
To analyze the relationship between brokerage and substance use, I used questions about
recent experiences of alcohol and cigarette use. Students reported the frequency of alcohol and
cigarette use during the past 30 days as an ordinal variable: “During the past 30 days, on how
many days did you smoke cigarettes?” with seven categories (1 – 0 days, 2 – 1 or 2 days, 3 – 3 to
5 days, 4 – 6 to 9 days, 5 – 10 to 19 days, 6 – 20 to 29 days, and 7 – all 30 days). Compared to
using the ordinal variable, a binary variable helps reduce complexity in multiple imputation and
leads to the same conclusions. Thus, I used a binary variable of alcohol and cigarette use where 0

48
indicates 0 days of drinking/smoking and 1 indicates 1 or more days of drinking/smoking during
the past 30 days. The likelihood of alcohol consumption and cigarette smoking were examined in
separated models.
Independent Variables: EV-Brokerage, and Numbers and Directionality of a Broker’s Ties
The primary independent variables for my research questions are brokerage scores, quantity of a
broker’s friendship ties, and directionality of a broker’s ties. To calculate these variables, I use
friendship nomination information and calculate them with a basic network analysis. Since each
student has been assigned a unique identification number, researchers are able to create ties
between students using the friendship nominations. Once the students and their ties are known, I
draw a social network that shows the number and directionality of ties between students.
I used the EV-brokerage measure (Everett and Valente 2016) to calculate and assign
brokerage scores. The EV-brokerage measure is the best tool for this study for several reasons
compared to other brokerage measures such as a tie strength or structural hole. For example,
Granovetter (1977) evaluated the strength of interpersonal ties and considered those who have
weak ties as brokers. Burt (1992, 2004) provided the constraint measure that identifies those who
fill a structural hole, a lack of a direct connection between two individuals or groups, as brokers.
Both methods emphasize the role of a broker in connecting across distant people or groups, but
these measures are at the local level and are unsuitable for examining larger level network effects
such as network structure or structural position in an entire network (Appendix 3 Table A). As
for the global level measures, Freeman’s betweenness centrality (1978) is one of the methods in
identifying brokers, using the shortest path calculation. Because this betweenness centrality
measure calculates paths between individuals in an entire network, we can identify those who
have the shortest paths connecting with all or almost all others across a network, also often called

49
long ties, as brokers. However, this measure is sensitive to network size and the number of ties
that a person has.
To fill the gap in existing brokerage metrics, Valente and Fujimoto (2010) proposed the
VF-bridging measure, which systematically deletes each tie of an individual and calculates
change in network cohesion. Then, an individual gets assigned an average value of the changes
for all ties. Taking the average value of changes for all ties mitigates the impact of having many
ties and this approach helps reduce the potential effect of the quantity of ties in calculating
brokerage power. More recently, the EV-brokerage measure, introduced by Everett and Valente
(2016), helps improve a few limitations of the VF-bridging measure to better identify brokerage
potential for certain types of networks (e.g., pendants). Considering that the current study aims to
identify those who have long ties as brokers, the EV-brokerage measure is the best method that
utilizes the shortest path calculation and controls for the potential effects of the quantity of ties
and network size.
The EV-brokerage measure indicates the degree to which a person holds a broker
position in a social network, and it is calculated as follows: 1) calculate standard node
betweenness
11
; 2) double each score and add n-1 to all non-zero nodes where n is network size
12
;
and 3) divide non-zero nodes by their degrees. Because it is difficult to interpret these scores
intuitively, I converted these raw EV-brokerage scores into normalized scores for each school,
named “EV-normalized score.” To convert raw EV-brokerage scores to EV-normalized scores, I
divided everyone’s raw EV-brokerage scores by the highest raw EV-brokerage scores in each

11
The standard node betweenness is calculated as 𝐶 𝑏 (𝑖 ) = ∑ (
𝑡 𝑗𝑖𝑘 𝑡 𝑗𝑘
)
𝑗 =𝑘 where 𝑡 𝑗𝑘
denotes the total number of shortest
paths in a (network) graph and 𝑡 𝑗𝑖𝑘 is the number of shortest paths connecting j to k that pass through vertex i.
12
Each node is divided by (n
2
- 1)/4 if n is odd and (n
2
- 2)/4 if n is an even number, where n refers to a network
size.

50
school and then multiplied by 100. The scores indicate percentages of brokerage power within
school; for example, 0 of the EV-normalized score means zero (i.e., the lowest) brokerage power
in school, while 100 of the EV-normalized score means the highest brokerage power in school.
As for a broker’s ties, there are two types of a broker’s ties to be examined: total number
of friends the broker has and the total number of friends who reported using alcohol/cigarettes.
To see these two types of ties, I created two interaction terms: 1) an interaction term between the
EV-normalized scores and a total number of friends (i.e., total degree), and 2) an interaction term
between the EV- normalized score and a total number of friends who reported alcohol or
cigarette use, network exposure. Because indegree (i.e., numbers of being nominated) and
outdegree (i.e., numbers of nomination) indicate directionality of friendship ties, I created two
interaction terms between the EV- normalized measures and a total number of in-nomination,
and the EV-normalized measure and a total number of out-nomination respectively.
Control Variables and Multiple Imputation
To rule out potential effects of students’ socio-demographic characteristics and substance
use of their peers and family on the likelihoods of substance use, I included them as control
variables: 1) race/ethnicity, 2) gender, 3) SES, 4) self-reported academic grades, 5)
extracurricular activity participation (sports, arts, academics), 6) peer substance use, 7) sibling
substance use, and 8) adult (parent) substance use. The survey offered 20 options for
race/ethnicity,
13
and I initially collapsed these 20 categories into 6 categories (e.g. Asian, Black,
Hispanic, Multiracial, Native American/Pacific Islander, and White). As explained above, most
of the students from the survey were Hispanic or Asian, and only a small percentage of them

13
1. American Indian/Alaska Native, 2. Asian, 3. Black/African American, 4. Central American, 5. Chicano or
Chicana, 6. Chinese, 7. Chinese-American, 8. Hispanic, 9. Japanese, 10. Japanese-American, 11. Korean, 12.
Korean-American, 13. Latino/Latina, 14. Mexican, 15. Mexican-American, 16. Native Hawaiian/Pacific Islander,
17. South American, 18. Vietnamese, 19. Vietnamese-American, 20. White

51
were from the other four groups. Therefore, I recoded the race/ethnicity variable with three
categories: Hispanic, non-Hispanic Asian, and non-Hispanic/non-Asian. I used the original
survey results with no change for the gender variable. For the SES control variable, I relied on
two main concepts, which are parents’ educational attainment and free lunch eligibility. Based
on parents’ educational attainment questionnaires, I created the parent education variable with
three categories: 1) less than high school, 2) high school diploma, and 3) college or higher. Since
the survey asked for the educational attainments of both the mothers and fathers, I used the
highest level of educational attainment between the mother or father. When only one parent’s
information was available I used a binary variable—whether students were eligible for free lunch
or not—from the survey results with no change. As for the academic grades variable, the original
survey question offered students nine options
14
and I collapsed these nine categories into four
categories: 1) Mostly A’s and B’s, 2) Mostly B’s and C’s, 3) Mostly C’s and D’s, and 4) Mostly
D’s and F’s. The survey also asked students whether they participated in any of 36 school-based
activities or clubs. Based on prior studies (Schaefer et al. 2011; Schaefer, Simpkins, and Ettekal
2018), these extracurricular activities (ECAs) are sorted by three types of activities ⎯sports, arts,
academics ⎯participation in each coded as a separate binary variable
15
.
To construct a variable of peers’ substance use, I used friendship network data and linked
information of peers’ substance use behaviors. Then, I counted a total number of friends who
used alcohol and cigarettes in the past 30 days. As for siblings’ substance use, I recorded whether

14
1. Mostly A’s, 2. Mostly A’s and B’s, 3. Mostly B’s, 4. Mostly B’s and C’s, 5. Mostly C’s, 6. Mostly C’s and D’s,
7. Mostly D’s, 8. Mostly F’s.
15
1. Sports: Badminton, basketball, baseball, cheerleading, cross country, dancing, exercises, football, hiking,
soccer, softball, swimming/diving, tennis, running/racing/track and field, volleyball, wrestling; 2. Arts: band,
chorus/choir, drama club, orchestra; 3. academics: book club, computer club, California scholarship federation
(CSF), debate team, French club, history club, honor society, key club, Latin club, math club, newspaper, science
club, Spanish club, student council, yearbook

52
or not the student’s sibling(s) consumed alcohol or smoked cigarettes once a month or more
using a binary variable. Similarly, I created a parent substance use variable that indicates 1 as
either or both their parent(s) consumed alcohol or smoked cigarettes weekly or more.
As briefly discussed in previous studies, the results of bivariate analyses support potential
associations between the control variables and substance use (Appendix 3 Table B and Table C).
For example, race/ethnicity has significant relationships with alcohol consumption (f = 27.94; p
<.001) and cigarette smoking (f = 4.26; p <.05) respectively. Cigarette use shows significant
differences by gender (t = 3.60; p <.001). Academic performance is inversely correlated with
substance use, not surprisingly. Participation in sports has positive associations with the
likelihood of alcohol use (p <.001), whereas participation in academics has negative associations
with the likelihood of cigarette use (p <.05). Additionally, the likelihood of alcohol and cigarette
use are positively associated with family members’ alcohol and cigarette use respectively.
Accordingly, I included these control variables into the logistic regression models to see main
effects of brokerage, numbers of a broker’s ties, and directionality of a broker’s ties on substance
use.
Prior to multiple imputation, I dropped 14 cases that have missing information in all of
network data, which are crucial variables in this study. Although the size of the missing cases is
small (1.09%) and many studies on adolescent health network excluded the missing cases
(Bearman, Moody, and Stovel 2004; Gile and Handcock 2017), I tested if excluding 14 cases
leads to any potential biases or different results. Based on Appendix 3 Table D, descriptive
statistics of 14 respondents are not significantly different from those of 1,265 respondents who
have network information. Furthermore, the main outcomes for logistic regression models are
also no major differences between the sample with excluding 14 cases and the sample with 14

53
cases).
Of 1,265 students, I manually imputed missing values in race/ethnicity, gender, parent’s
educational attainment, free lunch eligibility using data at earlier waves. Because characteristics
of these variables show no or small changes over time, using information at earlier waves (wave
1 and 2) can be one of the good sources in imputing missing values. After manual imputations, I
conducted multiple imputations to handle missing values on race, gender, parent’s educational
attainment, free lunch eligibility, GPA, peer substance use, sibling substance use, and parent
substance use. Even after manual imputation, about 0.8% of the sample had no data in parent’s
educational attainment and gender (Appendix 3 Table E). About 0.6% of the sample is missing
free lunch eligibility information and 2.6% of the sample is missing race/ethnicity. Even though
substance use information is highly sensitive, missing rates fall into between 3 to 8 percent. The
missing rates on peer substance use are about 4% and missing rates on adolescent’s substance
use are about 5 to 8%. In addition, missing rates on parent alcohol and cigarette use (~8%) as
well as sibling alcohol (~8%) and cigarette use (~6%) are smaller than 10%.
In the imputation process, three variables are used as auxiliary variables—two continuous
variables for sport and academic participation, and eigenvector centrality scores—because these
auxiliary variables have no missing observations with relatively high correlations with imputed
variables and are not used in the main analyses. Continuous variables for sport and academic
activity participation do not have missing values and highly correlated with several imputed
variables. Eigenvector centrality is one of the social network centrality measures that indicate the
degree to which a person is connected to “influencers/popular persons.” It often highly correlated
with brokerage scores and total degree values. I utilized multiple imputations by chained
equations (MICE) with 50 imputation datasets. MICE is a better fit than multivariate normal

54
distribution (MVN) to impute the variables because the variables for imputation consist of
binary, categorical, and continuous variables and MICE allowed me to use separate conditional
distribution for each imputed variable (UCLA, Institute for Digital Research & Education,
Statistical Consulting 2020).
Analytic Strategy
I analyzed the data with logistic regression models rather than multilevel logistic
regression models. Although the students are nested in schools and residuals of student-level
factors are likely to be clustered by school, the number of the schools (five schools) in the data is
too small to analyze using multilevel modeling and the models do not include any school-level
factors. Further, there are two additional pieces of evidence that supported logistic regression
models over multilevel models. First, standard errors were not significantly differences between
regular standard errors and clustered standard errors, indicating no issue in using OLS models.
Second, when I have constructed multilevel models and examined their Intraclass correlation
(ICC) values, the ICC values were close to 0 and this indicates very low variability at level 2
factors.
𝑙𝑜𝑔𝑖𝑡 {Pr (𝑆𝑢𝑏𝑠𝑡𝑎𝑛𝑐𝑒 𝑈𝑠𝑒 𝑖 = 1|𝑥 𝑖 )} = 
0
+ 
1
𝐵𝑟𝑜𝑘𝑒𝑟𝑎𝑔𝑒 𝑖 + 
𝑘 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑖 + 
𝑖
(1)
To examine a relationship between broker scores and substance use, I constructed a full
model that includes a brokerage score variable as a primary predictor and control variables, using
the equation (1). Because other primary predictors—numbers and directionality of a broker’s
ties—are interaction terms, the full models for these interaction terms used equation (2) that
contains not only the interaction terms but also the EV-broker measure and the number and
directionality of a broker’s ties. In the equation (1), the 
1
represents coefficients of the broker

55
scores variable in Model 1. In the equation (2), the 
1
represents coefficients of the interaction
terms—the number of a broker’s friendship ties, the number of a broker’s ties with those who
reported using substances, the number of a broker’s indegree ties, the number of a broker’s
outdegree ties—, 
2
represents coefficients of brokerage scores, and 
3
represents coefficients of
the number of friendship ties (e.g., numbers of friends, numbers of friends who reported using
substances, numbers of indegree, numbers of outdegree).
𝑙𝑜𝑔𝑖𝑡 {Pr (𝑆𝑢𝑏𝑠𝑡𝑎𝑛𝑐𝑒 𝑈𝑠𝑒 𝑖 = 1|𝑥 𝑖 )} = 
0
+ 
1
𝐵𝑟𝑜𝑘𝑒𝑟𝑎𝑔𝑒 ∗
# 𝑜𝑓 𝐹𝑟𝑖𝑒𝑛𝑑𝑠 ℎ𝑖𝑝 𝑇𝑖𝑒𝑠 (𝑡𝑜𝑡𝑎𝑙 𝑓𝑟 𝑖 𝑒𝑛𝑑𝑠 , 𝑠𝑢𝑏𝑠𝑡𝑎𝑛𝑐𝑒 − 𝑢𝑠𝑖𝑛𝑔 𝑝𝑒𝑒𝑟𝑠 , 𝑖𝑛𝑑𝑒𝑔𝑟𝑒𝑒 ,
𝑜𝑢𝑡𝑑𝑒𝑔𝑟𝑒𝑒 )
𝑖 + 
2
𝐵𝑟𝑜𝑘𝑒𝑟𝑎𝑔𝑒 𝑖 + 
3
𝐹𝑟𝑖𝑒𝑛𝑑𝑠 ℎ𝑖𝑝 𝑇𝑖𝑒𝑠 𝑖 + 
𝑘 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑖 + 
𝑖 (2)

Results
Descriptive Findings
Table 3.1 provides descriptive statistics about substance use behaviors, friendship
networks, and socio-demographic characteristics of the students. Although the students in this
study were from five different high schools, the prevalence rates of the students’ alcohol and
cigarette use are similar across schools and to the rates of the nationally representative sample
(Johnston et al. 2021)
16
. For example, 34% and 13% of 12th-grade students in this study
respectively reported that they had used alcohol and cigarettes during the past 30 days whereas
39% and 16% of a nationally representative sample in the 2013 Monitoring the Future survey
reported alcohol use and cigarette use during the past 30 days.
Table 3.1. Descriptive Sample Statistics Among 12th-grade students in Five Southern
California Schools: Social Network Study, Spring 2013 (n = 1,265)

16
According to data from the Monitoring the Future, 39.2%, 16.3%, and 36.4% of 12th-graders in 2013 reported the
use of alcohol, cigarette, and marijuana.
Variable

56

Table 3.1 shows that the sample has 13.95 mean scores (SD=18.15) of the EV-
normalized measure with a range of 0 to 100. This indicates that the average brokerage score is
low as 14% power although variation across brokerage scores (SD=18.15) is quite large. Because
of multiplication between numbers of friend ties and EV-normalized scores, the interaction terms
are bigger than 100. For example, the average number of friendship ties (i.e., total degree) is
eight, but the average value of a broker’s friendship ties is 148.6 derived from multiplying
Dependent Variables Mean/% SD Min Max
Alcohol use (binary) 34.11% 0 1
Cigarette use (binary) 13.10% 0 1
Independent Variables
EV-Brokerage scores 13.95 18.15 0.00 100.00
EV brokerage * Total degree 148.60 232.77 0.00 2100.00
EV brokerage * Alcohol using peers 35.84 63.46 0.00 600.00
EV brokerage * Cigarette using peers 13.78 37.43 0.00 590.06
EV brokerage * Indegree 62.64 102.72 0.00 1180.11
EV brokerage * Outdegree 85.96 161.13 0.00 1800.00
Total degree (numbers of friends) 8.00 4.68 0.00 36.00
User friends (alcohol use) 1.77 1.69 0.00 12.00
User friends (cigarette use) .67 .98 0.00 7.00
Indegree (numbers of being nominated) 2.73 2.81 0.00 19.00
Outdegree (numbers of nomination) 5.27 3.88 0.00 19.00
Control Variables
Race/Ethnicity
Hispanic 75.16%
Asian 23.80%
Non-Hispanic/Non-Asian 1.04%
Female 53.55% 0 1
Parent’s Education
No high school diploma 51.79%
High school graduate 21.74%
College or higher degree 26.47%
Free lunch eligibility 88.65% 0 1
Total GPA 3.16 0.79 1 4
Participation in Sports 44.11% 0 1
Participation in Arts 18.02% 0 1
Participation in Academics 28.46% 0 1
% of Peers who drink alcohol 30.77 28.30 0 100
% of Peers who smoke cigarette 14.18 20.72 0 100
Sibling alcohol use 37.43% 0 1
Sibling cigarette use 17.10% 0 1
Adult alcohol use 43.97% 0 1
Adult cigarette use 25.14% 0 1

57
between the number of friendship ties and the EV-normalized scores. As mentioned, the average
number of friends is eight (SD=4.68), but the average number of friends who reported using
alcohol is 1.77 (SD=1.69) and those who reported smoking cigarettes is .67 (SD=.98). The table
also shows that the average numbers of indegree and outdegree do not exceed 6 although the
students were able to nominate up to 19 best friends; for example, the mean value of indegree is
2.73 (SD = 2.81) and of outdegree 5.27 (SD = 3.88), indicating that students nominated more
than five others as their friends, whereas students received about three nominations as friends.
Many of the students in this study were either Hispanic or Asian. Almost three-quarters
of the respondents (75%) are Hispanic and one-quarter of the respondents (24%) are non-
Hispanic Asian. The male to female ratio is about even (54% of female). About one quarter of
the students have parent(s) who earned a college degree, whereas more than half of them
reported their parent(s) did not have a high school diploma. 44% of the students participated in
sport activities, whereas 18% and 28% of the survey respondents reported participation in art and
academic clubs respectively. As for the prevalence rates of substance use among family
members, alcohol and cigarette use rates of parents were higher than those of siblings, and both
groups used alcohol more than cigarettes; for example, 44% of students reported that their
parents drank alcohol, while 37% of the students reported that their siblings did. On the other
hand, 17% of students answered that their sibling(s) used cigarettes, which is half the rate of
alcohol use, and 25% of students responded that their parent(s) smoked cigarettes.
Broker’s Total Numbers of Friends on Alcohol and Cigarette Use
Table 3.2 presents the results of logistic regression models predicting the likelihood of a
broker’s alcohol use. In Model 1, there is no significant effect of brokerage and numbers of
friends on alcohol consumption, indicating that neither the brokerage scores nor numbers of

58
friends explain differences in likelihood of alcohol consumption. Model 2 also shows no
significant effect for the interaction term between brokerage scores and the number of friends on
alcohol consumption. This means that the numbers of a broker’s friends are not statistically
related to likelihood of alcohol consumption. However, the marginal effects of the numbers of a
broker’s friends on alcohol use provide further details on this relationship. Figure 3.1 presents
the different pattern between low-brokerage groups and high-brokerage groups. For the high-
brokerage groups or potential brokers, they seem to drink less alcohol as their numbers of friends
increase; however, the opposite tendency is found in the low-brokerage groups. For the low
brokerage groups, they are more likely to drink alcohol as they have more friends.
Table 3.2. Logistic Regression Models of Alcohol Use: Broker vs. Broker’s Friend Ties
(N = 1,265)
Model 1 Model 2
Dependent Variable: Alcohol Use
Primary Predictors OR 95% CI OR 95% CI
Broker scores 1.00 .99−1.01 1.01 .99−1.02
Broker * Total degree - - 1.00 1.00−1.00
Total degree (# of friends) - - 1.00 .96−1.05
Control Variables
Female 1.08 .82−1.42 1.08 .82−1.42
Race (ref: Hispanic)
Asian .38*** .26 −.62 .39*** .26−.58
Non-Asian/Non-Hispanic 1.69 .65−4.41 1.67 .64−4.38
Parent’s edu (ref: No HS)
High school graduate 1.00 .71−1.41 1.00 .71−1.41
College or higher degree .98 .70−1.37 .98 .70−1.38
Free-lunch eligibility .99 .64−1.54 .98 .63−1.53
Total GPA (ref: D’s and F’s)
Mostly C’s and D’s .48 .18−1.28 .48 .18−1.27
Mostly B’s and C’s .42+ .16−1.10 .42+ .16−1.10
Mostly A’s and B’s .33* .12 −.86 .33* .12−.86
Participation in sports 1.43* 1.08 −1.88 1.43* 1.08−1.88
Participation in arts .82 .57−1.19 .82 .57−1.18
Participation in academics .82 .59−1.15 .83 .59−1.16
Sibling alcohol use 1.49* 1.10 −2.03 1.49* 1.10−2.03
Sibling cigarette use 1.28 .86−1.90 1.27 .86−1.89

59
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

Figure 3.1. Comparison of Probability of Drinking Depending on Numbers of Ties Between
Low- and High-brokerage Groups

Table 3.3 presents the likelihood of cigarette use depending on brokerage scores and
numbers of peers. As with the likelihood of alcohol consumption in Table 3.2, brokerage scores
(OR = 1.01, 95% CI = .99-1.02; Model 4) and number of friends (OR = .98, 95% CI = .92-1.03;
Model 4) do not seem to have significant impact on the likelihood of cigarette use. This means
that students with high brokerage scores are not significantly more or less likely to smoke
Adult alcohol use 1.60** 1.21 −2.12 1.61** 1.21−2.13
Adult cigarette use 1.21 .88−1.67 1.21 .88−1.67
Intercept .86 .31−2.44 .86 .29−2.50
Overall model evaluation
df

17 19
Pseudo R
2
.080 .082

2

5.18*** 4.68***

60
cigarettes compared to those with low brokerage scores. The number of friends also does not
substantially impact the likelihood of cigarette use, indicating that those who have lots of friends
do not necessarily engage more in cigarette use than those who reported having few or no
friends. Furthermore, there is no main effect of the interaction term between brokerage scores
and number of friends on cigarette use. Similarly, no significant difference is found between
low- and high-brokerage groups even when examining marginal effects of the interaction terms.
Table 3.3. Logistic Regression Models of Cigarette Use: Broker vs. Broker’s Friend Ties
(N= 1,265)
Model 3 Model 4
Dependent Variable: Cigarette Use
Primary Predictors OR 95% CI OR 95% CI
Broker scores 1.01 1.00-1.02 1.01 .99-1.02
Broker * Total degree - - 1.00 1.00-1.00
Total degree (# of friends) - - .98 .92-1.03
Control Variables
Female .57** .38-.85 .57** .39-.86
Race (ref: Hispanic)
Asian .58+ .32-1.04 .57+ .32-1.03
Non-Asian/Non-Hispanic 1.15 .24-5.55 1.20 .25-5.84
Parent’s edu (ref: No HS)
High school graduate 1.46 .91-2.33 1.46 .91-2.34
College or higher degree 1.39 .88-2.20 1.40 .89-2.22
Free-lunch eligibility .69 .40-1.22 .71 .40-1.24
Total GPA (ref: D’s and F’s)
Mostly C’s and D’s 1.22 .37-4.02 1.21 .37-4.00
Mostly B’s and C’s 1.02 .32-3.29 1.03 .32-3.32
Mostly A’s and B’s .40 .12-1.35 .40 .12-1.35
Participation in sports 1.06 .72-1.56 1.06 .72-1.56
Participation in arts .81 .47-1.40 .81 .47-1.40
Participation in academics .53* .30-.92 .55* .31-.96
Sibling alcohol use .99 .63-1.56 1.00 .63-1.57
Sibling cigarette use 2.60*** 1.72-4.23 2.58*** 1.58-4.22
Adult alcohol use 1.08 .72-1.63 1.09 .72-1.63
Adult cigarette use 2.03*** 1.33-3.08 2.03*** 1.34-3.09
Intercept .19* .05-.70 .22* .06-.82
Overall model evaluation
df

17 19
Pseudo R
2
.12 .137

61
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

Although there is no main effect of brokerage and numbers of friends on substance use, I
found a few control variables—GPA, extracurricular activities, family members’ substance
use—that statistically associate with the likelihoods of alcohol and cigarette use. First,
academically high-performing students were less likely to drink alcohol compared to those
whose GPAs were mostly D’s or Fs. Additionally, participation in sports and academics seems to
affect the likelihood of substance use. For likelihood of alcohol consumption, sport participants
are 43% more likely to use alcohol compared to non-participants (95% CI = 1.08-1.88; Table
3.2) and, participants in academic activities are about 45% less likely to smoke (OR= .53, 95%
CI = .30-.92 in Model 3; OR = .55, 95% CI = .31-96 in Model 4 in Table 3.3). Parent’s and
sibling’s substance use also matters. For example, the likelihood of alcohol consumption
increases by 49% and 61% when students’ sibling(s) (95% CI = 1.10-2.03, p <.05; Model 2 in
Table 3.2) and parent(s) consume alcohol (95% CI = 1.21-2.12, p <.01; Model 2 in Table 3.2).
And, adolescents are 2.5 and 2 times more likely to smoke when their sibling(s) (95% CI = 1.58-
4.22, p <.001; Model 4 in Table 3.3) and parents (95% CI = 1.34-3.09, p <.001; Model 4 in Table
3.3) respectively also smoke and parent(s).
Broker’s Substance-Using Peers on Likelihood of Alcohol and Cigarette Use
Table 3.4 displays the likelihood of alcohol consumption depending on the number of a
student’s friends who reported using alcohol and cigarettes. In Model 5 and Model 6, the results
show no significant main effect of numbers of substance-using friends on alcohol consumption
and no interaction effects between brokerage and substance-using friends. These results mean
that having many substance-using peers does not necessarily have significant (linear)
relationship with the likelihood of alcohol consumption for students and even brokers. However,

2

4.87*** 4.40***

62
the marginal effects of the interaction terms present that there are different patterns on alcohol
consumption between low-brokerage and high-brokerage groups. Based on Figure 3.2, those
whose brokerage scores are low are more likely to drink alcohol as they have more friends who
drink alcohol. On the other hand, high-brokerage group or potential brokers are less likely to
drink alcohol as their numbers of friends who drink alcohol. Figure 3.3 also supports this
tendency in the interaction term between brokerage and the number of friends who smoke
cigarettes. Overall, having substance-using peers may lead to increase the likelihood of alcohol
consumption for low-brokerage groups, whereas high-brokerage groups or potential brokers are
less likely to drink alcohol as they have more substance-using friends.
Table 3.4. Logistic Regression Models of Alcohol Use: Broker’s User Friends Who
Reported Alcohol and Cigarettes (N=1,265)
Model 5 Model 6
Dependent Variable: Alcohol Use
Primary Predictors OR 95% CI OR 95% CI
Broker * user friends
(alcohol)
1.00 .99-1.00 - -
Broker * user friends
(cigarette)
- - 1.00 .99-1.00
Broker scores 1.01 1.00-1.02 1.00 .99-1.01
# of user friends (alcohol) 1.04 .92-1.18 - -
# of user friends (cigarette) - - 1.00 .81-1.24
Control Variables
Female 1.07 .81-1.41 1.08 .82-1.43
Race (ref: Hispanic)
Asian .38*** .26-.57 .39*** .26-.58
Non-Asian/Non-Hispanic 1.64 .62-4.31 1.69 .64-4.44
Parent’s edu (ref: No HS)
High school graduate 1.01 .72-1.43 1.00 .71-1.41
College or higher degree .97 .69-1.37 .95 .70-1.37
Free-lunch eligibility .98 .63-1.52 .98 .63-1.53
Total GPA (ref: D’s and F’s)
Mostly C’s and D’s .48 .18-1.29 .48 .18-1.28
Mostly B’s and C’s .43+ .17-1.12 .43+ .16-1.10
Mostly A’s and B’s .33* .13-.87 .33* .12-.86
Participation in sports 1.43* 1.08-1.88 1.43* 1.08-1.88
Participation in arts .82 .57-1.19 .82 .56-1.18

63
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)
Figure 3.2. Comparison of Probability of Drinking Depending on Numbers of Peers Who
Drink Alcohol Between Low- and High-brokerage Groups

Participation in academics .82 .58-1.15 .82 .59-1.15
Sibling alcohol use 1.50** 1.10-2.04 1.49* 1.09-2.02
Sibling cigarette use 1.28 .86-1.89 1.28 .86-1.89
Adult alcohol use 1.61*** 1.22-2.14 1.60** 1.20-2.12
Adult cigarette use 1.20 .87-1.66 1.21 .88-1.68
Intercept .82 .29-2.35 .87 .31-2.48
Overall model evaluation
df

19 19
Pseudo R
2
.083 .082

2

4.69*** 4.65***

64
Figure 3.3. Comparison of Probability of Drinking Depending on Numbers of Peers Who
Smoke Cigarettes Between Low- and High-brokerage Groups

Table 3.5 shows the relationship between likelihood of smoking and the numbers of
friends who reported drinking and smoking. Not surprisingly, the numbers of substance-using
friends seem to be an important factor in the likelihood of smoking. However, having more
substance-using peers leads to a decreased likelihood of smoking; for example, a student is 28%
less likely to smoke when the numbers of friends who smoke increase (OR = .72, 95% CI = .52-
1.00, p <.05). Similarly, a student is 14% less likely to smoke when a student has an increase of
the number of friends who consume alcohol (OR = .86, 95% CI = .72-1.02, p <.1). Additionally,
there is no significant interaction effect between brokerage and the number of substance-using
friends. The examination of marginal effects also supports this finding; as the number of
substance-using peers increase, the likelihood of smoking cigarettes decreases for both low- and

65
high-brokerage groups. In other words, having more substance-using peers generally serves the
deterrent effect in smoking cigarettes, but these effects are not significantly different between
non-brokers and brokers.
Table 3.5. Logistic Regression Models of Cigarette Use: Broker’s User Friends Who
Reported Alcohol and Cigarettes (N=1,265)
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)
Model 7 Model 8
Dependent Variable: Cigarette Use
Primary Predictors OR 95% CI OR 95% CI
Broker * user friends
(alcohol)
1.00 1.00-1.01 - -
Broker * user friends
(cigarette)
- - 1.00 1.00-1.01
Broker scores 1.01 .99-1.02 1.01 1.00-1.02
# of user friends (alcohol) .86+ .71-1.02 - -
# of user friends (cigarette) - - .72* .52-1.00
Control Variables
Female .57** .38-.85 .57** .38-.85
Race (ref: Hispanic)
Asian .55* .31-.99 .56+ .31-1.01
Non-Asian/Non-Hispanic 1.25 .26-5.99 1.42 .30-6.83
Parent’s edu (ref: No HS)
High school graduate 1.48 .92-2.38 1.47 .92-2.37
College or higher degree 1.42 .90-2.25 1.40 .88-2.22
Free-lunch eligibility .70 .40-1.23 .71 .40-1.25
Total GPA (ref: D’s and F’s)
Mostly C’s and D’s 1.24 .38-4.09 1.15 .35-3.77
Mostly B’s and C’s 1.05 .33-3.40 .99 .31-3.19
Mostly A’s and B’s .41 .12-1.39 .38 .11-1.28
Participation in sports 1.04 .71-1.53 1.05 .71-1.54
Participation in arts .78 .45-1.35 .77 .44-1.34
Participation in academics .56* .32-.97 .56* .32-.97
Sibling alcohol use .99 .63-1.56 .97 .61-1.53
Sibling cigarette use 2.52*** 1.54-4.12 2.57*** 1.57-4.20
Adult alcohol use 1.07 .71-1.61 1.07 .71-1.62
Adult cigarette use 2.08*** 1.37-3.17 2.11*** 1.38-3.23
Intercept .23* .06-.84 .23* .06-.85
Overall model evaluation
df

19 19
Pseudo R
2
.13 .13

2

4.46*** 4.54***

66
Broker’s Indegree and Outdegree on Alcohol and Cigarette Use
Table 3.6 presents the analysis of the main effects of the number of indegree and
outdegree ties and its interaction term with brokerage scores on alcohol consumption. Neither the
indegree nor outdegree shows a significant effect on the likelihood of alcohol consumption. This
finding means that any change in indegree or outdegree ties does not lead to significant
differences in likelihood of a student’s alcohol consumption. The results also show no interaction
effect between brokerage and indegree/outdegree ties, indicating that a broker’s indegree or
outdegree ties do not substantially matter to the likelihood of alcohol consumption. However,
brokerage scores seem to be important in understanding alcohol consumption. Students are more
likely to consume alcohol as their brokerage scores increase, controlling potential impacts of
other variables including indegree or outdegree (OR = 1.01, 95% CI = 1.00-1.02, p<. 05; Model
9).
Table 3.6. Logistic Regression Models of Alcohol Use: Broker’s Indegree vs. Broker’s
Outdegree (N= 1,265)
Model 9 Model 10
Dependent Variable: Alcohol Use
Primary Predictors OR 95% CI OR 95% CI
Broker * Indegree 1.00 .99-1.00 - -
Broker * Outdegree - - 1.00 1.00-1.00
Broker scores 1.01* 1.00-1.02 1.00 .99-1.01
Indegree .99 .92-1.06 - -
Outdegree - - 1.01 .97-1.06
Control Variables
Female 1.08 .82-1.43 1.08 .82-1.42
Race (ref: Hispanic)
Asian .38*** .25-.57 .39*** .26-.58
Non-Asian/Non-Hispanic 1.73 .66-4.53 1.67 .63-4.37
Parent’s edu (ref: No HS)
High school graduate 1.02 .73-1.44 1.01 .72-1.42
College or higher degree .99 .71-1.40 .98 .70-1.38
Free-lunch eligibility .98 .63-1.53 .99 .64-1.54
Total GPA (ref: D’s and F’s)
Mostly C’s and D’s .48 .18-1.28 .48 .18-1.29

67
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

Figure 3.4 provides details on the marginal effect for the interaction terms between
indegree ties and brokerage on the likelihood of alcohol consumption. Overall, students are less
likely to consume alcohol as their numbers of indegree ties increase. Although this pattern is
generally found across all different brokerage groups, there is one distinction between low-
brokerage groups and high-brokerage groups. When high-brokerage groups or potential brokers
have no indegree tie, the likelihood of alcohol consumption is relatively higher than low-
brokerage groups. However, the likelihood of alcohol consumption substantially decreases as
their indegree ties increase. Low-brokerage groups also show decreases in the likelihood of
alcohol consumption as their indegree ties increase, but the decreasing rates are smaller
compared to the high-brokerage groups.
Compared to Figure 3.4, the association between the number of outdegree ties and the
likelihood of alcohol use is positively associated across different brokerage groups (Figure 3.5).
However, this tendency fades as brokerage scores increase; for example, those whose brokerage
scores are low are more likely to use alcohol as their outdegree ties increase. Adolescents whose
brokerage scores are more likely to consume alcohol as they nominate more friends, but the
Mostly B’s and C’s .59+ .16-1.08 .43+ .16-1.10
Mostly A’s and B’s .41* .12-.85 .33* .13-.86
Participation in sports 1.41* 1.07-1.86 1.43* 1.08-1.88
Participation in arts .82 .56-1.19 .82 .57-1.19
Participation in academics .83 .59-1.16 .81 .58-1.14
Sibling alcohol use 1.48* 1.09-2.02 1.49* 1.09-2.03
Sibling cigarette use 1.27 .86-1.89 1.28 .86-1.90
Adult alcohol use 1.62*** 1.22-2.15 1.59** 1.20-2.11
Adult cigarette use 1.62 .87-1.67 1.21 .87-1.67
Intercept .90 .32-2.55 .82 .28-2.38
Overall model evaluation
Df

19 19
Pseudo R
2
.085 .082

2

4.87*** 4.66***

68
increasing rates are relatively smaller than those of low-brokerage groups. In sum, the likelihood
of alcohol use decreases as the number of indegree ties increases, whereas it increases as the
number of outdegree ties increases; however, the effects of indegree ties are stronger and the
effects of outdegree ties are weaker for potential brokers compared to low-brokerage groups.

Figure 3.4. Comparison of Probability of Drinking Depending on Numbers of Indegrees
Between Low- and High-brokerage Groups

69
Figure 3.5. Comparison of Probability of Drinking Depending on Numbers of Outdegrees
Between Low- and High-brokerage Groups

Table 3.7 presents the results of the main effects of indegree and outdegree ties on
cigarette use and the interaction effect between brokerage and indegree/outdegree ties. Although
the results do not identify any significant roles of indegree/outdegree ties and the interaction
effects on smoking, brokerage is the salient factor that seems to matter in the likelihood of
cigarette use. For example, the likelihood that the student smokes increases by 1% when one unit
of the brokerage scores increases (OR = 1.01, 95% CI = 1.01-1.02; Model 9) when controlling
other variables. Regardless of non-significant interaction effects in Table 3.7, Figure 3.6 and
Figure 3.7 provide marginal effects for the interaction effects that show what conditions may
lead to an important effect on the likelihood that a broker smokes.

70
Table 3.7. Logistic Regression Models of Cigarette Use: Indegree Brokers vs. Outdegree
Brokers (N= 1,265)
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

Model 9 Model 10
Dependent Variable: Cigarette Use
Primary Predictors OR 95% CI OR 95% CI
Broker * Indegree 1.00 .99-1.00 - -
Broker * Outdegree - - 1.00 1.00-1.00
Broker scores 1.01* 1.00-1.02 1.00 .99-1.01
Indegree .99 .92-1.06 - -
Outdegree - - 1.01 .97-1.06
Control Variables
Female 1.08 .82-1.43 1.08 .82-1.42
Race (ref: Hispanic)
Asian .38*** .25-.57 .39*** .26-.58
Non-Asian/Non-Hispanic 1.73 .66-4.53 1.67 .63-4.37
Parent’s edu (ref: No HS)
High school graduate 1.02 .73-1.44 1.01 .72-1.42
College or higher degree .99 .71-1.40 .98 .70-1.38
Free-lunch eligibility .98 .63-1.53 .99 .64-1.54
Total GPA (ref: D’s and F’s)
Mostly C’s and D’s .48 .18-1.28 .48 .18-1.29
Mostly B’s and C’s .59+ .16-1.08 .43+ .16-1.10
Mostly A’s and B’s .41* .12-.85 .33* .13-.86
Participation in sports 1.41* 1.07-1.86 1.43* 1.08-1.88
Participation in arts .82 .56-1.19 .82 .57-1.19
Participation in academics .83 .59-1.16 .81 .58-1.14
Sibling alcohol use 1.48* 1.09-2.02 1.49* 1.09-2.03
Sibling cigarette use 1.27 .86-1.89 1.28 .86-1.90
Adult alcohol use 1.62*** 1.22-2.15 1.59** 1.20-2.11
Adult cigarette use 1.62 .87-1.67 1.21 .87-1.67
Intercept .90 .32-2.55 .82 .28-2.38
Overall model evaluation
Df

19 19
Pseudo R
2
.085 .082

2

4.87*** 4.66***

71
Figure 3.6. Comparison of Probability of Smoking Depending on Numbers of Indegrees
Between Low- and High-brokerage Groups

In Figure 3.6 that shows marginal effects, there is an opposite pattern that students are
less likely to smoke between low- and high-brokerage groups. When adolescents whose
brokerage scores are low have more indegree ties, they are more likely to smoke cigarettes.
However, high-brokerage groups or potential brokers are less likely to smoke cigarettes as their
indegree ties increase. Interestingly, the opposite pattern between low- and high-brokerage
groups is also found in the effects of outdegree ties on likelihood of smoking (Figure 3.7). Figure
3.7 shows marginal effects of outdegree ties on the likelihood that a broker smokes. While low-
brokerage groups are less likely to use cigarettes as their numbers of outdegree ties increase,
high-brokerage groups are more likely to smoke. These results indicate that adolescents whose

72
brokerage scores are low or non-brokers tend to smoke when they receive more friendship
nominations. On the other hand, non-brokers are less likely to smoke when they nominate many
friends. However, potential brokers seem to show the opposite patterns. When potential brokers
receive more friendship nominations, they are less likely to smoke cigarettes. Additionally,
potential brokers are more likely to smoke cigarettes as they submit more names as their friends.
Figure 3.7. Comparison of Probability of Smoking Depending on Numbers of Outdegrees
Between Low- and High-brokerage Groups

Discussion
Five decades after Granovetter presented his weak ties theory (1973), scholars have
raised the question of whether and how a broker engages in risky behaviors. Several studies have
attempted to answer this question, but the mechanism of how a broker engages in substance use

73
has not been sufficiently addressed due to a lack of examination on the network properties of a
broker’s ties. However, theoretical frameworks such as the contagion theory and social
acceptance explanation emphasize the importance of understanding the network properties of a
broker’s ties, particularly tie degrees and directionality, on substance use behaviors. Centola and
Macy (2007) explained that brokers tend to engage in or transmit risky behaviors when they have
wide ties, or multiple connections with others. Additionally, brokers who seek more friendships
may be more willing to take risky behaviors like substance use considering their unstable and
precarious relations with peers and peer groups. Based on past literature, the present study tests
whether and how the number and directionality of a broker’s friendship ties are associated with
the likelihood of alcohol and cigarette use. Using the 2013 Social Networking Study data, I first
attempt to identify any positive relationship between the total numbers of brokers’ friends and
the likelihood of their alcohol consumption and cigarette smoking to see whether contagion
theory helps explain a broker’s substance use behaviors. Similarly, I explore whether the number
of substance-using friends is positively associated with the likelihood of a broker engaging in
substance use. Considering the importance of status-seeking and social acceptance from peers
among brokers, I also investigate whether and how the likelihood of brokers’ substance use may
differ depending on the number of indegree and outdegree ties of brokers.
The results show that no interaction term is statistically significant in the main effect
models, but these results do not indicate there is no meaningful relationship. Instead, the
relationship between a broker’s friendship and the likelihood of the broker’s substance use may
not be linearly associated; or, a certain range of this relationship may show statistical
significance, whereas the other range may not. Although there is no main effect for a broker’s
friendship on substance use, additional factors help explain an adolescent’s alcohol and cigarette

74
use. The findings also reveal the positive association between brokerage scores and substance
use when controlling for the potential effects of indegree ties and other socio-demographic
characteristics of students. This indicates that students from high-brokerage groups are more
likely to drink alcohol and smoke cigarettes than those from low-brokerage groups when
controlling effects of their indegree ties and socio-demographic backgrounds. However, those
assumptions may not be realistic in understanding adolescent substance use behaviors because it
is rare that students have the same socio-demographic characteristics, brokerage scores, and
indegree ties. Furthermore, an absence of significant main effects related to interaction terms
does not necessarily mean no meaningful relationship between a broker’s friendship and
substance use and therefore I also explored marginal effects of the interaction terms.
To uncover any hidden relationships, I highlight several marginal effects of the
interaction terms that show significant associations between brokers’ friendships and the
likelihood of substance use. First, I found that there is an opposite pattern for the effects of
friendship ties on alcohol drinking behaviors between low- and high-brokerage groups. For
example, potential brokers are less likely to drink alcohol as the number of their friends increases
regardless of whether they use substances, but low-brokerage groups are more likely to drink
alcohol when they have more friends. The results support the conventional belief that having
more friends leads to increase the likelihood of alcohol consumption for the low-brokerage
groups, but not for the high-brokerage groups. Instead, having more friends seems to serve the
deterrent effect in consuming alcohol for potential brokers. These reverse patterns are also found
in the examination of indegree and outdegree ties on cigarette smoking behaviors. For example,
when the number of indegree ties increase, potential brokers are less likely to smoke cigarettes
whereas non-brokers are more likely to smoke. Additionally, the number of outdegree ties for

75
high-brokerage groups is positively associated with the likelihood of cigarette smoking, but low-
brokerage groups show the negative association between outdegree ties and cigarette smoking.
As for alcohol consumption, potential brokers seem to drink alcohol more when they nominate
more friends, but they are less likely to smoke when they receive more friendship nominations.
The results indicate that potential brokers tend to engage in alcohol and cigarette use behaviors
when they seek more friendships. On the other hand, potential brokers are less likely to engage in
substance use behaviors as they receive many friendship nominations from other peers.
These marginal effects of interaction terms do not necessarily support Centola and
Macy’s contagion theory in explaining substance-using behaviors among high school students in
Southern California. Neither the number of friends nor the number of substance-using friends
show the positive association with the probabilities of alcohol and cigarette use for potential
brokers. Put differently, having more friends or substance-using friends leads to decrease the
chances of using alcohol and cigarettes for potential brokers, whereas non-brokers are more
likely to use alcohol and cigarettes as their numbers of friends or substance-using peers increase.
Instead, this study demonstrates the social acceptance explanation for how brokers
engage in smoking behaviors based on the unique characteristic of a broker’s position. Because a
broker’s ties are usually unstable and volatile (Burt 2002a), status-seeking and social acceptance
might be relatively more important to brokers than non-brokers and therefore brokers who seek
status or acceptance might be more willing to take risky behaviors. The positive association
between a broker’s outdegree ties and substance use can support this explanation, showing that a
broker who nominates relatively more friends engages in more alcohol and cigarette use. In the
same vein, having many nominations from others or being highly recognized by peers does not
necessarily increase the likelihood that brokers smoke because they do not need to seek status or

76
social acceptance for friendships by engaging in risky behaviors.
The results here also show that the relationships between the number and directionality of
ties and substance use seem to be different between high-brokerage and low-brokerage groups.
These results may indicate that brokers with many friends are potentially different from popular
individuals with no or low brokerage power regarding substance use behaviors. As discussed
from Figure 3.1 through Figure 3.7, there are distinct differences between low-brokerage (left)
and high-brokerage (right) groups as their number of total ties, indegrees, and outdegrees
increase. For example, having more outdegree ties is positively associated with the likelihood of
cigarette smoking for high-brokerage groups, whereas it is negatively associated with the
likelihood of smoking for low-brokerage groups (Figure 3.7); and the likelihood of smoking for
high-brokerage groups substantially decreases when they have more indegree ties, whereas the
likelihood of smoking for low-brokerage groups increases (Figure 3.6). Stated differently,
potential brokers are more willing to engage in smoking cigarettes as they want to build more
friendships. However, non-brokers do not show more cigarette smoking even when they seek
more friendships. As for indegree ties, having more friendship nominations may deter substance
use for potential brokers while having no impact on non-brokers. These differences are not
unexpected considering differences in qualities of their ties with other adolescents and potential
status in peer groups. While brokers’ ties with others are usually precarious and weak, friendship
ties of popular individuals are not necessarily weak or unstable like those of brokers. Instead,
popular or high-central individuals, who tend to “receive” many friendship nominations, are
highly visible and recognized by other peers in social networks. Because of their unstable
standing in peer groups, brokers may perceive peer recognition and pursue friendship differently
compared to non-brokers although further examination is needed to confirm this argument in

77
future studies.
Overall, the present study illustrates the mechanisms underpinning how a broker engages
in risky behaviors, based on social network theories. In addition to empirical evidence from this
study, I have several suggestions for future studies to further validate these mechanisms. First, it
would be worthwhile to evaluate the likelihood of a broker’s substance use with other types of
broker measures. This study relies on the EV-brokerage measure, which is based on calculating
the shortest path between individuals to assign brokerage scores, to compare brokerage power
between individuals. However, this continuous measurement does not provide a clear criterion
for researchers to identify brokers. Therefore, the current study simply compares low- and high-
brokerage groups on substance use and assumes high-brokerage groups as potential brokers
without identifying actual brokers. This limitation may be reduced by using other types of broker
measurements including a binary measure. For example, Osgood et al. (2014) applied a cluster-
connector method that identifies clustered groups in networks and labels those who connect
members of two or more different groups as brokers in examination of a broker’s substance use.
Because this binary measure explicitly labels those who are and are not brokers, it is
straightforward to examine the likelihood of brokers’ substance use compared to those of non-
brokers. Despite advantages of the cluster-connector method, the EV-brokerage measure is still
considered the best option for this study because segregation levels in the friendship networks of
Social Networking Study data are quite low, and this binary measure is less effective in
identifying clustered groups and brokers. However, future studies can further provide empirical
evidence for the mechanisms with this cluster-connector method when their social network
datasets are highly segregated.
The second suggestion is related to the first point: using different adolescent samples and

78
friendship networks. In this study, I selected the Social Networking Study data due to a number
of unique advantages. For example, all the students were from similar neighborhoods and the
same school district in Southern California, indicating that this unique characteristic may reduce
the effects of potential confounders in exploring research questions. As stated, however, low
segregation levels in friendship networks limit empirical analysis to leveraging just a few
established measures of brokerage. More importantly, this data is not derived from a national
representative sample but mostly Hispanic and Asian students from racial/ethnic minority
dominant schools and low-income families. Therefore, future studies may want to consider using
a national representative sample such as the National Longitudinal Study of Adolescent to Adult
Health for generalization of the results and providing additional empirical support.
Finally, longitudinal analysis would allow us to understand and expand the findings as
causal relationships. The current study relies on a cross-sectional analysis, using the single wave
(wave 4) of Social Networking Study data. Although the cross-sectional analysis in this study
finds meaningful associations between a broker’s friendship and the likelihood that the broker
drinks alcohol and smokes cigarettes, the results do not directly reveal any causal relationships.
To identify whether a broker’s friendship actually affects the likelihood of substance use
behaviors, future studies can use longitudinal methodologies such as the dynamic network
analysis (Carley 2003). One of the big challenges for social network analysis is to examine
several different layers of information at the same time. Even for a cross-sectional analysis for
social network data, researchers should consider “relational” aspects (e.g., numbers of friendship
ties, directionality of ties, network positions) and “attribute” aspects (e.g., socio-demographic
characteristics) at the same time. Thus, it has been considered a difficult task to include temporal
aspects in social network analyses. However, recent social network analyses have adopted new

79
approaches including dynamic network analysis, which allows researchers to see the temporal,
relational, and attribute aspects of individuals at the same time. A longitudinal analysis study can
help us explore the primary cause for an adolescent’s substance use and better understand the
mechanisms that explain the key factors that drive a broker to engage in risky behaviors.

80
CHAPTER 4
STRUCTURAL AND RELATIONAL EMBEDDEDNESS ON SUBSTANCE USE:
AN EXAMINATION OF BROKERAGE AND CROSS-GENDER FRIENDSHIPS ON
ADOLESCENT ALCOHOL AND CIGARETTE USE

Introduction
As network data and theories have been introduced, many studies of peer influence have
examined the effect of network position on adolescent substance use behaviors. In adolescent
friendship relations and structures, some network positions may be relatively accessible as a
source of substance use behaviors. This is particularly true with a broker—an actor who links
individuals or groups otherwise disconnected—position (Granovetter 1977; Centola and Macy
2007; Osgood et al. 2014; Henry and Kobus 2007). Because brokers are connected to many
(distant) others via their weak yet long ties, these indirect ties allow greater chances for brokers
to access substances.
In addition to brokers’ weak and indirect ties, their immediate connections are also
diverse, which helps increase chances of having substance-using peers, compared to those in
other network positions such as in-group members who mostly connect with similar others.
Accordingly, scholars have supported that those who have cross-gender friendships can serve as
brokers due to their connections between opposite gender groups and their cross-gender
friendships help promote changes in drinking behaviors (Kreager and Haynie 2011; Kreager,
Haynie, and Hopfer 2013). In fact, a large body of literature finds the positive association
between cross-gender friendships and the likelihood of substance use behaviors among
adolescents (Molloy et al. 2014; Mercken et al. 2010; Grard et al. 2018; Mrug, Borch, and
Cillessen 2011; Gaughan 2006; Poulin, Denault, and Pedersen 2011; Malow-Iroff 2006).
However, it is not clear whether and how brokers’ ties—heterogenous and long ties—lead to a

81
greater likelihood of substance use due to lack of examination of these structural and relational
aspects of brokers. Additionally, there are gender differences not only in the prevalence of
substance use and proportion of cross-gender friends but also the role of cross-gender friendships
on substance use (Gaughan 2006; Mrug, Borch, and Cillessen 2011; Poulin, Denault, and
Pedersen 2011; Kreager and Haynie 2011), and therefore it is worthwhile to investigate these
research questions by gender.
The present study examines the impact of both structural and relational attributes of
brokers on substance use behavior by asking three questions: 1) whether adolescent brokers
engage in more substance use because of their long ties, 2) whether adolescent brokers engage in
more substance use because of their cross-gender friendships, and 3) whether brokers show
different structural and/or relational characteristics by gender with respect to substance use
behaviors. Using Wave 1 of the National Longitudinal Study of Adolescent to Adult Health, I
aim to examine the likelihood of alcohol drinking, cigarette smoking, and being drunk. Based on
literature regarding cross-gender friendships as a moderator between a broker’s long ties and
substance use (as presented in Figure 1), I add cross-gender friendships in the main models and
examine them for male and female student respectively.
Figure 4.1. A Conceptual Model of a Broker’s Alcohol and Cigarette Use
Broker’s Long Ties
Broker’s Heterogeneous Ties
(Cross-gender Friendships)
Alcohol Consumption
Cigarette Smoking

82

Structural and Relational Dimensions of Brokers and the Likelihood of Substance Use
Over the past few decades, the broker role in social networks has been demonstrated to be
an advantageous position for accessing information and resources. In the literature on brokers,
two aspects are highlighted to explain how a broker benefits from its position: (1) a relational
aspect: a broker’s heterogeneous ties that can connect with dissimilar or diverse others, and (2) a
structural aspect: a broker’s long ties that can reach out as far as possible to distant individuals or
groups in networks. Because people usually interact with similar others, segregated groups or
clusters in social networks often have homogeneous characteristics (McPherson, Smith-Lovin,
and Cook 2001a). On the other hand, a broker who connects between these separate individuals
or groups tend to link or befriend people or groups from different backgrounds. Distant
individuals or groups can be connected through a broker, and therefore a broker has direct or
indirect paths that link with all or most individuals in networks. Based on this relational and
structural aspect, brokers tend to have chances to access new ideas and diverse opportunities
from distant and diverse groups and thus bring many benefits to them (Granovetter 1977; Burt
2004; 1992). Considering that a broker's heterogeneous and long ties promote accessibility to
favorable information or people, we can also ask whether brokers are likely to show risky
behaviors such as substance use because their ties may also allow them to have greater exposure
to diverse people, including those who use substances or can provide.
For this question, there are two different results that explain the likelihood of a broker’s
substance use. The first says that broker position has a positive association with substance use
(Henry and Kobus 2007; Osgood et al. 2013; Kreager and Haynie 2011; Kreager, Haynie, and
Hopfer 2013). In Henry and Kobus’s study (2007), brokers were more likely to consume alcohol
and smoke cigarettes, and Osgood et al. (2014) identified the positive association between a

83
broker position and marijuana use. In the same vein, brokers in the studies of Kreager and
colleagues (Kreager and Haynie 2011; Kreager, Haynie, and Hopfer 2013), corroborated that a
broker position is sensitive to not only its immediate, direct relations but also indirect relations.
For example, brokers, particularly girls, were more susceptible to drinking behaviors of their
peers and partners as well as friends of their partners. In contrast, other studies provide evidence
that a broker position does not necessarily relate to greater likelihood of substance use (Ennett
and Bauman 1993; Fang et al. 2003). These scholars found that there is no significant association
between a broker position and the likelihood of substance use; instead of brokers, isolates
showed high levels of likelihood of substance use in the studies. The latter sounds conflicting
against the former one, but the explanations of why an isolate shows more smoking cigarettes is
in the same line of the former one’s argument. According to the latter explanations, scholars
elucidated that isolates tended to smoke cigarettes more compared to other network positions
because they have connections with others from outside the school. This indicates that some
isolates may not have in-school friendships, but they can play as brokers that connect in-school
and out-of-school individuals.
Table 4.1. A Summary of Previous Studies on a Broker’s Substance Use

17
Adolescents who answered “yes” to the question, “Do you smoke cigarettes now?”, and who also indicated they
had one or more packs of cigarettes in their lifetime.
18
Adolescents who responded to have experiences of smoking; that is, those who have ever smoked.
Authors
(year)
Types/Measures of
Brokers
Types of
Substances
Findings
Ennett and
Bauman
(1993)
Liaisons who have at
least two links with
clique members
Cigarette
smoking
17

- Isolates were more likely to smoke
cigarettes compared to clique members
or brokers
- High probabilities of smoking
cigarettes for isolates are due to
connections with others from outside
the school
Fang et al.
(2003)
Liaisons (same as
Ennett and Bauman
1993)
Cigarette
smoking
18

- Overall isolates were more likely to
have experimented with cigarettes
compared to group members or liaisons
- Among male 10
th
graders, group

84

These two different hypotheses, summarized in Table 4.1., also provide inconsistent
speculations on “how or why brokers have higher (or lower) show substance use,” although no
speculation has been empirically explored yet. The first prong argued that the structural aspect—
a broker’s long ties—enables brokers to have direct and/or indirect connections with substance-
using peers in networks, leading them to have greater exposure to substance use behaviors
(Henry and Kobus 2007; Osgood et al. 2014). However, it is still questionable whether a broker’s

19
A binary variable for each type of substance to explain whether a student had (not) used the substance in the
preceding 6 months.
20
A binary measure taken from responses to the question, “Over the past 12 months, on how many days did you
drink five or more drinks in a row?”
21
A question of drinking: “During the past month, how many times have you had beer, wine, wine coolers, or other
liquor?”; smoking: “During the past month, how many times have you smoked any cigarette?”; drunkenness:
“During the past month, how many times have you been drunk from drinking wine, wine coolers, or other liquor?”
22
A binary variable using a question of alcohol drinking: “During the past month, how many times have you had
beer, wine, wine coolers, or other liquor?”; cigarette smoking: “During the past month, how many times have you
smoked any cigarette?”; marijuana use: “During the past month, how many times have you smoked any marijuana
(pot, reefer, weed, blunts)?”
members or liaisons had smoked more
than isolates
Henry and
Kobus (2007)
Liaisons who were
connected to at least
two other students,
neither of whom were
connected to the other
Tobacco,
alcohol,
marijuana, and
inhalant use
19

- Brokers were more likely to use
tobacco than group members or isolates
and were more likely to use alcohol than
isolates
- Three social positions did not differ on
their use of marijuana or inhalants
Kreager and
Haynie (2011)
Those who are in
romantic partnership
Alcohol
drinking
20

- Brokers are more susceptible to
drinking behaviors of their partners and
friends of their partners
Kreager,
Haynie, and
Hopfer (2013)
Those who are in
romantic partnership
Alcohol
drinking,
cigarette
smoking, and
drunkenness
21

- Brokers are more susceptible to
drinking behaviors of their partners and
friends of their partners (same result in
Kreager and Haynie 2011)
- Brokers are more susceptible to
smoking behaviors of their partners, but
not friends of their partners
Osgood et al.
(2013)
Liaisons who connect
different clustered
communities
Alcohol, tobacco,
and marijuana
use
22

- Group members were more likely to
drink than isolates and liaisons
- Isolates were more likely to use
cigarettes than group members
- Liaisons were more likely to use
marijuana than group members

85
long ties can also help reduce the likelihood of substance use due to greater exposure or
accessibility to non-substance users. In contrast, the second prong highlights the relational aspect
of isolates to explain why isolates use more substances compared to brokers or popular
adolescents. While isolates usually have no or very few school friends who are mostly non-
substance users, they have social ties with out-of-school cliques that have easy access to
substances. Stated differently, isolates’ immediate ties with substance users lead them to engage
in more substance use behaviors. However, it is possible brokers also should show a high degree
of substance use because they may have not only immediate connections with substance users
via heterogeneous ties but also direct/indirect ties with substance users via long ties. Because no
study has examined both aspects of brokers in identifying the potential mechanism about a
broker’s substance use, the above questions remain unanswered. Thus, the current study
examines whether and how the relational and structural aspect of a broker help explain a
potential mechanism for a broker’s substance use. Considering that cross-gender friendships are
one of the key factors that explain the likelihood of adolescent substance use, this study
examines whether a structural and relational dimension of a broker help explain the likelihood of
a broker’s substance use.

Cross-gender Friendships and Substance Use
Homophily, defined as the tendency for individuals to become friends with similar others,
is one of the robust findings in adolescent friendship networks. A large body of literature shows
many different types of homophily, such as race/ethnicity (Goodreau, Kitts, and Morris 2009;
Quillian and Campbell 2003; Quillian and Redd 2009; Joyner and Kao 2000) and gender-based
homophily (Shrum, Cheek, and Hunter 1988; Brashears 2008; Mayhew et al. 1995; Smith,

86
McPherson, and Smith-Lovin 2014). Although adolescents generally strongly prefer friendships
with similar others, gender homophily has unique characteristics compared to other types of
homophily. For example, gender homophily appears the strongest in pre- and early adolescence
even given roughly equal numbers of both genders in social settings, while race and ethnic
homophily are often influenced by the structural effects of category size (McPherson, Smith-
Lovin, and Cook 2001a). However, the same-sex friendships decline over time because of
growing interests and exposure to other-sex peers during adolescence and late adolescence
(Arndorfer and Stormshak 2008; Kalmijn 2002; Dunphy 1963). Additionally, all-girl or all-boy
friendship groups are more tight-knit with strong and mutual ties, while cross-gender friendships
(CGFs) are loosely connected (Molloy et al. 2014; Kreager and Haynie 2011; Kreager, Haynie,
and Hopfer 2013). If CGFs are weakly linked and numbers of CGFs are growing during
adolescence and late adolescence, can CGFs be network bridges in adolescent friendship
networks and therefore adolescents who have CGFs tend to show different behaviors?
To answer this question, a few studies provide theoretical and empirical support for the
potential role of CGFs in serving as network bridges. Dunphy has addressed the role of CGFs as
a bridge in his ideal-typical model of mixed-gender peer group development in adolescence
(1963). From early through late adolescence, adolescents start dating and mixed-gender groups
begin to form. As romantic partnerships and CGFs are forming, exposure to new norms or
behaviors via these ties increases. Dunphy notes that dating ties or CGFs remain structurally
weak and therefore often serve as bridges in peer networks. The recent studies have further
supported the role of CGFs as bridges and identified that adolescents who have CGFs or
romantic relationships are more likely to adopt risky behaviors such as substance use (Kreager
and Haynie 2011; Kreager, Haynie, and Hopfer 2013; Payne and Cornwell 2007). Using the

87
concept of Granovetter’s weak ties, Kreager and Haynie (2011) defined romantic relationships as
network bridges or liaisons and these bridges promote changes in drinking behaviors. Because
romantic relationships connect adolescents to “different” or “new” peer groups, adolescents in a
peer group of couples tend to show high levels of drinking behaviors. Additionally, Grard et al.
(2018) found that having other-sex friendships was associated with binge drinking, smoking, and
cannabis use for female students, and smoking for male students. Overall, research concludes
that having CGFs is significantly associated with the increased likelihood of adolescent
substance use (Molloy et al. 2014; Mercken et al. 2010; Grard et al. 2018; Mrug, Borch, and
Cillessen 2011; Gaughan 2006; Poulin, Denault, and Pedersen 2011; Malow-Iroff 2006).
Although prior literature has identified the relationship between CGFs and the likelihood
of adolescent substance use, most of the studies have primarily looked at immediate ties or
relational aspects of those who have CGFs in explaining substance use. Stated differently, it is
still unclear whether and how the structural position of those who have CGFs helps explain the
likelihood of substance use. Despite lack of studies on the structural aspect of CGFs, few
scholars have addressed a potential structural position of CGFs in explaining high levels of
substance use among those who have CGFs. Because mixed-sex groups are less popular but
more accessible to diverse individuals and groups compared to same-sex peer groups, those who
have CGFs tend to adopt new norms or behaviors with respect to alcohol consumption and
cigarette smoking (Molloy et al. 2014). Because network literature illustrates that less popular
network positions that are loosely connected with diverse people or groups often deal with
pressure and double commitment to multiple group memberships (Osgood et al. 2014; Henry and
Kobus 2007; Ennett and Bauman 1993; Fang et al. 2003), it is not surprising to see the positive
association between CGFs and the likelihood of substance use behaviors.

88
Prior studies also highlighted that this positive association between CGFs and the
likelihood of substance use is more pronounced among girls. In Mrug, Borch, and Cillessen’s
study (2011), girls who had CGFs were five times more likely to smoke cigarettes than those
who had same-sex friends (SSFs). When compared to male counterparts, girls are still more
likely to engage in substance use; for example, girls with CGFs were more likely to consume
alcohol due to their male friends whereas boys with CGFs were not influenced by their female
friends (Gaughan 2006). According to Arndorfer and Stormshak (2008), the likelihood of
substance use increases when female students have CGFs, whereas male students are more likely
to use substances when they have more same-sex friendships. Moreover, Kreager and Haynie
(2011) claimed that boys were more influenced by their other-sex partners, but in more positive
ways; that is, they were less likely to use substances thanks to romantic partnerships. Poulin,
Denault, and Pedersen (2011) also supported this notion, showing that CGFs may result in risky
behaviors for girls, whereas CGFs may be less associated with risky behaviors for boys. Scholars
disagree that these gender differences are derived from biological sex. Instead, masculinity is
correlated with active participation in risky behaviors among boys (Dempster 2011; Mullen et al.
2007; Sanders 2011). Using Connell’s hegemonic masculinity and West and Zimmerman’s doing
gender concept, feminist scholars explain that substance use in adolescence is a means of
achieving masculinity. In sum, boys generally show higher levels of substance use than girls, but
girls who have CGFs are also likely to use substances. The summary is presented in Table 4.2.
Table 4.2. A Summary of Previous Studies on Cross-gender Friendships
Authors (year) Findings
Mrug, Broch, and
Cilessen (2011)
- Girls who have cross-gender friends were more likely to smoke cigarettes
compared to girls who have same-sex friends
Gaughan 2006
- Girls who have cross-gender friends were more likely to consume alcohol
compared to boys who have cross-gender friends

89

Based on the reviewed literature, the present study aims to answer three main questions:
1) whether adolescent brokers engage in more substance use just simply because of their
structural position that connect distant peers as far as possible, 2) whether adolescent brokers
engage in more substance use because their immediate ties are more likely to be different,
diverse friendships such as CGFs, and 3) whether brokers show different structural and/or
relational aspects by gender in terms of substance use behaviors.

Data, Measures, and Methods
Data and Sample
This research uses data from the National Longitudinal Study of Adolescent to Adult
Health (i.e., Add Health), a study of a national representative sample of U.S. adolescents.
Because the current study intends to examine adolescent friendship networks and their substance
use behaviors, I chose the Wave 1 of Add Health that includes 90,118 adolescents across 144
middle and high schools who were in grades 7 to 12 during the 1994-95 school year. The
students were asked about their socio-demographic characteristics and health behaviors with
their friend relations; the Wave 1 data also includes survey information from school
administrators to gather school information such as size, region, and student/teacher
compositions. I used four datasets—in-school survey, friendship nomination data, school
Arndorfer and
Stormshak (2008)
- Girls who have cross-gender friends were more likely to use substance
- Boys who have same-gender friends were more likely to use substances
Kreager and Haynie
(2011)
- Boys who have cross-gender friends or a romantic partner were less likely
to drink alcohol compared to boys who do not have
Poulin, Denault, and
Pedersen (2011)
- Cross-gender friendship is positively associated with girls, but negatively
associated with boys

90
information, and school administrator survey—to create friendship networks, connecting them
with individual and school information. The original datasets encompass 90,118 students with
172 schools. After cleaning invalid IDs
23
and excluding schools with low response rates (lower
than 75 percent), the final sample is 61,608 students within 100 schools
24
which was further
reduced by 417 students due to missing data on the gender variable resulting in a final sample
size of 61,191 students in 100 schools. Further details are provided in the multiple imputation
section.
Dependent Variables: Alcohol and Cigarette Use and Drunkenness
The dependent variables are alcohol consumption, and cigarette smoking. To measure
these concepts, I used two questions that asked students how often they 1) have smoked
cigarettes, and 2) have had an alcoholic drink like beer, wine, or liquor during the past twelve
months. Because many of the students responded with no use of alcohol or cigarettes, I use a
binary variable of alcohol and cigarette use and drunkenness where 0 indicates never used or got
drunk, and 1 indicates one or more days of having ever used substances or ever drunk over the
past twelve months. Because drunkenness, considered as more extreme levels of alcohol
consumption, may cause behavioral problems and other health issues, it is worthy to examine
this variable in the main analysis to see how brokerage and cross-gender friendships differently
(or, similarly) explain drunkenness compared to no alcohol use or ever use.
Independent and Moderator Variables: Brokerage and Cross-gender Friendship
The primary independent variable is brokerage. Although several measures such as

23
Among three IDs (student ID, survey questionnaire ID, school ID) in the datasets, I confirmed invalid values in
student IDs and survey questionnaire IDs that hindered connecting the datasets. After cleaning invalid these IDs, the
datasets include 80,026 students within 141 schools.
24
To prevent potential biases that a small number of students represents the school population (Haas, Schaefer, and
Kornienko 2010; Schaefer et al. 2011), I included only schools from which 75 percent or more students participated
in the survey.

91
betweenness centrality and EV-brokerage are available to calculate brokerage (Freeman 1977;
Everett and Valente 2016), a cluster-connector approach is used to identify “cluster-connects” as
brokers who link members of two or more different clusters. In a first step, I sorted out clustered
groups using a cluster_edge_betweenness function from igraph R package. As one of the
community detection methods, this cluster_edge_betweenness function measures the number of
shortest paths from one vertex to another and removes edges with the highest edge betweenness
scores. Once this function is conducted, the results provide information of a cluster size (the
numbers of people in each cluster) and a unique number of each cluster (the identification
number of each cluster) for every individual in the network. Prior to labeling those who connect
two or more clusters as brokers, I excluded small clusters that have only one or two individuals
in groups. Then, I checked all individuals to see if they link two or more different clusters and, if
so, labelled them as cluster connectors.
This cluster connector method is the best fit for this study due to its several unique
features. First, this method primarily reflects the structural aspects of a broker. Based on the
shortest-path calculation, cluster connectors lie on the shorter or shortest paths between other
individuals, indicating that cluster connectors are those who have long ties that can reach to
distant or faraway others. Furthermore, it is clear to see who brokers are based on this method.
The cluster connector method is a binary variable that tells us whether individuals in networks
connect two or more members of different clusters or not. Thus, it is a more intuitive tool to be
interpreted compared to other brokerage measures such as betweenness and EV-brokerage that
give us continuous levels of brokerage power or the degree to which an individual holds a broker
position.
As for the moderator variable, I used the proportion of cross-gender friends. A number of

92
cross-gender friends is another good measure of cross-gender friendship, but I decided to use the
proportion measure rather than the count measure to reduce any potential impacts of popularity
(i.e., many friendship ties) in examining the role of cross-gender friendship in a broker’s
substance use. To measure the proportion of cross-gender friends, I used friendship network data
and linked information about peers’ gender. Then, I divided the total number of cross-gender
friends by the total number of friends and multiplied by 100. Thus, 0 of this variable indicates
that a student has no cross-gender friends, whereas a score of 100 on the variable indicates that
all of a student’s friends are of the opposite gender from him/her.
Control Variables and Multiple Imputation
I included eight student-level and three school-level control variables in the main
analyses. First, the student-level control variables are: 1) gender, 2) grade, 3) race/ethnicity, 4)
parent’s educational attainment, 5) sport activity participation, 6) art activity participation, 7)
academic activity participation, 8) total GPA.
The gender is a binary variable based on the original survey questionnaire in Add Health,
asking whether respondents are male or female. As for the grade variable, the survey let
respondents choose between grade 6 and grade 12 and a similar number of the sample is quite
equally distributed over each category. I constructed this as a continuous variable because I
expected a positive association between levels of grade and a likelihood of substance use instead
of comparing a particular grade level with other grades. The survey also allowed respondents to
select their ethnicity—whether they are of Hispanic or Spanish origin—and race—whether they
are White, Black/African American, Asian/Pacific Islander, American Indian/Native American,
and other. I combined respondents who answered yes for the American Indian/Native American
and Other race questions into other due to small numbers of each. Then, I created the race

93
variable with five categories: 1) non-Hispanic White, 2) non-Hispanic Black, 3) Hispanic, 4)
non-Hispanic Asian, 5) non-Hispanic others. As for parents’ educational attainment, the survey
asked respondents “how far in school did mother/father go.” I used the highest level of a parent’s
educational attainment between the mother or father or if only one parent’s information was
available. Then, I created the parent education variable with three categories: 1) less than high
school, 2) high school diploma or equivalent level, and 3) college or higher level. The survey
also asked respondents whether they participated in any of 30 school-based activities or clubs.
Using the literature (Schaefer et al. 2011; Schaefer, Simpkins, and Ettekal 2018), I categorized
these 36 activities or club participation into three types—sports, arts, academics—as a separate
binary variable.
25
As for the total GPA, respondents were asked about their letter grades in
English/Language Arts, Mathematics, History/Social Sciences, Science respectively. Using these
questions, I calculated the average of the student’s grades in these four subjects (1=D, 4=A).
The school-level control variables are: 1) type of school (i.e., public vs. private), 2)
urbanicity of school, and 3) level of gender-heterogeneity. The type of school is a binary variable
that whether schools were public or private. The urbanicity of school variable has three
categories that include Urban, Suburban, and Rural; The last variable, gender-heterogeneity, is
continuous variables. Gender-heterogeneity represents a level of distribution by different gender
in school and I used Blau’s heterogeneity index (1−∑𝑝 i
2
), where 𝑝 i is the proportion of group
members in each of the 𝑖 categories (Blau 1977) to calculate this gender-heterogeneity for each
school. A high heterogeneity index means that the ratio of male/female is close to equal, whereas

25
1. Sports: Badminton, basketball, baseball, cheerleading, cross country, dancing, exercises, football, hiking,
soccer, softball, swimming/diving, tennis, running/racing/track and field, volleyball, wrestling; 2. Arts: band,
chorus/choir, drama club, orchestra; 3. academics: book club, computer club, California scholarship federation
(CGF), debate team, French club, history club, honor society, key club, Latin club, math club, newspaper, science
club, Spanish club, student council, yearbook

94
a low index indicates that a single gender dominates.
Not surprisingly, some variables have numbers of missing cases. According to the
Appendix A, about 17 percent of the sample have missing information about the proportion of
cross-gender friendships; about 15 and 12 percentage of the sample does not have information on
parent’s educational attainment and GPA information. A smaller than 10 percent of the sample
has missing information about these four variables—alcohol (6.00%) and cigarette (5.76%) use,
drunkenness (6.54%), and race/ethnicity (6.83%). For grade and gender, missing cases are lower
than one percent of the sample.
To prevent any potential biases due to missingness, I conducted a multiple imputation to
handle missing values on the percentage of cross-gender friendship, parent’s educational
attainment, total GPA, race/ethnicity, drunkenness, alcohol use, cigarette use, grade, and
gender. In the imputation process, I used three variables as auxiliary variables: numbers of sport
activity/club participation, numbers of art activity/club participation, numbers of academic
activity/club participation. I chose these variables because they are highly correlated with
imputed variables with no missing cases. I conducted the multiple imputation by chained
equations (MICE) with five imputation datasets, considering the large size of the sample. MICE
is a better approach for imputing the datasets compared to multivariate normal distribution
(MVN) because imputed variables are a variety of binary, categorical, and continuous variables,
and MICE allowed me to use separate conditional distribution for each imputed variable (UCLA,
Institute for Digital Research & Education, Statistical Consulting 2020).
Although I have conducted the multiple imputation for all the missing variables, the final
sample size is 61,191 students, not 61,608 students, because missing cases in gender were not
included in the main analyses. That is because I examined a single gender of the likelihood of

95
alcohol and cigarette use—that is, female and male students of alcohol/cigarette use separately—
for comparing gender differences and these analyses do not let me use imputed data. Despite the
small number of missing cases in the gender variable, I have checked whether missing cases and
non-missing cases in the gender variable have significantly different characteristics and the
examination indicates no significant difference between these two sub-samples.
Analytic Strategy
To identify any significant association between brokerage and substance use and
moderating effects of cross-gender friendships, I used multilevel logistic regression models
where the student is level 1 and the school is level 2 (Raudenbush and Bryk 2002). For the
nested datasets with high clustering of residuals by schools, a multilevel model should be
employed; furthermore, the dependent variables are binary responses and therefore multilevel
logistic regression models are the best fit to examine the questions in this study. To examine
moderating effects of cross-gender friendships, I included the interaction term between
brokerage and cross-gender friendships using equation (1).
𝑙𝑜𝑔𝑖𝑡 {Pr (𝑆𝑢𝑏𝑠𝑡𝑎𝑛𝑐𝑒 𝑈𝑠𝑒 𝑖𝑗
= 1|𝑥 𝑖𝑗
)} = 𝛾 00
+ 𝛾 10
𝐵𝑟𝑜𝑘𝑒𝑟 𝑖𝑗
+
𝛾 20
𝐶𝑟𝑜𝑠𝑠𝑔𝑒𝑛𝑑𝑒𝑟 𝑓𝑟𝑖𝑒𝑛𝑑𝑠 ℎ𝑖 𝑝 𝑖𝑗
+ 𝛾 30
𝐵𝑟𝑜𝑘𝑒𝑟 ∗ 𝐶𝑟𝑜𝑠𝑠𝑔𝑒𝑛𝑑𝑒𝑟 𝑓𝑟𝑖𝑒𝑛𝑑𝑠 ℎ𝑖𝑝
𝑖𝑗
+
𝛾 𝑘 0
𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 𝑖𝑗
+ 𝑢 0𝑗 + 𝑟 𝑖𝑗
(1)
Although I conducted the main analyses using equation (1), the unconditional models and
the basic models are included in the Appendix C to check any significant changes as adding
moderating and control variables. Prior to the main analyses, I first conducted the unconditional
models without independent and control variables and the basic models that include only the
independent variable (Appendix C). For the main analyses, I conducted the multiple imputation
procedure and multilevel logistic regression analyses using the STATA 14.1.

96

Results
Descriptive Results
Table 4.3 provides a descriptive summary of substance use behaviors, friendship
networks, and socio-demographic characteristics of the students. About half of the sample
responded that they had consumed alcohol during the past 12 months (53.52%), whereas about
30 percent of the sample reported having ever been drunk. As for cigarette use, about one-third
of the sample reported smoking cigarettes (34.99%). Considering the past 12 months as the
duration of substance use, the high prevalence of substance use is not surprising; furthermore,
alcohol is the most-used substance and being drunk is an extreme version of alcohol
consumption and therefore the lower rate of drunkenness than of alcohol consumption seems
predictable. Based on the cluster connector method, brokers are about half of the sample. The
average proportion of cross-gender friendships was around 35%, and the students reported about
two cross-gender friends on average (SD = 1.72) with a range of 0 to 9.
Table 4.3. Descriptive Sample Statistics of Wave 1 in Add Health Data, 1994-95 (n = 61,608
of students, N = 100 of schools)
Variable
Dependent Variables Mean/% SD Min Max
Alcohol use (binary) 53.52% 0 1
Cigarette use (binary) 34.99% 0 1
Being Drunk (binary) 29.69% 0 1
Independent and Mediating Variables
Broker (cluster connector) 50.73% 0 1
# of Cross-gender Friends 2.03 1.72 0 9
% of Cross-gender Friends 34.74 26.77 0.00 100
Control Variables – Student-level
Male 49.91% 0 1
Grade 9.51 1.63 6 12
Race/Ethnicity
White 55.28%
Black 15.74%
Hispanic 16.22%
Asian/Pacific Islander 5.91%
Other 6.86%

97

The sample is equally distributed by gender. As for race/ethnicity, the majority of the
sample was White (55.28%); Black and Hispanic groups constituted about 16% and two other
groups, Asian and Other groups, constituted 5.91% and 6.85% respectively. Parents of the
students in the study had high school diplomas (43.62%) or college or higher degrees (44.11%),
whereas 12.28% of the sample reported that their parents had no high school diploma or
education. The average GPAs were 2.81 (SD = 0.80) on the four subjects: English/Language
Arts, Mathematics, History/Social Sciences, Science. More than half of the sample participated
in sports (54.92%), whereas around 30% of the sample participated in arts (26.54%) and in
academics (31.74%). Among 100 schools, 10 percent of the sample schools are private schools.
Most of the sample schools are from suburban areas (59%), but 28% of the sample is urban
schools and 13% is rural schools. The school-level gender heterogeneity is .50 (SD=.051).
Multilevel Model Results
Using multilevel logistic regression models, Table 4.4 through Table 4.6 present the
association between brokerage and substance use, and the moderating effects of cross-gender
friendships on this association. The likelihood of alcohol consumption, cigarette smoking, and
drunkenness is predicted in Table 4.4, Table 4.5, and Table 4.6 respectively. I begin the analyses
Parent’s Education
No high school diploma 12.28%
High school graduate 43.62%
College or higher degree 44.11%
Total GPA 2.81 0.80 1.00 4.00
Participation in Sports 54.92% 0 1
Participation in Arts 26.54% 0 1
Participation in Academics 31.74% 0 1
Control Variables – School-level
Private school 10.00% 0 1
Urbanicity
Urban 28.00%
Suburban 59.00%
Rural 13.00%
Gender-heterogeneity .50 .051 .0065 .53

98
with a decomposition of the variance components, using variances in the unconditional models
(Appendix 4 Table C) and in full models (Table 4-6). First, I found that a large portion of the
variance in the likelihood of substance use was due to differences within schools. For female
student alcohol use, the student- and school-level covariates account for 71% of the variance at
the student-level, but only for approximately 12% of variance at school-level, and 4% of total
variance. Similarly, 53% of variance at the student-level and 4% of the variance at the school-
level are explained by the covariates regarding cigarette use for the female students
26
. As for the
variance in drunkenness, the covariates account for 76% of variance at the student-level, whereas
13% of variance at the school-level is explained by the covariates. This variance pattern is also
found in male students’ substance use behaviors. For example, 42% of variances at the student-
level are explained by the covariates, whereas 1% of the variance at the school-level are
explained by the covariates with respect to the likelihood of cigarette use.
Table 4.4. Odds Ratios on the Likelihood of Alcohol Consumption by Gender (N = 61,608)

26
I compare the variances of the unconditional model (Alcohol Use in Appendix B) and the full model (Model 1)
for alcohol use; that is, the variance is calculated by 1 – (variance in a full model/variance in unconditional model).
First is 1-(.29
2
/.58
2
), the second one is from 1-(1.81
2
/1.81
2
), and the last overall variance is from 1-(.29
2
+
1.81
2
)/(.58
2
+ 1.81
2
) for alcohol consumption.
27
OR = 1.005339, 95% CI = 1.003911-1.006768 (Male); OR = 1.005948, 95% CI = 1.004479-1.00742 (Female)
28
OR = 1.000498, 95% CI = .9987908-1.002207 (Male); OR = .9975587, 95% CI = .9958173-.9993031 (Female)
Alcohol Use
Primary Predictors OR (95% CI) OR (95% CI)
Male Female
Broker (cluster connector) .97 (.91-1.04) 1.06 (.99-1.14)
% of Cross-gender friends
27
1.01*** (1.00-1.01) 1.01*** (1.00-1.01)
Broker*CGFs
28
1.00 (1.00-1.00) 1.00** (1.00- 1.00)
Control Variables
Grade 1.35*** (1.32-1.38) 1.35*** (1.32-1.38)
Race (ref: White)
Black .72*** (.65-.79) .71*** (.65-.77)
Hispanic 1.01 (.93-1.10) .86*** (.78-.94)
Asian .62*** (.54-.70) .53*** (.47-.60)
Other 1.16* (1.03-1.30) 1.06 (.96-1.18)
Parent’s edu (ref: No HS)

99
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)
Table 4.4 shows no significant association between a broker’s long ties and the likelihood
of alcohol consumption for both male and female groups when controlling the effects of cross-
gender friendships. When adolescents have no cross-gender friendships or all of their friends are
the same gender, the likelihood of alcohol consumption does not substantially matter depending
on whether adolescents connect clusters or not. Simply speaking, a broker’s long ties solely do
not affect the likelihood of alcohol consumption. However, both male and female non-brokers
were more likely to report alcohol consumption as their cross-gender friendships increase. For
example, the likelihood of alcohol consumption increases by about 0.53 percent for male “non-
brokers” (OR=1.01, 95% CI=1.00-1.01, p <.001; footnote 11) and about 0.59 percent for female
“non-brokers” (OR=1.01, 95% CI=1.00-1.01, p <.001; footnote 11) for every one percent
increase in their cross-gender friendships. These increases in alcohol consumption may sound
negligible, but are in fact significant in explaining alcohol consumption. Compared to male non-
brokers who had no cross-gender friendships (i.e., CGFs are zero), male non-brokers whose
friends were all female (i.e., CGFs are 100) were 53 percent more likely to drink alcohol. In the
same vein, female non-brokers whose friends were all male were 59 percent more likely to drink
High school graduate .97 (.87-1.08) .93 (.85-1.03)
College or higher degree .95 (.86-1.05) .88* (.80-.98)
Participation in sports 1.21*** (1.14-1.28) 1.18*** (1.12-1.24)
Participation in arts .76*** (.72-.81) .76*** (.72-.80)
Participation in academics .97 (.92-1.04) 1.04 (.99-1.10)
Total GPA .67*** (.64-.69) .61*** (.59-.64)
Private School (vs. Public) 1.19 (.92-1.54) 1.05 (.81-1.37)
Urbanity (ref: Urban)
Suburban 1.11 (.95-1.31) 1.16+ (.98-1.36)
Rural 1.15 (.91-1.46) .93 (.73-1.20)
Gender heterogeneity .30+ (.080-1.14) .0062 (.00001-3.45)
Intercept .30** (.14-.63) 3.08 (.13-75.37)
√

11
.33 .33
√𝜃
1.84 1.84
ρ .031 .031

100
alcohol compared to female non-brokers who had no cross-gender friends.
For male students, the effect of cross-gender friendships on alcohol consumption is the
same between brokers and non-brokers because the interaction term between broker scores and
CGFs is not statistically significant. However, the interaction term is significant and negative for
female students. This indicates that female brokers were less likely to drink alcohol, compared to
their female non-broker counterparts. More specifically, the likelihood of alcohol consumption
increases by 0.35 percent for female brokers, and 0.59 percent for female non-brokers when their
cross-gender friendship increases by one percent (footnote 12). Stated differently, female brokers
whose friends were all male were 35 percent more likely to drink alcohol compared to female
brokers who had no CGFs. Similarly, female non-brokers whose friends were all male were 59
percent more likely to drink alcohol compared to female non-brokers whose friends were all
same genders. Compared to the results in Appendix 4 Table D, there is no major difference in
significance or direction of these associations except female brokers; the odds ratios of female
brokers were positive in Table 4, whereas the odds ratios were negative in Appendix D. These
results may indicate that there was no substantial moderating effect of cross-gender friendships
on brokers’ alcohol use, but were some moderating effect on female brokers.
Table 4.5. Odds Ratios on the Likelihood of Cigarette Smoking by Gender (N = 61,608)

29
OR = 1.002506, 95% CI = 1.001063-1.003952 (Male); OR = 1.003289, 95% CI = 1.001854-1.004726 (Female)
30
OR = 1.001251, 95% CI = .9993984-1.003107 (Male); OR = .9997606, 95% CI = .9980064-1.001518 (Female)
Cigarette Use
Primary Predictors OR (95% CI) OR (95% CI)
Male Female
Broker (cluster connector) .92* (.86-.98) 1.00 (.93-1.07)
% of Cross-gender friends
29
1.00*** (1.00-1.00) 1.00*** (1.00-1.00)
Broker*CGFs
30
1.00 (1.00-1.00) 1.00 (1.00-1.00)
Control Variables
Grade 1.12*** (1.10-1.15) 1.14*** (1.12-1.17)
Race (ref: White)
Black .55*** (.50-.61) .39*** (.35-.42)

101
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)
Table 4.5 provides the results of how likely male and female students were to smoke
cigarettes in the last year depending on their brokerage status and their proportion of cross-
gender friendships. Male brokers without cross-gender friends were less likely to smoke
cigarettes (OR=.92, 95% CI=.86-.98, p <.05), whereas there is no substantial difference in
smoking behaviors between female brokers and female non-brokers when controlling the effect
of CGFs. Both male and female non-brokers were more likely to smoke cigarettes when they had
cross-gender friends. In Table 5, male non-brokers were 0.25 percent more likely to smoke
cigarettes (OR=1.00, 95% CI=1.00-1.00, p <.001; footnote 13), and female non-brokers were
0.33 percent more likely to smoke cigarettes (OR=1.00, 95% CI=1.00-1.00, p <.001) when their
cross-gender friendships increase by one percent. As explained above, these increases point to
significant differences between non-brokers who have all same sex peers and non-brokers who
have opposite sex peers; for example, male non-brokers whose friends were all female were 25
Hispanic .69** (.78-.95) .68*** (.62-.75)
Asian .72*** (.62-.84) .53*** (.46-.61)
Other 1.16** (1.05-1.29) .96 (.87-1.07)
Parent’s edu (ref: No HS)
High school graduate .91* (.82-1.00) .96 (.88-1.04)
College or higher degree .94 (.85-1.03) .90* (.83-.98)
Participation in sports .85*** (.80-.90) 1.00 (.95-1.06)
Participation in arts .92* (.86-.98) .76*** (.72-.81)
Participation in academics .96 (.90-1.03) .96 (.91-1.02)
Total GPA .60*** (.57-.63) .52*** (.50-.54)
Private School (vs. Public) 1.01 (.79-1.31) 1.01 (.75-1.35)
Urbanity (ref: Urban)
Suburban 1.13 (.96-1.32) 1.17+ (.98-1.41)
Rural 1.31* (1.04-1.65) 1.37* (1.05-1.80)
Gender heterogeneity .18** (.051-.64) .16 (.0025-9.99)
Intercept 1.84+ (.90-3.0 2.78 (.33-23.14)
√

11
.32 .37
√𝜃
1.82 1.81
ρ .030 .040

102
percent more likely to smoke cigarettes compared to male non-brokers who had no CGFs; and,
female non-brokers whose friends were all male were 33 percent more likely to smoke cigarettes
compared to female non-brokers who had only same gender friends.
The results show that there is no statistical differences in the likelihood of cigarette
smoking between brokers and non-brokers and no gender difference as well because the
interaction terms between brokers and CGFs are not statistically significant. Furthermore, there
is no moderating effect of cross-gender friendship for both genders, indicating that both male and
female brokers show an increased likelihood of smoking cigarettes, similar to their non-broker
counterparts. Although there is no change in direction or significance compared to the results of
models without the interaction term in Appendix 4 Table E, adding the interaction term leads to
increase in the odds ratios of male students’ cigarette use. This means that considering a broker’s
cross-gender friendship in the model may reinforce the direct effect between brokers’ long ties as
presented in Table 4.5.
Table 4.6. Odds Ratios on the Likelihood of Drunkenness by Gender (N = 61,608)

31
OR = 1.003397, 95% CI = 1.001863-1.004933 (Male); OR = 1.004496, 95% CI = 1.002963-1.006031 (Female)
32
OR = 1.001776, 95% CI = 1.000074 -1.003481 (Male); OR = .9997732, 95% CI = .9978535-1.001697 (Female)
Drunkenness
Primary Predictors OR (95% CI) OR (95% CI)
Male Female
Broker (cluster connector) .84*** (.78-.90) .92* (.85-.99)
% of Cross-gender friends
31
1.00*** (1.00-1.00) 1.00*** (1.00-1.01)
Broker*CGFs
32
1.00* (1.00-1.00) 1.00 (1.00-1.00)
Control Variables
Grade 1.46*** (1.43-1.50) 1.40*** (1.36-1.43)
Race (ref: White)
Black .61*** (.54-.68) .53*** (.47-.59)
Hispanic .93 (.83-1.03) .78*** (.69-.88)
Asian .48*** (.41-.55) .46*** (.39-.54)
Other 1.20** (1.06-1.37) .97 (.87-1.09)
Parent’s edu (ref: No HS)
High school graduate .87* (.78-.97) .90* (.82-.99)
College or higher degree .84** (.76-.94) .88* (.80-.97)

103
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

Table 4.6 presents the results of whether and how broker position and cross-gender
friendship help explain the likelihood of drunkenness. When students had no friends of the
opposite gender, both male and female brokers were less likely to have been drunk. For example,
male brokers were 16 percent less likely to have been drunk (OR=.84, 95% CI=.78-.90, p <.001)
and female brokers were 8 percent less likely to have been drunk (OR=.92, 95% CI=.85-.99,
p<.05) compared to their non-broker counterparts. For non-brokers, both male and female
students had a positive association between cross-gender friendships and drunkenness; male non-
brokers were 0.34 percent more likely to have been drunk as their cross-gender friendship
increases by one percent, indicating that male non-brokers whose friends were all opposite
gender were 34 percent more likely to being drunk compared to male non-brokers whose friends
were all same gender. Similarly, the likelihood of female non-brokers’ drunkenness increases by
0.45 percent for every percent increase in their CGF rate, indicating that female non-brokers
whose friends were all male were 45 percent more likely to being drunk than female non-brokers
whose friends were all female.
Although brokers who have no cross-gender friends were less likely to have been drunk
Participation in sports 1.15*** (1.08-1.22) 1.16*** (1.10-1.24)
Participation in arts .71*** (.66-.77) .65*** (.61-.69)
Participation in academics .97 (.91-1.04) 1.02 (.95-1.08)
Total GPA .64*** (.61-.66) .57*** (.55-.60)
Private School (vs. Public) 1.16 (.86-1.55) 1.06 (.77-1.44)
Urbanity (ref: Urban)
Suburban 1.25* (1.04-1.50) 1.32** (1.08-1.60)
Rural 1.16* (1.07-1.85) 1.20 (.90-1.60)
Gender heterogeneity .42 (.096-1.89) .0045+ (.000018-1.14)
Intercept .053*** (.023-.12) 1.02 (.062-16.95)
√

11
.37 .38
√𝜃
1.84 1.89
ρ .039 .039

104
as stated earlier, having cross-gender friendships leads brokers, particularly male brokers, to
have greater odds of having been drunk. For example, male brokers were 0.18 percent (0.34
percent for male non-brokers) more likely to have been drunk when their cross-gender friendship
increases by one percent (OR=1.00, 95% CI=1.00-1.00, p <.05; footnote 15). Put differently,
male brokers whose friends were all girls were 18 percent more likely to being drunk than male
brokers whose friends were all boys. When adding the interaction term between brokerage and
cross-gender friendship into the main model, there is no change of statistical significance or
direction compared to the results in Appendix F. However, it is worth noting that adding the
interaction term impacts male students more than female students; for example, the odds ratios of
male students were 12 percent more likely to being drunk in the model without the interaction
term and 16 percent in the model including the interaction term as their CGFs increase by one
percent. On the other hand, female students did not show any substantial changes after including
the interaction term in the model.
Overall, the findings suggest that those who have long ties have a negative association
with cigarette use and drunkenness when students have no opposite gender friends. Additionally,
this negative association is more pronounced among male broker students compared to their
female counterparts. Although there are differences by gender and broker status, it is clear that
having cross-gender friends results in an increase in the likelihood of substance use behaviors.
Lastly, cross-gender friendship was found to moderate male students’ drunkenness.
In addition to the main findings, there are a few variables that have a significant
relationship with substance use behaviors. Above all, students seem to engage in more
substances as their year in school increases. Racial and ethnic groups show significantly different
likelihoods of substance use; most of the minority groups reported lower odds of substance use

105
compared to their White counterparts. Not surprisingly, students whose parents have higher
educational attainment such as high school or college degrees are less likely to engage in
substance use behaviors compared to those whose parents have no high school diploma. A
student’s GPAs is negatively associated with the likelihood of substance use. While participating
in arts activities is negatively associated with substance use, sport and academic activities seem
to be different; for example, sport actives are positively associated with alcohol consumption or
having been drunk for both genders, but there is no substantial difference in substance use
behaviors between academic activity participants and non-participants. Lastly, students from
rural areas reported higher chances of smoking cigarettes and students from suburban areas are
more likely to have been drunk compared to urban students.

Conclusion
Recent advancements in social network analyses have helped explain whether and how a
network position is associated with substance use behaviors (Henneberger, Mushonga, and
Preston 2021; Osgood et al. 2014; Valente, Unger, and Johnson 2005; Valente, Gallaher, and
Mouttapa 2004; Fujimoto and Valente 2012). Based on extensive research on the roles of brokers
in receiving or transmitting favorable information and behaviors, this study questions whether
and how brokers can also engage in risky behaviors like substance use. Theoretically, the
structural aspect of brokers’ ties—long ties—helps them reach out to distant individuals so that
they can have greater chances of access to substances or people who have or use substances.
Additionally, the relational aspect of brokers—heterogeneous ties such as cross-gender
friendships—help introduce and stimulate experimental behaviors to adolescent brokers (Kreager
and Haynie 2011; Kreager, Haynie, and Hopfer 2013). Ironically, brokers’ long ties can also

106
have greater access to peers who do not use substances or have negative norms or attitudes
against substances and may not show substance use behaviors. In addition, risky behaviors may
require higher thresholds to be adopted than do favorable behaviors and need additional
conditions for adoption and brokers may therefore not show substance use behaviors without
special conditions (Centola and Macy 2007).
Multilevel logistic regression analyses showed the mixed results of the relationship
between brokers’ long ties and the likelihood of substance use. While brokers’ long ties did not
have statistically significant associations with the likelihood of alcohol use for both genders,
these long ties showed negative associations with the likelihood of drunkenness for both gender
and male students’ smoking. These findings indicate that brokers’ long ties do not necessarily
provoke their substance use behavior; instead, students who have long ties are less likely to be
drunk and male students especially are also less likely to smoke cigarettes compared to those
who do not have long ties. Regardless, there are a few possible explanations for this unique role
of brokers’ long ties on substance use. As stated above, brokers’ long ties may promote access
not only to people who use or have substances but also to people who do not use or have
negative attitudes or norms against substances. Generally, many people in networks are
substance nonusers and therefore students who have long ties may have more influences from
substance nonusers than substance users. Another explanation is that their long ties may just be
too weak to impact the likelihood of brokers’ substance use behaviors. It is also possible that
risky behaviors like substance use require high peer exposures (high thresholds) such as more
supports and confirmations from many immediate peer connections to rationalize adoption of
risky behaviors. Accordingly, loosely connected with “friends of friends” (or, friends of friends
of friends) or rare contacts with distant individuals may not help enhance credibility or

107
legitimacy of substance use behaviors and therefore long ties may not impact the likelihood of
brokers’ substance use.
The role of cross-gender friendships also seemed to have mixed results on adolescent
brokers’ substance use behaviors. The results showed that having cross-gender friends increased
the likelihood of substance use for both brokers and non-brokers, and these incremental paces
were not significantly different between brokers and non-brokers. However, there are exceptions
for female brokers’ alcohol use and male brokers’ drunkenness. For example, female brokers
were more likely to drink alcohol as numbers of their cross-gender friends increased, but their
incremental level of odds ratio was slower than that of female non-brokers. Male brokers were
also more likely to have been drunk with an increase of cross-gender friends, but the increased
rate was bigger compared to that of male non-brokers. In addition to these findings, this study
identified the role of cross-gender friendship as a moderating factor for male students’ smoking
and being drunk. The results indicated that direct effects of brokers’ long ties on smoking and
being drunk for boys became stronger by the inclusion of the interaction term. Put differently,
the protective effects of brokers’ long ties, which reduced the likelihood of smoking and being
drunk for boys, became bigger when male brokers had cross-gender friends. Overall, cross-
gender friends for brokers play key roles that not only increase the likelihood of substance use
behaviors but also reinforce protective effects of brokers’ long ties on substance use. For
continuing the earlier discussion about brokers’ long ties on substance use, this conclusion is not
surprising because cross-gender friendships are brokers’ immediate ties that may have stronger
impact on brokers’ risky behaviors than their long ties.
Given the diverse levels of adolescent substance use, I suggest that future research may
need to further examine different levels of alcohol and cigarette use and drunkenness regarding

108
the structural and relational aspect of brokers. The primary dependent variables in this study
were binary variables that indicate whether adolescents have engaged in drinking alcohol,
smoking cigarettes, and being drunk or not in the past 12 months. Considering that the sample
from Add Health data used in this study is the large size with multilevel datasets and the fact that
I also applied the multiple imputation procedure to the datasets, it is not practically feasible to
conduct multilevel multinomial logistic regression analyses. Although the multilevel logistic
regression analyses have been successfully conducted and the results of the analyses are easily
interpreted, little is known about whether and how brokers’ long and heterogeneous ties
differently explain substance use behaviors depending on frequencies of substance use.
My second suggestion is to test the research questions using more recent datasets.
Although Add Health data is the best fit to explore my research questions, the questionnaires
were collected in 1994-1995. I believe that this adolescent sample collected almost three decades
ago can still provide meaningful relationships between brokers and adolescent substance use, but
using more recent data will be able to help suggest implications and develop public policies that
serve adolescent health today. More importantly, students were asked their “male” friends first
then “female” friends subsequently, which may cause potential biases for this purpose of the
study. Accordingly, it is expected to see any different or better results if a future study uses a
dataset that collects both female and male friends equally.
Third, future research should consider a longitudinal analysis to further identify a causal
relationship between broker positions and substance use. The present study used the single wave
(Wave 1) to explore the research questions. However, considering the high decay rates of broker
positions over time (Kilduff, Burt, Tasselli 2013, p. 19-20), it is worth investigating whether and
how the relationship between broker position and substance use can be changed (or stable) over

109
time.
In spite of these limitations, this current study attempted to disentangle a potential
mechanism of whether and how the structural and relational aspect of brokers explains the
likelihood of substance use using the national representative sample and theoretical foundations.
Based on classical and recent theories of brokers on risky behaviors, the present study highlights
the important role of brokers’ immediate ties, particularly ties with cross-gender friendships, in
explaining their substance use behaviors. Accordingly, the findings of this study contribute to the
growing body of literature about a broker’s role on non-favorable or risky behaviors to help
design effective prevention and intervention programs for adolescent substance use.

110
CHAPTER 5
CONCLUSION

Summary
Granovetter’s “Strength of Weak Ties (Granovetter 1973)” is still one of the most-cited
articles in the social sciences (Healy 2014). His idea—those who loosely connect with distant
others have more access to resources and information—is straightforward and simple, but
groundbreaking in the field of social networking. Researchers had long believed in the power of
strong ties and of those who occupy central or popular positions in social networks. However,
Granovetter’s weak ties theory introduced the importance of a broker as a gatekeeper, liaison,
mediator, itinerant, or coordinator and highlighted the benefits of access to diverse resources and
information (Granovetter 1977; Gould and Fernandez 1989). Since Granovetter’s introduction of
weak ties, scholars have also come to recognize the potential costs of the broker position.
Because brokers loosely connect with distant and separate individuals, their social ties are often
unstable and precarious (Burt 2002a; Barnes et al. 2016; Aral 2016). High volatility and decay
rates of a broker’s ties may result in diminishing social capital and even promoting a willingness
of taking risky behaviors to be accepted by others. Furthermore, there is a lack of robust
discussion on who occupies this (dis)advantageous position and what environments help brokers
emerge in friendship networks. Accordingly, the main purpose of this dissertation is to address
the potential characteristics of brokers and the environments that promote the emergence of
brokers. This dissertation also examines whether and how brokers show a high risk for engaging
in risky behaviors, particularly substance use behaviors.
The second chapter of this dissertation describes the general characteristics of brokers
and how extracurricular activities help (or hinder) adolescents in occupying broker positions.

111
The results indicate that racial/ethnic minority students (except Asian students), students from
low socio-economic status (SES) family backgrounds, and student who abstain from sports but
participate in arts are highly correlated characteristics of potential brokers. Although high status
is conceptualized differently depending on the environment of each school, the general
perception of high-status adolescents—for example, affluent, athletic students—help us
understand that brokers are not necessarily always high-status adolescents. This finding supports
the assumption from social exchange theory that high-status adolescents do not aim to be brokers
and instead prefer to build social closure with similar others. While high-status adolescents can
perceive no or little benefits from being brokers, racial/ethnic minority students or disadvantaged
status students may serve as brokers due to the potential benefits from connecting with dissimilar
others including high-status peers.
This chapter also supports the idea that extracurricular activities function as a social place
for the emergence of brokers, with the caveat that the effect of this opportunity varies depending
on a student’s race and ethnicity. Because students participate in sport, art, and academic
activities at different rates, some activities provide more chances to broker development
depending on race and ethnicity. Moreover, each type of ECAs promote status-leveling effects,
which mitigate racial/ethnic biases and stereotypes toward a certain race or ethnicity, differently
depending on a student’s race/ethnicity. In sum, participating in ECAs generally helps increase
the chance of being brokers, but each type of ECA differs in its impact on generating more (or
fewer) chances to being connected with different others depending on a student’s race or
ethnicity.
The third chapter explores whether brokers are at a higher risk of using alcohol and
cigarettes. Based on contagion theory and the social acceptance hypothesis, I examine the

112
association of the number and directionality of a broker’s ties with the likelihood of alcohol and
cigarette use. The results present no linearly significant relationship between the number/
directionality of brokers’ ties and their alcohol and cigarette use. Furthermore, the results from
the marginal effects analyses do not find any evidence of Centola and Macy’s contagion theory,
for both alcohol and cigarette use behaviors. Potential brokers or high-brokerage groups tend to
use less alcohol and cigarettes as their number of friends increase. When the number of
substance-using peers increase for potential brokers, they are also less likely to use substances.
Instead, the results provide the supporting evidence of the social acceptance explanation in
understanding a broker’s substance use, especially cigarette use behavior. This chapter presents
that the number of outdegree connections (i.e., nominating others as friends) for potential brokers
is positively associated with cigarette smoking, indicating that brokers who seek more
friendships are more likely to try cigarette smoking. On the other hand, potential brokers do not
necessarily engage in substance use when they receive many friendship nominations.
Conversely, these popular brokers who are nominated by many other peers are less likely to try
cigarette smoking and alcohol drinking.
The fourth chapter examines whether and how the structural (long ties) and relational
(heterogeneous ties) characteristics of a broker help explain the likelihood of a broker’s alcohol
and cigarette use. Not surprisingly, this chapter identifies different roles of long ties and gender
heterogeneous ties. Brokers’ long ties serve as a protective function against alcohol abuse for
both genders and cigarette smoking for boys. On the other hand, heterogeneous ties, cross-gender
friendships in this chapter, are positively associated with alcohol and cigarette use for both
gender groups and for brokers and non-brokers. This indicates that both boys and girls are more
likely to use alcohol and cigarettes as they have more friends of opposite genders. Additionally,

113
there is no substantial difference in the effects of cross-gender friendships between brokers and
non-brokers in explaining cigarette smoking. However, the association between cross-gender
friendships and alcohol abuse is stronger for male brokers than female brokers. Thus, these
results support the hypothesis that long ties do not necessarily provoke brokers to engage in
substance use. Instead, brokers’ heterogeneous ties tend to promote the likelihood of drinking
alcohol and smoking cigarettes, but their alcohol and cigarette use behaviors should be discussed
in consideration of gender differences.

Limitations
As I completed my dissertation, I encountered several issues that may be addressed in
future studies. This dissertation addresses the potential characteristics of adolescent brokers and
potential mechanisms that explain whether and how brokers use substances. This dissertation
relied on cross-sectional data (Wave 1 of Add Health data for chapters 2 and 4; Wave 4 of the
Social Network Study data for chapter 3) that identifies adolescent brokers and examines their
substance use behaviors at a single point in time. Considering that a broker’s ties are often
unstable and their decay rates are high over time (Burt, Kilduff, and Tasselli 2013; Burt 2002a),
using a single wave of the datasets may lead to limited conclusions in understanding adolescent
brokers and their behavior patterns. More importantly, cross-sectional examinations lead me to
carefully interpret the results solely as associations rather than causal relations. This indicates
that it is still uncertain whether extracurricular activities increase chances of becoming brokers. It
is also unclear whether becoming a broker increases the probability of using substances, or if
using substances causes an individual to become a broker. In this context, a longitudinal analysis
would allow us to understand and expand the findings in this dissertation as causal relationships.

114
For example, using panel data helps researchers further examine why and how brokers emerge
and disappear regarding individual- and school-factors. Furthermore, a longitudinal analysis
would also help us explore the primary cause of adolescent substance use and better understand
the mechanisms that explain the key factors surrounding broker’s substance use behaviors.
Second, a binary response for substance use may lead to limitations in exploring the
likelihood of a broker’s substance use. In the third and fourth chapter, the dependent variables
are binary measures that indicate whether adolescents have engaged in drinking alcohol and
smoking cigarettes in the past 30 days (in the third chapter) or past 12 months (in the fourth
chapter). I conducted logistic regression analyses for the third chapter and multilevel logistic
regression analyses for the fourth chapter because it is not practically feasible to conduct
multilevel multinomial logistic regressions considering large number of missing cases in such a
large sample size. Using binary variables provides valid results because many of the adolescents
responded that they did not drink or smoke, and I also included “drunkenness” in the fourth
chapter to see any different relationships between brokers and excessive alcohol use. However,
the results may not sufficiently capture how brokerage is associated with diverse levels of
adolescent substance use. In particular, adolescents were asked about their substance use
behaviors during the past 12 months in the fourth chapter and there might be some differences
between one-time-users who used alcohol or cigarettes once and heavy users reporting daily use
of alcohol or cigarettes in the past year. Accordingly, future studies should further consider
nominal outcome variables such as frequencies of substance use to detect the associations
between a broker and the likelihood of alcohol and cigarette use more thoroughly.
Lastly, I should note that future studies would benefit from using different adolescent
samples and friendship networks to validate the conclusions from this dissertation. In this

115
dissertation, the second and the fourth chapter use Add Health data—a nationally representative
sample—and the third chapter uses the Social Networking Study data—a sample of students
from five Southern California high schools within the same school district. If I further examine
potential characteristics of brokers as I did in the second chapter using the Social Networking
Study data, results will help reveal whether “status” is still a salient factor for understanding the
characteristics of adolescent brokers. One of the findings from the second chapter—high-status
adolescents do not serve broker positions—gives rise to one critical question: does a standard
definition of high and low-status apply uniformly across a heterogenous population of schools?
Social status among adolescents may differ by school environments or student populations; what
may be considered “high-status” at one school could be fundamentally different at a different
school. However, the second chapter did not closely examine the different interpretations of
high-status or low-status. Instead, I relied on a general perception of high-status adolescents,
such as being White, affluent, and athletic, but this definition may not work for all U.S. schools.
However, using different datasets such as the Social Networking Study data, which pulls from a
demographically dissimilar population, could help affirm (or rebut) the conclusion about the
relationship between a student’s status and the likelihood of being a broker.
Furthermore, using more recent datasets may help utilize the findings for adolescent
substance use programs and policies. Above all, the results from the wave 1 of Add Health data,
collected in 1994-1995, may not be able to reflect substance use behaviors of adolescents today
because the prevalence and patterns of their substance use behaviors have been changed over last
few decades. For example, adolescents today tend to build online friendships more than before
and these online friendships can have a strong impact on their behavior. Although alcohol and
cigarettes are still the most-used substances among adolescents, their prevalence rates are

116
steadily decreasing over time; but, more recently adolescents and young adults tend to use other
substances such as nicotine vaping and marijuana (Johnston et al. 2021; Anderson et al. 2019;
McKeganey and Barnard 2018). Thus, it is necessary to validate the results from this dissertation
using different adolescent samples and friendship network data.

117
REFERENCES
Allen, Joseph P., Maryfrances R. Porter, and F. Christy McFarland. 2006. “Leaders and
Followers in Adolescent Close Friendships: Susceptibility to Peer Influence as a
Predictor of Risky Behavior, Friendship Instability, and Depression.” Development and
Psychopathology 18 (1): 155–72.

Allport, Gordon. 1954. The Nature of Prejudice. Cambridge, MA: Addison-Wesley.

Almaatouq, Abdullah, Laura Radaelli, Alex Pentland, and Erez Shmueli. 2016. “The Role of
Reciprocity and Directionality of Friendship Ties in Promoting Behavioral Change.” In
International Conference on Social Computing, Behavioral-Cultural Modeling and
Prediction and Behavior Representation in Modeling and Simulation, 33–41. Springer.

Anderson, D. Mark, Benjamin Hansen, Daniel I. Rees, and Joseph J. Sabia. 2019. “Association
of Marijuana Laws with Teen Marijuana Use: New Estimates from the Youth Risk
Behavior Surveys.” JAMA Pediatrics 173 (9): 879–81.

Aral, Sinan. 2016. “The Future of Weak Ties.” American Journal of Sociology 121 (6): 1931–39.

Arndorfer, Cara Lee, and Elizabeth A. Stormshak. 2008. “Same-Sex versus Other-Sex Best
Friendship in Early Adolescence: Longitudinal Predictors of Antisocial Behavior
throughout Adolescence.” Journal of Youth and Adolescence 37 (9): 1059–70.

Barnes, Michele, Kolter Kalberg, Minling Pan, and PingSun Leung. 2016. “When Is Brokerage
Negatively Associated with Economic Benefits? Ethnic Diversity, Competition, and
Common-Pool Resources.” Social Networks 45 (March): 55–65.
https://doi.org/10.1016/j.socnet.2015.11.004.

Barnes-Mauthe, Michele, Steven Allen Gray, Shawn Arita, John Lynham, and PingSun Leung.
2015. “What Determines Social Capital in a Social–Ecological System? Insights from a
Network Perspective.” Environmental Management 55 (2): 392–410.

Bearman, Peter S., James Moody, and Katherine Stovel. 2004. “Chains of Affection: The
Structure of Adolescent Romantic and Sexual Networks 1.” American Journal of
Sociology 110 (1): 44–91.

Berger, Christian, and Jan Kornelis Dijkstra. 2013. “Competition, Envy, or Snobbism? How
Popularity and Friendships Shape Antipathy Networks of Adolescents.” Journal of
Research on Adolescence 23 (3): 586–95.

Blau, Peter Michael. 1977. Inequality and Heterogeneity: A Primitive Theory of Social Structure.
Vol. 7. Free Press New York.

Branch, Curtis W., Priti Tayal, and Carla Triplett. 2000. “The Relationship of Ethnic Identity and
Ego Identity Status among Adolescents and Young Adults.” International Journal of
Intercultural Relations 24 (6): 777–90.

118

Brashears, Matthew E. 2008. “Gender and Homophily: Differences in Male and Female
Association in Blau Space.” Social Science Research 37 (2): 400–415.

Briggs, Xavier de Souza. 2002. “Bridging Networks, Social Capital, and Racial Segregation in
America.”

———. 2007. “‘Some of My Best Friends Are…’: Interracial Friendships, Class, and
Segregation in America.” City & Community 6 (4): 263–90.

Brown, B. Bradford, and Erin L. Dietz. 2009. “Informal Peer Groups in Middle Childhood and
Adolescence.” Handbook of Peer Interactions, Relationships, and Groups, 361–76.

Brown, Deborah Wright, and Alison M. Konrad. 2001. “Granovetter Was Right: The Importance
of Weak Ties to a Contemporary Job Search.” Group & Organization Management 26
(4): 434–62.

Burt, Ronald S. 1992. Structural Holes: The Social Structure of Competition. Harvard university
press.

———. 2002a. “Bridge Decay.” Social Networks 24 (4): 333–63.

———. 2002b. “The Social Capital of Structural Holes.” The New Economic Sociology:
Developments in an Emerging Field, 148–90.

Burt, Ronald S. 2004. “Structural Holes and Good Ideas.” American Journal of Sociology 110
(2): 349–99.

Burt, Ronald S., Martin Kilduff, and Stefano Tasselli. 2013. “Social Network Analysis:
Foundations and Frontiers on Advantage.” Annual Review of Psychology 64: 527–47.

Carley, Kathleen M. 2003. Dynamic Network Analysis. na.

Centola, Damon, and Michael Macy. 2007. “Complex Contagions and the Weakness of Long
Ties.” American Journal of Sociology 113 (3): 702–34. https://doi.org/10.1086/521848.

Clotfelter, Charles T. 2002. “Interracial Contact in High School Extracurricular Activities.” The
Urban Review 34 (1): 25–46.

Coleman, James. 1988. “Social Capital in the Creation of Human Capital.” American Journal of
Sociology 94: 95–120.

Coleman, James, Thomas Hoffer, and Sally Kilgore. 1982. High School Achievement: Public,
Catholic, and Private Schools Compared. New York: Basic Books.

Copeland, Molly, Jacob C. Fisher, James Moody, and Mark E. Feinberg. 2018. “Different Kinds
of Lonely: Dimensions of Isolation and Substance Use in Adolescence.” Journal of Youth
and Adolescence 47 (8): 1755–70.

119

Cornwell, Benjamin. 2009. “Good Health and the Bridging of Structural Holes.” Social Networks
31 (1): 92–103.

Crowell, Linda F. 2004. “Weak Ties: A Mechanism for Helping Women Expand Their Social
Networks and Increase Their Capital.” The Social Science Journal 41 (1): 15–28.

Dempster, Steve. 2011. “I Drink, Therefore I’m Man: Gender Discourses, Alcohol and the
Construction of British Undergraduate Masculinities.” Gender and Education 23 (5):
635–53.

DiMaggio, Paul, and Filiz Garip. 2011. “How Network Externalities Can Exacerbate Intergroup
Inequality.” American Journal of Sociology 116 (6): 1887–1933.
https://doi.org/10.1086/659653.

———. 2012. “Network Effects and Social Inequality.” Annual Review of Sociology 38 (1): 93–
118. https://doi.org/10.1146/annurev.soc.012809.102545.

Dunphy, Dexter C. 1963. “The Social Structure of Urban Adolescent Peer Groups.” Sociometry,
230–46.

Eder, Donna. 1985. “The Cycle of Popularity: Interpersonal Relations among Female
Adolescents.” Sociology of Education, 154–65.

Eder, Donna, and David A. Kinney. 1995. “The Effect of Middle School Extra Curricular
Activities on Adolescents’ Popularity and Peer Status.” Youth & Society 26 (3): 298–324.

Elpus, Kenneth, and Carlos R. Abril. 2011. “High School Music Ensemble Students in the
United States: A Demographic Profile.” Journal of Research in Music Education 59 (2):
128–45.

Ennett, Susan T., and Karl E. Bauman. 1993. “Peer Group Structure and Adolescent Cigarette
Smoking: A Social Network Analysis.” Journal of Health and Social Behavior, 226–36.

Ensminger, Margaret E., Christopher B. Forrest, Anne W. Riley, Myungsa Kang, Bert F. Green,
Barbara Starfield, and Sheryl A. Ryan. 2000. “The Validity of Measures of
Socioeconomic Status of Adolescents.” Journal of Adolescent Research 15 (3): 392–419.

Everett, Martin G., and Thomas W. Valente. 2016. “Bridging, Brokerage and Betweenness.”
Social Networks 44 (January): 202–8. https://doi.org/10.1016/j.socnet.2015.09.001.

Fagbule, O. F., K. K. Kanmodi, V. O. Samuel, T. O. Isola, E. O. Aliemeke, M. E. Ogbeide, K. E.
Ogunniyi, L. A. Nnyanzi, H. O. Adewuyi, and F. B. Lawal. 2021. “Prevalence and
Predictors of Cigarette Smoking and Alcohol Use among Secondary School Students in
Nigeria.” Annals of Ibadan Postgraduate Medicine 19 (2): 112–23.

120
Fang, Xiaoyi, Xiaoming Li, Bonita Stanton, and Qi Dong. 2003. “Social Network Positions and
Smoking Experimentation among Chinese Adolescents.” American Journal of Health
Behavior 27 (3): 257–67.

Fink, Clay, Aurora Schmidt, Vladimir Barash, Christopher Cameron, and Michael Macy. 2016.
“Complex Contagions and the Diffusion of Popular Twitter Hashtags in Nigeria.” Social
Network Analysis and Mining 6 (1): 1.

Finkelstein, Daniel M., Laura D. Kubzansky, and Elizabeth Goodman. 2006. “Social Status,
Stress, and Adolescent Smoking.” Journal of Adolescent Health 39 (5): 678–85.

Fleming, Lee, Santiago Mingo, and David Chen. 2007. “Collaborative Brokerage, Generative
Creativity, and Creative Success.” Administrative Science Quarterly 52 (3): 443–75.

Freeman, Linton C. 1977. “A Set of Measures of Centrality Based on Betweenness.” Sociometry
40 (1): 35–41. https://doi.org/10.2307/3033543.

———. 1978. “Segregation in Social Networks.” Sociological Methods & Research 6 (4): 411–
29.

Fujimoto, Kayo, and Thomas W Valente. 2012. “Decomposing the Components of Friendship
and Friends’ Influence on Adolescent Drinking and Smoking.” The Journal of Adolescent
Health 51 (2): 136–43. https://doi.org/10.1016/j.jadohealth.2011.11.013.

Gaughan, Monica. 2006. “The Gender Structure of Adolescent Peer Influence on Drinking.”
Journal of Health and Social Behavior 47 (1): 47–61.

Ghasemiesfeh, Golnaz, Roozbeh Ebrahimi, and Jie Gao. 2013. “Complex Contagion and the
Weakness of Long Ties in Social Networks: Revisited.” In Proceedings of the Fourteenth
ACM Conference on Electronic Commerce, 507–24.

Ghavami, Negin, and Rashmita S. Mistry. 2019. “Urban Ethnically Diverse Adolescents’
Perceptions of Social Class at the Intersection of Race, Gender, and Sexual Orientation.”
Developmental Psychology 55 (3): 457.

Gile, Krista J., and Mark S. Handcock. 2017. “Analysis of Networks with Missing Data with
Application to the National Longitudinal Study of Adolescent Health.” Journal of the
Royal Statistical Society: Series C (Applied Statistics) 66 (3): 501–19.

Goodman, Elizabeth, Nancy E. Adler, Ichiro Kawachi, A. Lindsay Frazier, Bin Huang, and
Graham A. Colditz. 2001. “Adolescents’ Perceptions of Social Status: Development and
Evaluation of a New Indicator.” Pediatrics 108 (2): e31–e31.

Goodreau, Steven M., James A. Kitts, and Martina Morris. 2009. “Birds of a Feather, or Friend
of a Friend? Using Exponential Random Graph Models to Investigate Adolescent Social
Networks.” Demography 46 (1): 103–25.

121
Gould, Roger V., and Roberto M. Fernandez. 1989. “Structures of Mediation: A Formal
Approach to Brokerage in Transaction Networks.” Sociological Methodology 19: 89–
126. https://doi.org/10.2307/270949.

Granovetter, Mark. 1973. “The Strength of Weak Ties.” American Journal of Sociology 78 (6):
1360–80.

———. 1977. “The Strength of Weak Ties.” In Social Networks, 347–67. Elsevier.

Grard, Adeline, Anton Kunst, Mirte Kuipers, Matthias Richter, Arja Rimpela, Bruno Federico,
and Vincent Lorant. 2018. “Same-Sex Friendship, School Gender Composition, and
Substance Use: A Social Network Study of 50 European Schools.” Substance Use &
Misuse 53 (6): 998–1007.

Haas, Steven A., David R. Schaefer, and Olga Kornienko. 2010. “Health and the Structure of
Adolescent Social Networks.” Journal of Health and Social Behavior 51 (4): 424–39.

Hallinan, Maureen T., and Ruy A. Teixeira. 1987. “Students’ Interracial Friendships: Individual
Characteristics, Structural Effects, and Racial Differences.” American Journal of
Education 95 (4): 563–83.

Healy, Kieran. 2014. “Sociology’s Most Cited Papers by Decade.” November 15, 2014.
https://kieranhealy.org/blog/archives/2014/11/15/top-ten-by-decade/.

Henneberger, Angela K., Dawnsha R. Mushonga, and Alison M. Preston. 2021. “Peer Influence
and Adolescent Substance Use: A Systematic Review of Dynamic Social Network
Research.” Adolescent Research Review 6 (1): 57–73.

Henry, David B., and Kimberly Kobus. 2007. “Early Adolescent Social Networks and Substance
Use.” The Journal of Early Adolescence 27 (3): 346–62.

Holland, Megan M. 2012. “Only Here for the Day: The Social Integration of Minority Students
at a Majority White High School.” Sociology of Education 85 (2): 101–20.

Huang, Grace C., Daniel Soto, Kayo Fujimoto, and Thomas W. Valente. 2014. “The Interplay of
Friendship Networks and Social Networking Sites: Longitudinal Analysis of Selection
and Influence Effects on Adolescent Smoking and Alcohol Use.” American Journal of
Public Health 104 (8): e51–59.

Hussong, Andrea M. 2002. “Differentiating Peer Contexts and Risk for Adolescent Substance
Use.” Journal of Youth and Adolescence 31: 207–20.

Ingels, Steven J., Ben W. Dalton, and Laura LoGerfo. 2008. “Trends among High School
Seniors, 1972-2004. NCES 2008-320.” National Center for Education Statistics.

Ispa-Landa, Simone. 2013. “Gender, Race, and Justifications for Group Exclusion: Urban Black
Students Bussed to Affluent Suburban Schools.” Sociology of Education 86 (3): 218–33.

122
Johnston, Lloyd D., Richard A. Miech, Patrick M. O’Malley, Jerald G. Bachman, John E.
Schulenberg, and Megan E. Patrick. 2021. “Monitoring the Future National Survey
Results on Drug Use, 1975-2020: Overview, Key Findings on Adolescent Drug Use.”
Institute for Social Research.

Joyner, Kara, and Grace Kao. 2000. “School Racial Composition and Adolescent Racial
Homophily.” Social Science Quarterly, 810–25.

Kalmijn, Matthijs. 2002. “Sex Segregation of Friendship Networks. Individual and Structural
Determinants of Having Cross-Sex Friends.” European Sociological Review 18 (1): 101–
17.

Kao, Grace, and Kara Joyner. 2004. “Do Race and Ethnicity Matter among Friends? Activities
among Interracial, Interethnic, and Intraethnic Adolescent Friends.” Sociological
Quarterly 45 (3): 557–73.

Killeya-Jones, Ley A., Ryo Nakajima, and Philip R. Costanzo. 2007. “Peer Standing and
Substance Use in Early-Adolescent Grade-Level Networks: A Short-Term Longitudinal
Study.” Prevention Science 8: 11–23.

Kreager, Derek A., and Dana L. Haynie. 2011. “Dangerous Liaisons? Dating and Drinking
Diffusion in Adolescent Peer Networks.” American Sociological Review 76 (5): 737–63.

Kreager, Derek A., Dana L. Haynie, and Suellen Hopfer. 2013. “Dating and Substance Use in
Adolescent Peer Networks: A Replication and Extension.” Addiction 108 (3): 638–47.

Lee, Chih-Ting, Tsai-Wei Chen, Yi-Fang Yu, Carol Strong, Chung-Ying Lin, Yun-Hsuan
Chang, Yi-Ping Hsieh, Yi-Ching Lin, Josue Jaru Ubeda Herrera, and Meng-Che Tsai.
2022. “Reciprocal Peer Network Processes on Substance Use and Delinquent Behavior in
Adolescence: Analysis from a Longitudinal Youth Cohort Study.” International Journal
of Mental Health and Addiction, 1–14.

Lee, Stacey J. 2015. Unraveling the" Model Minority" Stereotype: Listening to Asian American
Youth. 2nd ed. Teachers College Press.

Lin, Nan. 1999. “Building a Network Theory of Social Capital.” Connections 22 (1): 28–51.
———. 2002. Social Capital: A Theory of Social Structure and Action. Vol. 19. Cambridge
university press.

Lundborg, Petter. 2006. “Having the Wrong Friends? Peer Effects in Adolescent Substance
Use.” Journal of Health Economics 25 (2): 214–33.

Mahoney, Joseph L., Deborah Lowe Vandell, Sandra Simpkins, and Nicole Zarrett. 2009.
“Adolescent Out-of-School Activities.” In Handbook of Adolescent Psychology:
Contextual Influences on Adolescent Development, Vol. 2, 3rd Ed, 228–69. Hoboken, NJ,
US: John Wiley & Sons Inc.

123
Malow-Iroff, Micheline S. 2006. “Cross-Sex Best Friendship Influences on Early Adolescent
Cigarette and Alcohol Expectancies and Use.” The Journal of Psychology 140 (3): 209–
27.

Mangino, William. 2009. “The Downside of Social Closure: Brokerage, Parental Influence, and
Delinquency Among African American Boys.” Sociology of Education 82 (2): 147–72.
https://doi.org/10.1177/003804070908200203.

Mathys, Cécile, William J. Burk, and Antonius HN Cillessen. 2013. “Popularity as a Moderator
of Peer Selection and Socialization of Adolescent Alcohol, Marijuana, and Tobacco
Use.” Journal of Research on Adolescence 23 (3): 513–23.

Mayhew, Bruce H., J. Miller McPherson, Thomas Rotolo, and Lynn Smith-Lovin. 1995. “Sex
and Race Homogeneity in Naturally Occurring Groups.” Social Forces 74 (1): 15–52.

McDonald, Steve. 2011. “What’s in the ‘Old Boys’ Network? Accessing Social Capital in
Gendered and Racialized Networks.” Social Networks 33 (4): 317–30.

McFarland, Daniel A., James Moody, David Diehl, Jeffrey A. Smith, and Reuben J. Thomas.
2014. “Network Ecology and Adolescent Social Structure.” American Sociological
Review 79 (6): 1088–1121.

McKeganey, Neil, and Marina Barnard. 2018. “Change and Continuity in Vaping and Smoking
by Young People: A Qualitative Case Study of a Friendship Group.” International
Journal of Environmental Research and Public Health 15 (5): 1008.

McNeal Jr, Ralph B. 1998. “High School Extracurricular Activities: Closed Structures and
Stratifying Patterns of Participation.” The Journal of Educational Research 91 (3): 183–
91.

McNeal, Ralph B. 1995. “Extracurricular Activities and High School Dropouts.” Sociology of
Education; Albany 68 (1): 62.

McPherson, Miller, Lynn Smith-Lovin, and James M. Cook. 2001a. “Birds of a Feather:
Homophily in Social Networks.” Annual Review of Sociology, 415–44.

McPherson, and Lynn Smith-Lovin. 1987. “Homophily in Voluntary Organizations: Status
Distance and the Composition of Face-to-Face Groups.” American Sociological Review
52 (3): 370. https://doi.org/10.2307/2095356.

McPherson, Lynn Smith-Lovin, and James M. Cook. 2001b. “Birds of a Feather: Homophily in
Social Networks.” Annual Review of Sociology 27: 415–44.

Meier, Ann, Benjamin Swartz Hartmann, and Ryan Larson. 2018. “A Quarter Century of
Participation in School-Based Extracurricular Activities: Inequalities by Race, Class,
Gender and Age?” Journal of Youth and Adolescence 47 (6): 1299–1316.

124
Mercken, Liesbeth, Tom AB Snijders, Christian Steglich, Erkki Vertiainen, and Hein De Vries.
2010. “Smoking-Based Selection and Influence in Gender-Segregated Friendship
Networks: A Social Network Analysis of Adolescent Smoking.” Addiction 105 (7):
1280–89.

Mickelson, Roslyn Arlin. 2001. “Subverting Swann: First-and Second-Generation Segregation in
the Charlotte-Mecklenburg Schools.” American Educational Research Journal 38 (2):
215–52.

Molloy, Lauren E., Scott D. Gest, Mark E. Feinberg, and D. Wayne Osgood. 2014. “Emergence
of Mixed-Sex Friendship Groups during Adolescence: Developmental Associations with
Substance Use and Delinquency.” Developmental Psychology 50 (11): 2449.

Molm, Linda D., and Karen S. Cook. 1995. “Social Exchange and Exchange Networks.”
Sociological Perspectives on Social Psychology 2: 209–35.

Molm, Linda D., David R. Schaefer, and Jessica L. Collett. 2007. “The Value of Reciprocity.”
Social Psychology Quarterly 70 (2): 199–217.

Montgomery, James D. 1992. “Job Search and Network Composition: Implications of the
Strength-Of-Weak-Ties Hypothesis.” American Sociological Review 57 (5): 586–96.
https://doi.org/10.2307/2095914.

Moody, James. 2001. “Race, School Integration, and Friendship Segregation in America.”
American Journal of Sociology 107 (3): 679–716. https://doi.org/10.1086/338954.

———. 2002. “The Importance of Relationship Timing for Diffusion.” Social Forces 81 (1):
25–56.

Morgan, P. R., and James M. McPartland. 1981. “The Extent of Classroom Segregation within
Desegregated Schools.”

Mouw, Ted. 2002. “Are Black Workers Missing the Connection? The Effect of Spatial Distance
and Employee Referrals on Interfirm Racial Segregation.” Demography 39 (3): 507–28.

———. 2009. “The Use of Social Networks among Hispanic Workers: An Indirect Test of the
Effect of Social Capital.” Unpublished Working Paper). Chapel Hill, NC: Department of
Sociology, University of North Carolina-Chapel Hill, 2010–04.

Mouw, Ted, and Barbara Entwisle. 2006. “Residential Segregation and Interracial Friendship in
Schools.” American Journal of Sociology 112 (2): 394–441.
https://doi.org/10.1086/506415.

Mrug, Sylvie, Casey Borch, and Antonius HN Cillessen. 2011. “Other-Sex Friendships in Late
Adolescence: Risky Associations for Substance Use and Sexual Debut?” Journal of
Youth and Adolescence 40 (7): 875–88.

125
Mullen, Kenneth, Jonathan Watson, Jan Swift, and David Black. 2007. “Young Men,
Masculinity and Alcohol.” Drugs: Education, Prevention and Policy 14 (2): 151–65.

Olds, R. Scott, and Dennis L. Thombs. 2001. “The Relationship of Adolescent Perceptions of
Peer Norms and Parent Involvement to Cigarette and Alcohol Use.” Journal of School
Health 71 (6): 223–28.

Osgood, D. Wayne, Mark E. Feinberg, Lacey N. Wallace, and James Moody. 2014. “Friendship
Group Position and Substance Use.” Addictive Behaviors 39 (5): 923–33.

Osgood, D. Wayne, Daniel T. Ragan, Lacey Wallace, Scott D. Gest, Mark E. Feinberg, and
James Moody. 2013. “Peers and the Emergence of Alcohol Use: Influence and Selection
Processes in Adolescent Friendship Networks.” Journal of Research on Adolescence 23
(3): 500–512.

Payne, Danielle C., and Benjamin Cornwell. 2007. “Reconsidering Peer Influences on
Delinquency: Do Less Proximate Contacts Matter?” Journal of Quantitative Criminology
23 (2): 127–49.

Pearson, Michael, Helen Sweeting, Patrick West, Robert Young, Jacki Gordon, and Katrina
Turner. 2006. “Adolescent Substance Use in Different Social and Peer Contexts: A Social
Network Analysis.” Drugs: Education, Prevention and Policy 13 (6): 519–36.

Poulin, François, and Alessandra Chan. 2010. “Friendship Stability and Change in Childhood
and Adolescence.” Developmental Review 30 (3): 257–72.

Poulin, François, Anne-Sophie Denault, and Sara Pedersen. 2011. “Longitudinal Associations
between Other-Sex Friendships and Substance Use in Adolescence.” Journal of Research
on Adolescence 21 (4): 776–88.

Quillian, Lincoln. 2002. “Why Is Black–White Residential Segregation so Persistent?: Evidence
on Three Theories from Migration Data.” Social Science Research 31 (2): 197–229.

Quillian, Lincoln, and Mary E. Campbell. 2003. “Beyond Black and White: The Present and
Future of Multiracial Friendship Segregation.” American Sociological Review 68 (4):
540–66. https://doi.org/10.2307/1519738.

Quillian, Lincoln, and Rozlyn Redd. 2009. “The Friendship Networks of Multiracial
Adolescents.” Social Science Research 38 (2): 279–95.

Quiroz, Pamela. 2000. “A Comparison of the Organizational and Cultural Contexts of
Extracurricular Participation and Sponsorship in Two High Schools.” Educational
Studies 31 (3): 255–68.

Raudenbush, Stephen W., and Anthony S. Bryk. 2002. Hierarchical Linear Models:
Applications and Data Analysis Methods. Vol. 1. Sage.

126
Reed, Mark D., and Pamela Wilcox Rountree. 1997. “Peer Pressure and Adolescent Substance
Use.” Journal of Quantitative Criminology 13: 143–80.

Ryan, Allison M. 2000. “Peer Groups as a Context for the Socialization of Adolescents’
Motivation, Engagement, and Achievement in School.” Educational Psychologist 35 (2):
101–11.

Sanders, Jolene M. 2011. “Coming of Age: How Adolescent Boys Construct Masculinities via
Substance Use, Juvenile Delinquency, and Recreation.” Journal of Ethnicity in Substance
Abuse 10 (1): 48–70.

Schaefer, David R. 2012. “Homophily through Nonreciprocity: Results of an Experiment.”
Social Forces 90 (4): 1271–95.

Schaefer, David R., Sandra D. Simpkins, and Andrea Vest Ettekal. 2018. “Can Extracurricular
Activities Reduce Adolescent Race/Ethnic Friendship Segregation?” In Social Networks
and the Life Course, 315–39. Springer.

Schaefer, David R., Sandra D. Simpkins, Andrea E. Vest, and Chara D. Price. 2011. “The
Contribution of Extracurricular Activities to Adolescent Friendships: New Insights
through Social Network Analysis.” Developmental Psychology 47 (4): 1141–52.
https://doi.org/10.1037/a0024091.

Shih, Regina A., Layla Parast, Eric R. Pedersen, Wendy M. Troxel, Joan S. Tucker, Jeremy NV
Miles, Lisa Kraus, and Elizabeth J. D’Amico. 2017. “Individual, Peer, and Family Factor
Modification of Neighborhood-Level Effects on Adolescent Alcohol, Cigarette, e-
Cigarette, and Marijuana Use.” Drug and Alcohol Dependence 180: 76–85.

Shrum, Wesley, Neil H. Cheek, and Saundra MacD. Hunter. 1988. “Friendship in School:
Gender and Racial Homophily.” Sociology of Education 61 (4): 227–39.
https://doi.org/10.2307/2112441.

Smith, Jeffrey A., Miller McPherson, and Lynn Smith-Lovin. 2014. “Social Distance in the
United States: Sex, Race, Religion, Age, and Education Homophily among Confidants,
1985 to 2004.” American Sociological Review 79 (3): 432–56.

Smith, Sandra Susan. 2005. “‘Don’t Put My Name on It’: Social Capital Activation and Job-
Finding Assistance among the Black Urban Poor.” American Journal of Sociology 111
(1): 1–57.

Soteriades, Elpidoforos S., and Joseph R. DiFranza. 2003. “Parent’s Socioeconomic Status,
Adolescents’ Disposable Income, and Adolescents’ Smoking Status in Massachusetts.”
American Journal of Public Health 93 (7): 1155–60.

Stanton-Salazar, Ricardo D., and Sanford M. Dornbusch. 1995. “Social Capital and the
Reproduction of Inequality: Information Networks among Mexican-Origin High School
Students.” Sociology of Education 68 (2): 116–35.

127

Stearns, Elizabeth, Claudia Buchmann, and Kara Bonneau. 2009. “Interracial Friendships in the
Transition to College: Do Birds of a Feather Flock Together Once They Leave the Nest?”
Sociology of Education 82 (2): 173–95.

Stovel, Katherine, and Lynette Shaw. 2012. “Brokerage.” Annual Review of Sociology 38: 139–
58.

Tyson, Karolyn. 2011. Integration Interrupted: Tracking, Black Students, and Acting White after
Brown. Oxford University Press.

UCLA, Institute for Digital Research & Education, Statistical Consulting. 2020. “Multiple
Imputation in Stata.” Multiple Imputation in Stata. June 4, 2020.
https://stats.idre.ucla.edu/stata/seminars/mi_in_stata_pt1_new/.

Valente, Thomas W., and Kayo Fujimoto. 2010. “Bridging: Locating Critical Connectors in a
Network.” Social Networks 32 (3): 212–20. https://doi.org/10.1016/j.socnet.2010.03.003.

Valente, Thomas W., Kayo Fujimoto, Jennifer B. Unger, Daniel W. Soto, and Daniella Meeker.
2013. “Variations in Network Boundary and Type: A Study of Adolescent Peer
Influences.” Social Networks 35 (3): 309–16.

Valente, Thomas W., Peggy Gallaher, and Michele Mouttapa. 2004. “Using Social Networks to
Understand and Prevent Substance Use: A Transdisciplinary Perspective.” Substance Use
& Misuse 39 (10–12): 1685–1712.

Valente, Thomas W., Jennifer B. Unger, and C. Anderson Johnson. 2005. “Do Popular Students
Smoke? The Association between Popularity and Smoking among Middle School
Students.” Journal of Adolescent Health 37 (4): 323–29.

Vaquera, Elizabeth, and Grace Kao. 2008. “Do You like Me as Much as I like You? Friendship
Reciprocity and Its Effects on School Outcomes among Adolescents.” Social Science
Research 37 (1): 55–72.

Vasconcelos, Vítor V., Simon A. Levin, and Flávio L. Pinheiro. 2019. “Consensus and
Polarization in Competing Complex Contagion Processes.” Journal of the Royal Society
Interface 16 (155): 20190196.

Von Hippel, Paul T. 2009. “8. How to Impute Interactions, Squares, and Other Transformed
Variables.” Sociological Methodology 39 (1): 265–91.

Xiao, Zhixing, and Anne S. Tsui. 2007. “When Brokers May Not Work: The Cultural
Contingency of Social Capital in Chinese High-Tech Firms.” Administrative Science
Quarterly 52 (1): 1–31.

128
APPENDIX CHAPTER 2
Appendix 2 Table A. Missing/Imputed Values

Variables Missing (%) Variables Missing (%)
Student Level (n=64,412) School Level (N=110)
Grade 443 (.7%) Racial heterogeneity 0
Male 433 (.7%) Average SES 0
Race/Ethnicity 1352 (2.1%) School type
(Public vs Private)
0
# of hetero-friendships 0 Urbanicity 0
Total GPA 8336 (12.9%) Region 0
SES 12924 (20.1%) Network Density 0
Interaction term: sport*race 1352 (2.1%) Network Size 0
Interaction term: art*race 1352 (2.1%)
Interaction term:
academic*race
1352 (2.1%)
Participation in Sports 0
Participation in Arts 0
Participation in Academic 0

129
Appendix Figure 2.2. Percentage of Cross-Racial Friendships by Race/Ethnicity

130
Appendix 2 Table B. Multilevel Models Predicting EV Brokerage Scores from Key Predictors
Including Each of the Student-level Control Variables
Note: # p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

Model 13 Model 14 Model 15 Model 16
Participation in
Sports
-0.24 (0.18) 0.31 (0.18)# 0.36 (0.18)* 0.45 (0.18)*
Participation in
Arts
0.66 (0.20)*** 1.25 (0.20)*** 0.96 (0.20)*** 1.04 (0.20)***
Participation in
Academics
-0.65 (0.19)*** 0.005 (0.19) -0.05 (0.19) -0.17 (0.19)
Race (ref: White)
Black 1.57 (0.32)*** 1.17 (0.32)*** 1.12 (0.32)*** 1.12 (0.32)***
Hispanic 2.13 (0.30)*** 1.72 (0.30)*** 1.70 (0.30)*** 1.75 (0.30)***
Asian -0.31 (0.45) -0.68 (0.45) -0.63 (0.45) -0.61 (0.45)
Other 1.72 (0.57)** 1.21 (0.57)*** 1.22 (0.57)* 1.29 (0.57)*
Multiracial 2.36 (0.38)*** 2.11 (0.38)*** 2.00 (0.38)*** 2.08 (0.38)***
SES -0.19 (0.03)*** -0.15 (0.03)*** -0.14 (0.03)*** -0.15 (0.03)***
Friend degree 0.37 (0.02)*** - - -
Male - 1.08 (0.18)*** - -
Grade - - -0.42 (0.08)*** -
Total GPA - - - 0.18 (0.11)#
√

11
45.50 45.54 45.33 45.53
√𝜃
21.28 21.35 21.36 21.36
ρ
0.82 0.82 0.82 0.82
Intercept 44.22 (4.35)*** 46.71 (4.35)*** 47.00 (4.35)*** 46.43 (4.35)***
Number of
imputations
10 10 10 10
Sample size –
students
64,412 64,412 64,412 64,412
Sample size -
schools
110 110 110 110

131
APPENDIX CHAPTER 3
Appendix 3 Table A. Brokerage Measures Classified by Centrality vs. Bridging and by Local vs.
Global Network

Centrality Bridging
Local level Degree

Inverse Constraint

Global level Betweenness

VF-bridging
EV-brokerage

132
Appendix 3 Table B. Bivariate Analysis Table of Substance Use (Alcohol and Cigarette)
ANOVA Alcohol
Mean diff (df b, df w) F value p value
Race (2,1151) 27.94 <.0001***
Hispanic (Ref) Mean: .39
NH Asian -.23 <.0001***
Other +.23 .24
Parent’s Education (2,1161) .02 .98
Less than HS (Ref) Mean: .34
HS Graduated -.001 .99
College or Higher -.002 .99
GPA (3,1109) 6.12 .0004 ***
Mostly D’s and F’s (Ref) Mean: .52
Mostly C’s and D’s -.14 .99
Mostly B’s and C’s -.14 .92
Mostly A’s and B’s .25 .059
Cigarette
Mean diff (df b, df w) F value p value
Race (2, 1183) 4.26 .014*
Hispanic (Ref) Mean: .14
NH Asian -.067 .012*
Other +.022 .99
Parent’s Education (2,1193) 2.48 .084
Less than HS (Ref) Mean: .11
HS Graduated +.034 .49
College or Higher +.049 .11
GPA (3,1173)) 11.13 <.0001***
Mostly D’s and F’s (Ref) Mean: .18
Mostly C’s and D’s +.0077 .99
Mostly B’s and C’s -.012 .99
Mostly A’s and B’s -.12 .38
t-test Alcohol
Mean(s.e) s.d. t value p value
Gender
No .53 (.018) .50 -.85 .39
Yes .55 (.025) .50
Free Lunch
Eligible?
No .88 (.012) .32 -.51 .61
Yes .89 (.016) .31
Sport club?
No .41 (.018) .49 -3.32 .0009 ***
Yes .51 (.03) .50
Art club?
No .20 (.014) .40 1.54 .12
Yes .16 (.018) .37
Academic club?
No .32 (.017) .47 3.71 .0002 ***
Yes .22 (.021) .41
Brokerage
No 50.84 (.98) 27.20 .050 .96
Yes 50.75 (1.44) 28.64
Popular brokerage No
473.84
(15.43)
428.44 .24 .81

133
Yes
467.44
(22.87)
455.03
Indegree brokerage
No
193.21
(7.82)
217.17 1.00 .32
Yes
179.77
(10.89)
216.78
Outdegree brokerage
No
280.62
(10.54)
292.56 -.038 .70
Yes
287.67
(15.68)
312.08
Total degree
No 8.04 (.16) 4.56 -.04 .96
Yes 8.05 (.25) 4.92
Indegree
No 2.83 (.10) 2.85 1.51 .13
Yes 2.56 (.14) 2.81
Outdegree
No 5.21 (.13) 3.73 -1.16 .24
Yes 5.49 (.20) 4.07
Cigarette use:
sibling
No .14 (.013) .34 -4.13 <.0001***
Yes .23 (.022) .42
Alcohol use: sibling No .32 (.017) .46 -5.72 <.0001***
Yes .48 (.025) .50
Cigarette use: adults No .23 (.016) .42 -2.22 .026*
Yes .29 (.023) .46
Alcohol use: adults No .38 (.017) .48 -6.09 <.0001***
Yes .56 (.0089) .50
Cigarette
Mean(s.e) s.d. t value p value
Gender No .56 (.015) .50 3.60 .0003 ***
Yes .41 (.040) .49
Free Lunch
Eligible?
No .89 (.0096) .31 1.42 .15
Yes .85 (.028) .35
Sport club? No .44 (.015) .50 -.98 .33
Yes .48 (.04) .50
Art club? No .19 (.013) .39 1.73 .08
Yes .13 (.027) .34
Academic club? No .31 (.014) .46 4.18 <.0001***
Yes .15 (.028) .35
Brokerage No 50.59 (.85) 27.47 -1.30 .19
Yes 53.66 (2.23) 27.95
Popular brokerage No
472.35
(13.46)
13.46 -.41 .68
Yes
487.81
(36.65)
36.65
Indegree brokerage No
190.45
(6.75)
217.77 .38 .70
Yes
183.40
(16.07)
201.42

134
* p <.05, ** p <.01, *** p <.001
Outdegree brokerage No
281.89
(9.14)
294.94 -.87 .38
Yes
304.41
(27.16)
340.28
Total degree No 8.12 (.15) 4.69 .72 .47
Yes 7.83 (.37) 4.67
Indegree No 2.78 (.09) 2.86 .61 .54
Yes 2.63 (.21) 2.58
Outdegree No 5.34 (.12) 3.86 .42 .67
Yes 5.20 (.31) 3.90
Cigarette use:
sibling
No .14 (.011) .35 -7.23 <.0001***
Yes .37 (.039) .48
Alcohol use: sibling No .35 (.015) .48 -3.43 .0006 ***
Yes .5 (.042) .50
Cigarette use: adults No .23 (.013) .42 -4.96 <.0001***
Yes .41 (.040) .49
Alcohol use: adults No .43 (.016) .49 -2.10 .036*
Yes .52 (.041) .50
Correlation Alcohol (p-value)
% of Cigarette use: peers .016*
% of Alcohol use: peers -.03*
Cigarette (p-value)
% of Cigarette use: peers -.007**
% of Alcohol use: peers -.07

135
Appendix 3 Table C. Logistic Regression Models of Control Variables on Substance Use
* p <.05, ** p <.01, *** p <.001
Variables Alcohol Cigarette
Coeff (s.e.) t value (p-value) Coefficient (s.e.) t value (p-value)
Female .14 (1.4) 1.03 (.30) -.61 (.19) -3.13 (.002)**
Race/Ethnicity (ref:
Hispanic)

Asian -.92 (.20) -4.72 (<.001)*** -.48 (.27) -1.75 (.08)
Non-Hispanic/Non-
Asian
1.01 (.61) 1.66 (.097) .30 (.87) .35 (.73)
Parent’s edu
(ref: No HS)

High school
graduated
-.017 (.17) -.10 (.92) .17 (.23) .73 (.46)
Some college + -.06 (.16) -.36 (.72) .21 (.23) .94 (.35)
Free lunch eligible .04 (.22) .16 (.87) -.35 (.28) -1.24 (.22)
GPA -.20 (.09) -2.21 (.027)* -.39 (.12) -3.32 (.001)***
Sports participation .45 (.14) 3.29 (.001)*** .11 (.19) .59 (.55)
Arts participation -.20 (.18) -1.12 (.26) -.28 (.27) -1.04 (.30)
Academics
participation
-.23 (.16) -1.39 (.16) -55 (.26) -2.13 (.03)*
Parent alcohol use .56 (.14) 4.01 (<.001)*** .14 (.20) .69 (.49)
Sibling alcohol use .39 (.15) 2.56 (.011)* .02 (.23) .10 (.92)
Parent cigarette use .16 (.16) 1.01 (.31) .76 (.20) 3.70 (<.001)***
Sibling cigarette use .30 (.20) 1.55 (.12) 1.03 (.24) 4.23 (<.001)***
% of peers who use
alcohol
.37 (.27) 1.34 (.18) .20 (.37) .54 (.59)
% of peers who use
cigarettes
-.11 (.38) -.28 (.78) -1.18 (.58) -2.02 (.43)
Intercept -.70 (.38) -1.82 (.07) -.49 (.50) -1.00 (.32)

136
Appendix 3 Table D. Descriptive Statistics of 14 Participants Who Have Missingness in Network
Data, comparing 1,265 Respondents Who Have Network Information
* p <.05, ** p <.01, *** p <.001
Missing cases
(Mean & SD)
Non-missing cases
(Mean & SD)
t value
(p-value)
Alcohol use .42 (.51) .34 (.47) .56 (p=.57)
Cigarette use .083 (.29) .13 (.34) .49 (p=.63)
Gender .50 (.52) .53 (.50) .23 (p=.82)
Race/Ethnicity 1.33 (.65) 1.28 (.49) .40 (p=.69)
Parent’s education 2.00 (.88) 1.73 (.84) 1.17 (p=.24)
Free lunch eligibility* .69 (.48) .89 (.32) 2.21 (p<.05)*
GPA 3.08 (.86) 3.16 (.79) .36 (p=.72)
Sport participation .36 (.50) .44 (.50) .63 (p=.53)
Art participation .14 (.36) .18 (.38) .36 (p=.72)
Academic participation .29 (.47) .28 (.45) .0093 (p=.99)
Sibling’s alcohol use .36 (.50) .37 (.48) .067 (p=.95)
Sibling’s cigarette use .17 (.39) .17 (.37) .018 (p=.99)
Parent’s alcohol use .55 (.52) .44 (.50) .71 (p=.48)
Parent’s cigarette use .42 (.51) .25 (.43) 1.32 (p=.19)

137
Appendix 3 Table E. Missing/Imputed Values

Variables Missing (%) Variables Missing (%)
Student Level (n=1,265)
Alcohol use in the last 30
days
100 (7.82%)
Cigarette use in the last 30
days
68 (5.32%)
Sibling alcohol use 103 (8.14%) Total GPA 78 (6.17%)
Adult alcohol use 97 (7.67%) Race/Ethnicity 33 (2.61%)
Sibling cigarette use 73 (5.77%) Parent’s educational
attainment
10 (0.79%)
Adult cigarette use 67 (5.30%) Free lunch eligibility 8 (0.63%)
Peers who smoke cigarette 45 (3.56%) Female 5 (0.40%)
Peers who drink alcohol 45 (3.56%)

138
Appendix 3 Table F. Logistic Regression Results between Brokers with Multiple Ties and High-
central Individuals with Low Brokerage

* p <.05, ** p <.01, *** p <.001
Indegree*high-brokerage
OR 95% CI
Top 20 Percentile in Brokerage 2.38* 1.06-5.35
Indegree .99 .92-1.08
Broker*indegree .88 .73-1.06
Indegree*low-brokerage
OR 95% CI
Bottom 20 Percentile in rokerage .50* .29-.89
Indegree .94 .87-1.02
Non-broker*indegree 1.18 .96-1.44

139
APPENDIX CHAPTER 4
Appendix 4 Table A. Missing/Imputed Values (N = 61,608)
Variables Missing value (%)
% of Cross-gender Friendships 10,556 (17.13%)
Parent’s educational attainment 9,403 (15.26 %)
Total GPA 7,441 (12.08%)
Race/Ethnicity 4,210 (6.83%)
Alcohol use in the past 12 months 3,700 (6.01%)
Cigarette use in the past 12 months 3,551 (5.76%)
Grade 422 (.68%)
Gender 417 (.68%)

140
Appendix 4 Table B. Descriptive Sample Statistics Differing by Gender
Variable Female Male Total Number (F:M)
Dependent Variables Mean/% Mean/%
Alcohol use (binary) 52.13% 54.99% 57,583 (29,337:28,246)
Cigarette use (binary) 34.33% 35.62% 57,731 (29,406:28,325)
Independent and Mediating Variables
Broker 53.73 47.87 61,191 (30,652:30,539)
# of Cross-gender Friends 1.97 2.11 50,783 (26,914:23,869)
% of Cross-gender Friends 33.37% 36.64% 50,783 (26,914:23,869)
Control Variables – Student-level
Grade 9.52 9.51 60,889 (30,534:30,355)
Race/Ethnicity 57,101 (28,659:28,442)
White 54.59% 56.11%
Black 17.09% 14.38%
Hispanic 15.61% 16.71%
Asian/Pacific Islander 5.75% 6.06%
Other 6.96% 6.73%
Parent’s Education 51,943 (26,430:25,513)
No high school diploma 13.67% 10.77%
High school graduate 43.71% 43.57%
College or higher degree 42.62% 45.67%
Total GPA 2.88 2.73 53,887 (27,435:26,452)
Participation in Sports 50.02% 59.99% 61,191 (30,652:30,539)
Participation in Arts 34.58% 18.53% 61,191 (30,652:30,539)
Participation in Academics 38.70% 24.82% 61,191 (30,652:30,539)

141
Appendix 4 Table C. Unconditional and Basic Models for Alcohol and Cigarette Use and
Drunkenness by Gender
* p
<.05,
** p
<.01,
*** p
<.01
Alcohol Use
OR (95% CI) OR (95% CI)
Male Female Male Female
Broker - - 1.01 (.96-1.06) 1.02 (.97-1.07)
Intercept 1.01 (.90-1.14) .90 (.80-1.02) 1.01 (.90-1.14)
.89+ (.79-
1.01)
√

11
.57 .61 .57 .61
√𝜃
.18 1.74 .18 1.74
ρ .91 .11 .91 .11
Cigarette Use
OR (95% CI) OR (95% CI)

Male Female Male Female
Broker - -
.95+ (.90-
1.01)
1.02 (.97-1.08)
Intercept
.48*** (.44-
.53)
.46*** (.42-
.52)
.50*** (.45-
.54)
.46*** (.41-
.51)
√

11
.42 .54 .42 .54
√𝜃
1.79 1.77 1.79 1.81
ρ
.052 .085 .052 .082
Drunkenness

OR (95% CI) OR (95% CI)
Male Female Male Female
Broker - -
.90*** (.85-
.95)
.93** (.88-.98)
Intercept
.34*** (.29-
.39)
.27*** (.23-
.31)
.35*** (.30-
.41)
.28*** (.24-
.33)
√

11
.74 .77 .74 .77
√𝜃
1.70 1.76 1.70 1.76
ρ .16 .16 .16 .16

142
Appendix 4 Table D. Odds Ratios on the Likelihood of Alcohol Consumption by Gender
Excluding the Interaction Term
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

33
OR = 1.005601, 95% CI = 1.00457-1.006634 (Male); OR = 1.004674, 95% CI = 1.003559-1.005791 (Female)
Alcohol Use
Primary Predictors OR (95% CI) OR (95% CI)
Male Female
Broker (cluster connector) .98 (.93-1.04) .99 (.94-1.05)
% of Cross-gender friends
33
1.01*** (1.00-1.01) 1.00*** (1.00-1.01)
Control Variables
Grade 1.35*** (1.32-1.38) 1.35*** (1.32-1.38)
Race (ref: White)
Black .72*** (.65-.79) .71*** (.65-.77)
Hispanic 1.01 (.93-1.10) .86*** (.78-.94)
Asian .62*** (.54-.70) .53*** (.47-.60)
Other 1.16* (1.03-1.30) 1.06 (.96-1.18)
Parent’s edu (ref: No HS)
High school graduate .97 (.87-1.08) .93 (.85-1.03)
College or higher degree .95 (.86-1.05) .89** (.80-.98)
Participation in sports 1.21*** (1.14-1.28) 1.18*** (1.12-1.24)
Participation in arts .76*** (.72-.81) .76*** (.72-.80)
Participation in academics .97 (.91-1.04) 1.04 (.99-1.10)
Total GPA .67*** (.64-.69) .61*** (.59-.64)
Private School (vs. Public) 1.19 (.92-1.54) 1.06 (.81-1.37)
Urbanity (ref: Urban)
Suburban 1.11 (.95-1.31) 1.16+ (.98-1.36)
Rural 1.15 (.91-1.46) .93 (.73-1.18)
Gender heterogeneity .30+ (.080-1.14) .0065 (.000012-3.58)
Intercept .30** (.14-.63) 3.07 (.13-74.51)
√

11
.33 .33
√𝜃
1.84 1.84
ρ .031 .031

143
Appendix 4 Table E. Odds Ratios on the Likelihood of Cigarette Smoking by Gender Excluding
the Interaction Term
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

34
OR = 1.003159, 95% CI = 1.002106-1.004213 (Male); OR = 1.003165, 95% CI = 1.001953-1.004379 (Female)
Cigarette Use
Primary Predictors OR (95% CI) OR (95% CI)
Male Female
Broker (cluster connector) .95+ (.89-1.01) .99 (.94-1.05)
% of Cross-gender friends
34
1.00*** (1.00-1.00) 1.00*** (1.00-1.00)
Control Variables
Grade 1.12*** (1.10-1.15) 1.14*** (1.12-1.17)
Race (ref: White)
Black .55*** (.50-.61) .39*** (.35-.42)
Hispanic .86** (.79-.95) .68*** (.62-.75)
Asian .72*** (.62-.84) .53*** (.46-.61)
Other 1.17** (1.05-1.29) .96 (.87-1.07)
Parent’s edu (ref: No HS)
High school graduate .91* (.82-1.00) .96 (.88-1.04)
College or higher degree .94 (.85-1.03) .90* (.83-.98)
Participation in sports .85*** (.80-.90) 1.00 (.95-1.06)
Participation in arts .91** (.85-.98) .76*** (.72-.81)
Participation in academics .96 (.90-1.03) .96 (.91-1.02)
Total GPA .60*** (.57-.63) .52*** (.50-.54)
Private School (vs. Public) 1.01 (.79-1.31) 1.01 (.75-1.35)
Urbanity (ref: Urban)
Suburban 1.13 (.96-1.32) 1.17+ (.98-1.41)
Rural 1.31* (1.04-1.65) 1.37* (1.05-1.80)
Gender heterogeneity .18** (.051-.63) .16 (.0025-10.02)
Intercept 1.82+ (.89-3.71) 2.78 (.33-23.12)
√

11
.32 .37
√𝜃
1.82 1.81
ρ .030 .040

144
Appendix 4 Table F. Odds Ratios on the Likelihood of Drunkenness by Gender Excluding the
Interaction Term
Note: + p <. 0.1, * p<.05 ** p<.01 *** p<.001 (two-tailed tests)

35
OR = 1.004308, 95% CI = 1.003197-1.00542 (Male); OR = 1.004381, 95% CI = 1.003212-1.005552 (Female)
Drunkenness
Primary Predictors OR (95% CI) OR (95% CI)
Male Female
Broker (cluster connector) .88*** (.83-.93) .91** (.86-.97)
% of Cross-gender friends
35
1.00*** (1.00-1.01) 1.00*** (1.00-1.00)
Control Variables
Grade 1.46*** (1.43-1.50) 1.40*** (1.36-1.43)
Race (ref: White)
Black .61*** (.54-.68) .53*** (.47-.59)
Hispanic .93 (.84-1.03) .78*** (.69-.88)
Asian .48*** (.41-.55) .46*** (.39-.54)
Other 1.20** (1.06-1.37) .97 (.87-1.09)
Parent’s edu (ref: No HS)
High school graduate .87** (.78-.97) .90* (.82-.99)
College or higher degree .84*** (.76-.93) .88* (.80-.97)
Participation in sports 1.15*** (1.08-1.22) 1.16*** (1.10-1.24)
Participation in arts .71*** (.66-.77) .65*** (.61-.69)
Participation in academics .97 (.91-1.04) 1.02 (.95-1.08)
Total GPA .64*** (.61-.66) .57*** (.55-.60)
Private School (vs. Public) 1.16 (.86-1.55) 1.06 (.77-1.47)
Urbanity (ref: Urban)
Suburban 1.25* (1.04-1.50) 1.32** (1.08-1.60)
Rural 1.41* (1.07-1.84) 1.20 (.90-1.60)
Gender heterogeneity .42 (.096-1.88) .0044+ (.000018-1.13)
Intercept .052*** (.023-.12) 1.02 (.062-16.96)
√

11
.37 .38
√𝜃
1.84 1.89
ρ .039 .039

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

Social self-control and adolescent substance use

PDF

Adolescent social networks, smoking, and loneliness

PDF

Social network influences on depressive symptoms among Chinese adolescents

PDF

Contextualizing social network influences on substance use among high risk adolescents

PDF

Exploring the role of peer influence, linguistic acculturation, and social networks in substance use

PDF

The egocentric social networks of Black adolescents and sources of racial socialization messages

PDF

Sociocultural stress, coping and substance use among Hispanic/Latino adolescents

PDF

Substance use disparities in adolescents of lower socioeconomic status: emerging trends in electronic cigarettes, alternative tobacco products, marijuana, and prescription drug use

PDF

The link between maternal depression and adolescent daughters' risk behavior: the mediating and moderating role of family

PDF

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

PDF

Examining the built and online environments to understand cannabis and e-cigarette availability, marketing, and product use in Southern California

PDF

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

PDF

Energy drink consumption, substance use and attention-deficit/hyperactivity disorder among adolescents

PDF

Affective decision-making predictive of Chinese adolescents drinking behaviors: a longitudinal study

PDF

Collaborative social networks in student affairs: an exploration of the outcomes and strategies associated with cross‐institutional collaboration

PDF

The interplay between social connection and substance use

PDF

Adolescent conduct problems and substance use: an examination of the risk pathway across the transition to high school

PDF

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

PDF

The Internet activities, gratifications, and health consequences related to compulsive Internet use

Asset Metadata

Creator Lee, Jihye Yoo (author)

Core Title Friendship network position on adolescent behaviors: an examination of a broker position and the likelihood of alcohol and cigarette use

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Sociology

Degree Conferral Date 2023-08

Publication Date 07/05/2023

Defense Date 06/05/2023

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

Tag add health,adolescent substance use,broker position,OAI-PMH Harvest,social network,social networking study data

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Owens, Ann (committee chair), Biblarz, Tim (committee member), Valente, Thomas (committee member)

Creator Email jihye.u@gmail.com,yoojihye@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113260746

Unique identifier UC113260746

Identifier etd-LeeJihyeYo-12013.pdf (filename)

Legacy Identifier etd-LeeJihyeYo-12013

Document Type Dissertation

Format theses (aat)

Rights Lee, Jihye Yoo

Internet Media Type application/pdf

Type texts

Source 20230705-usctheses-batch-1061 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

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

Repository Name University of Southern California Digital Library

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

Repository Email cisadmin@lib.usc.edu