Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Adolescent social networks, smoking, and loneliness
(USC Thesis Other)
Adolescent social networks, smoking, and loneliness
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
i
ADOLESCENT SOCIAL NETWORKS, SMOKING, AND LONELINESS
By
Stephanie Raye Dyal
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PREVENTIVE MEDICINE: HEALTH BEHAVIOR)
August 2016
ii
Dedication
This dissertation is dedicated to Gram, who devoted her life to raising her children, her
grandchildren, and her great-grandchildren. She always showed genuine interest in all our
endeavors and satisfied our curiosity with an infinite number of tales.
iii
Acknowledgements
Thank you to my dissertation committee, Drs. Valente, Bluthenthal, Rohrbach and Unger,
for supporting me throughout the dissertation process, qualifying exams, and graduate school as
a whole. I have learned from all of you and appreciate all the advice given as I proposed my
studies, opportunities to work with each of you individually on projects, and career advice and
encouragement. I especially thank Dr. Valente for providing much feedback on the studies
presented here, providing me many research opportunities, and continually offering support as
I’ve completed my graduate studies. You’ve been a wonderful advisor!
I would also like to thank all of my fellow students. I feel so lucky to have found myself
in such a welcoming work environment. It is only with all of your support, advice, and
encouragement that I have made it to here. I have greatly benefitted from your stories of your
own successes and struggles, your companionship, and even just the distraction that you all
provide to keep me going on tasks that I’m struggling to get done! You all have also been
wonderful people to collaborate with on project and to study with during classes. I hope to keep
in touch with all of you all as we move ahead in our careers and find ourselves on different paths.
I’m excited to see where we all end up!
Thank you to Marny Barovich for always looking out for me and the rest of the students,
whether it be giving class or career advice or making sure we were aware there was free food
somewhere in the building. Thank you to Andrew Zaw and the rest of the IT team for solving my
many computer problems. Thank you to Joyce and Consuelo for answering my many questions
about accounting. Thank you to Abigail for creating the budget for my grant. Thank you to all
the rest of the staff and faculty that I have missed naming here.
Thank you to Daniel Soto and Monica Pattarroyo for doing so much work to get the
Social Network Study data collected and entered! I also thank the rest of the research assistants
on the study, the schools and teachers for allowing us to take classroom time to survey the
students, and the students for participating in this research. Without willing research participants,
none of this would have been completed!
My dissertation work was supported by a two-year dissertation award from the Tobacco-
Related Disease Research Program (grant #23DT-0112). Thank you TRDRP for supporting my
dissertation research and for supporting my career development. Data for this dissertation was
collected as part of the Social Networks Study (PI: Valente). The Social Networks Study was
supported by National Institutes of Health grants RC1AA019239-01 and 03R01CA157577-02S1.
An earlier version of Chapter 2 was presented as a poster at the “Tobacco Control,
Research, and Education: Joining Forces to Address New Challenges” conference in
Sacramento, California in October 2015. An early version of Chapter 3 was presented as an oral
presentation at the Sunbelt conference in Brighton, England in June 2015. An early version of
Chapter 4 was presented as an oral presentation at the Sunbelt conference in Newport Beach,
California in April 2016. Thank you to all who provided feedback during these presentations and
to those who attended my presentations and showed interest in my work.
iv
TABLE OF CONTENTS
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Figures ................................................................................................................................ vi
List of Tables ................................................................................................................................ vii
Abstract ........................................................................................................................................ viii
CHAPTER 1: INTRODUCTION ................................................................................................... 1
Specific Aims .............................................................................................................................. 1
Study 1 ..................................................................................................................................... 2
Study 2 ..................................................................................................................................... 4
Study 3 ..................................................................................................................................... 5
Background and Significance...................................................................................................... 6
Scope of the Issue: Cigarette Use ............................................................................................ 6
Loneliness ................................................................................................................................ 7
Social Networks ..................................................................................................................... 11
Overview of Studies and Theoretical Background ................................................................ 17
CHAPTER 2: ASSOCIATIONS OF SOCIAL NETWORKS AND AFFECT WITH TOBACCO
AND OTHER SUBSTANCE USE IN ADOLESCENTS ............................................................ 21
Abstract ..................................................................................................................................... 21
Introduction ............................................................................................................................... 22
Methods ..................................................................................................................................... 23
Results ....................................................................................................................................... 29
Discussion ................................................................................................................................. 33
CHAPTER 3: ACCURACY AND INFLUENCE OF PERCEPTION OF PEER CIGARETTE
USE: MODERATION BY LONELINESS .................................................................................. 45
Abstract ..................................................................................................................................... 45
Introduction ............................................................................................................................... 46
Methods ..................................................................................................................................... 50
Results ....................................................................................................................................... 58
Discussion ................................................................................................................................. 60
CHAPTER 4: THE COEVOLUTION OF ADOLESCENT FRIENDSHIP NETWORKS,
LONELINESS, AND CIGARETTE SMOKING ......................................................................... 74
Abstract ..................................................................................................................................... 74
Introduction ............................................................................................................................... 74
v
Methods ..................................................................................................................................... 79
Results ....................................................................................................................................... 85
Discussion ................................................................................................................................. 87
CHAPTER 5: CONCLUSION ................................................................................................... 100
REFERENCES ........................................................................................................................... 106
vi
List of Figures
Figure 1: Conceptual model outlining associations among constructs included in the dissertation
studies 18
Figure 2: Confirmatory Factor Analysis Model 43
Figure 3: Main Structural Equation Model 44
vii
List of Tables
Table 1: Summary of article findings on the association between social networks and loneliness
15
Table 2: Demographic Characteristics of sample, using complete data by variable and the
imputed data set for analysis 36
Table 3: Means, standard deviation, and correlation matrix for variables included in SEM 37
Table 4: Results from structural equation model for entire sample 38
Table 5: Results from multigroup model assessing differences by ethnicity 39
Table 6: Results from multigroup model assessing differences by gender 41
Table 7: Demographics for imputed and original sample 67
Table 8: Regression results examining accuracy of participant report of friend's behavior 69
Table 9: Regressions for proportion of smokers/drinkers in network and participant’s
drinking/smoking behavior 70
Table 10: Interaction results for loneliness and depression predicting smoking and alcohol use 71
Table 11: Regression results for dyadic analyses 72
Table 12: Descriptive Statistics by School 92
Table 13: Effects included in Coevolution models 93
Table 14: Model Fit and Descriptives 95
Table 15: Results from Smoking, Loneliness, and Friendship Network Coevolution model by
school 96
Table 16: Meta-analysis results for SAOM of loneliness, smoking, and the friendship network 98
viii
Abstract
Past research has found that loneliness and social networks are both associated with
cigarette smoking. However, social factors and loneliness have not been studied concurrently as
correlates of smoking behavior, despite their potential association with each other. Furthermore,
past studies have often conflated loneliness and depression or did not include depression as a
covariate in analyses, although research suggests that loneliness and depression are correlated
and are associated with smoking. Examining social networks, loneliness, and depression together
as correlates of smoking may provide clarity into how they contribute to smoking behavior
individually and through interaction. This dissertation examines the associations among
loneliness, social networks, depression, and smoking in a sample of predominantly
Hispanic/Latino adolescents enrolled in five schools located in one school district in Southern
California, USA.
Study one is a structural equation model of social networks, parental communication,
loneliness, and depression as predictors of use of cigarettes, e-cigarettes, hookah, marijuana and
alcohol. Study two uses logistic and linear regression to assess the associations of social
networks, loneliness, and depression with accuracy in assessment of peer behavior, exposure to
peer smoking and alcohol use, participant’s own use of cigarettes and alcohol, and participant’s
use of cigarettes and alcohol with friends. Study three uses a stochastic actor-oriented model to
analyze the dynamics among social networks, loneliness, and smoking. Studies two and three
additionally consider potential interaction terms between peer influence and loneliness in the
prediction of smoking.
Contrary to past studies, we did not find that loneliness was associated with smoking and
other substance use. Exposure from peers to cigarette and alcohol use was associated with
ix
adolescents’ own cigarette and alcohol use, but loneliness was not a moderator of this
association. Furthermore, selection and influence processes were not found for cigarette use,
despite findings that having a friend who smokes is associated with an adolescent’s smoking
behavior. We found that centrality, parental communication, depression, and self-identifying as a
stoner/druggie were associated with smoking and other substance use. Future research directions
and implications of the findings are discussed.
1
CHAPTER 1: INTRODUCTION
Specific Aims
While smoking rates have dropped substantially over the past couple of decades,
adolescents continue to begin smoking at high rates despite notable health risks (National Center
for Chronic Disease Prevention and Health Promotion (US) Office on Smoking and Health,
2012). Understanding the risk factors for smoking and developing preventive interventions to
address these risk factors may result in a decrease in the uptake of cigarette smoking by
adolescents. Loneliness and social network characteristics are both important predictors of
smoking during adolescence (Christopherson & Conner, 2012; Seo & Huang, 2012). However,
social networks and loneliness have not been studied concurrently as predictors of cigarette
smoking, despite their potential association with one another and with smoking conjointly. This
study examines social network characteristics and loneliness as predictors of smoking within a
sample of predominately Hispanic/Latino adolescents attending five California high schools.
Findings from this study may enhance understanding of the etiology of cigarette use by
Hispanic/Latino adolescents and the larger adolescent population, contributing to youth tobacco
use prevention efforts.
Roughly 10% of Hispanic adolescents experience chronic high levels of loneliness (Benner,
2011), however, little research on loneliness has been conducted in the Hispanic adolescent
population. Much research concerning loneliness and health behaviors has either been conducted
in adults or has conflated loneliness with other negative affective states, leaving a gap in the
literature this study intends to address (VanderWeele, Hawkley, Thisted, & Cacioppo, 2011).
Additionally, loneliness and objective sociometric measures of social network quality and
quantity are not synonymous, yet conceptually associated, and have been found to be associated
2
in the few studies that have examined their association (Cacioppo, Fowler, & Christakis, 2009).
This dissertation will focus on studying social networks concurrently with perceived loneliness
in models predicting smoking which will assess how social networks and loneliness are
associated with one another and with smoking behavior, in a population that is understudied.
Parental communication, linguistic acculturation, depression, and personal identity may
all affect the level of loneliness experienced by adolescents, and the former three variables have
been found to be associated with loneliness in the literature (Benner, 2011; Matthews et al.,
2016; Segrin, Nevarez, Arroyo, & Harwood, 2012). All four of these variables have been shown
to be associated with smoking in past research (Chaiton, Cohen, O’Loughlin, & Rehm, 2009;
Cohen, Richardson, & LaBree, 1994; Epstein, Botvin, & Diaz, 1998; Sussman et al., 1990).
Studying these communication, identity, and affect constructs as predictors of smoking along
with social networks and loneliness, and exploring potential moderation effects within the
models will further address how social networks and loneliness may contribute to smoking
behavior.
I conducted three studies which explore the associations among social networks,
loneliness, and smoking in a sample of primarily Hispanic/Latino adolescents. Study aims and
hypotheses are detailed below. Hypotheses are in italics.
Study 1
Loneliness, social network indicators, depression, and parental communication will be modeled
as concurrent predictors of use of cigarettes, e-cigarettes, hookah, alcohol, and marijuana using
structural equation modeling (SEM).
3
1.1 To assess the associations between loneliness and social network indicators. We
hypothesize that loneliness and social network indicators will be negatively associated
with one another.
1.2 To assess the associations of loneliness, social connection, depression, and parental
communication with use of cigarettes, e-cigarettes, hookah, alcohol, and marijuana.
Loneliness, depression, and poor parental communication will be positively associated
with the substance use outcomes. Social connection will be negatively associated with the
substance use outcomes.
1.3 To assess if gender differences arise in the associations of loneliness, social connection,
depression, and parental communication with use of cigarettes, e-cigarettes, hookah,
alcohol, and marijuana. Gender differences may occur for the associations of social
network indices and loneliness with smoking and other substance use outcomes. Social
network pressures such as popularity and social isolation may affect substance use
behavior differently for males and females. Loneliness may also affect substance use
behavior differently for males and females, although this would be more likely with a
loneliness measure which includes the word lonely in it.
1.4 To assess if ethnic differences arise in the associations of loneliness, social network
indicators, depression, and parental communication with use of cigarettes, e-cigarettes,
hookah, alcohol, and marijuana. We do not have a hypothesis to suggest ethnic
differences; however we would like to emphasize findings relevant to the Hispanic/Latino
adolescent population and address any potential differences in findings by ethnicity.
4
Study 2
Loneliness, depression, identity, and social network indicators will be assessed as predictors of
accuracy between perceptions of peer smoking/drinking and actual peer smoking/drinking.
Loneliness will be assessed as a moderator for the associations of peer substance use, identity,
and social network indicators with participants’ smoking and drinking behaviors.
2.1 To assess if loneliness, out-degree, substance user identity, or depression are predictors of
accuracy in reporting of peer’s smoking and drinking behaviors. High loneliness will be
predictive of accurate perception of peer’s smoking and alcohol use because loneliness
increases attention/memory for social cues. High out-degree will be predictive of
accuracy in peer smoking and alcohol use perception.
2.2 To assess if loneliness, depression, out-degree, or identity is associated with the number
of friends nominated who smoke. Adolescents who identify with deviant identities will
have more peers who smoke. No other hypotheses are suggested.
2.3 To assess if peer smoking perceptions, loneliness, depression, or identity are associated
with smoking behavior. Higher perception of peer smoking will be associated with higher
odds of smoking. Self-identification with a deviant social identity will be associated with
higher odds of smoking.
2.4 To assess if loneliness or depression moderate the associations between peer smoking,
out-degree, and identity with smoking behavior and susceptibility. The association
between peer smoking and participant smoking will be stronger for those participants
who score higher on the loneliness scale. The association between the deviant identity
indicators and smoking will be stronger for those participants who score higher on the
loneliness scale.
5
Study 3
Loneliness, social network structure, parental communication, and linguistic acculturation will
be modeled as predictors of smoking using a stochastic actor-oriented model (SAOM).
3.1 To assess the stability of loneliness and how it is associated with network structure and
psychosocial variables in a co-evolution model of friendship network, loneliness and
smoking. We expect roughly 10% of the sample to exhibit stable, trait-like loneliness and
an additional 10% to exhibit loneliness increasing over time, as seen in recent studies
(Benner, 2011). Acculturation to the US culture and parental communication will be
negatively associated with loneliness. We have no a priori hypotheses concerning how
network structural characteristics will be associated with loneliness.
3.2 To assess the associations of loneliness, social networks, parental communication, and
linguistic acculturation as concurrent predictors of smoking in a co-evolution model of
network, loneliness, and smoking behavior. Loneliness and acculturation to the US
culture will be positively associated with smoking. Parental communication will be
protective against smoking. We expect both influence and selection to exist in a bi-
directional association between smoking and the network.
3.3 To assess if loneliness may serve as a potential moderator of network measures in the
prediction of smoking behavior. A significant interaction term may be found between a
peer influence indicator and loneliness for smoking behavior. We have no other a priori
hypotheses concerning interaction terms.
6
Background and Significance
Scope of the Issue: Cigarette Use
Cigarette smoking has been identified as a leading modifiable risk factor for cancer,
cardiovascular and lung disease, and reproductive problems in the United States and worldwide
(Office of the Surgeon General (US) & Office on Smoking and Health (US), 2004; Samet, 2013).
Almost one fifth, or 19.5% of US youth in grades 9-12 have smoked a cigarette in the past 30
days (National Center for Chronic Disease Prevention and Health Promotion (US) Office on
Smoking and Health, 2012). Smoking during adolescence is linked to an increased rate of
respiratory illness, decreased lung growth, and decreased physical fitness (National Center for
Chronic Disease Prevention and Health Promotion (US) Office on Smoking and Health, 2012).
Furthermore, adolescence is a key developmental phase for establishing smoking habits; the
majority of adult smokers (88%) began smoking prior to the age of 18 (National Center for
Chronic Disease Prevention and Health Promotion (US) Office on Smoking and Health, 2012).
Half of adult smokers die prematurely; cigarette smoking is attributed as the cause of 440,100
deaths annually in the United States (Office of the Surgeon General (US) & Office on Smoking
and Health (US), 2004). California has been a leading state in the reduction of smoking rates, yet
still 38,050 deaths in California from 1995-1999 were attributed to cigarette smoking (Office of
the Surgeon General (US) & Office on Smoking and Health (US), 2004; Pierce, Messer, White,
Cowling, & Thomas, 2011). The reduction of cigarette use would lead to reduced loss of life,
spending on healthcare, and morbidity.
Within California, Hispanic adolescents exhibit significantly higher prevalence of
lifetime cigarette use than white adolescents, however, smoking rates among Hispanics
nationwide are generally lower than non-Hispanic whites (Bethel & Schenker, 2005; Unger et
7
al., 2000). Rates of heavy smoking (defined as smoking 11+ cigarettes a day) have increased for
Hispanic adolescents in the US who smoke from 3.1% to 6.4% from 1991 to 2009 (Jones, Kann,
& Pechacek, 2011), a time period when population rates of heavy smoking have decreased
(Pierce et al., 2011). In 2012, persons identifying as Hispanic/Latino accounted for 38.2% of the
population of California, a 5.8% increase from 2000 (US Census Bureau, n.d.). Furthermore, the
Hispanic/Latino population is younger in comparison to the non-Hispanic White population
(Ramirez & de la Cruz, 2002), so Hispanics likely represent an even larger percentage of the
Californian adolescent population. Hispanic adolescents comprise a large, growing percentage of
the California population, have a disproportionately high prevalence of smoking, and a
disproportionate rise in heavy smoking rates. Research is needed to identify correlates of
smoking in Hispanic/Latino adolescents in order to develop appropriate prevention and cessation
programs to reduce smoking rates and disproportionate burden for this population.
Loneliness
Loneliness is defined by cognitive theorists as a negative affective state resulting from
the perception of oneself as socially isolated (Laursen & Hartl, 2013; Peplau & Perlman, 1982).
It has been differentiated from social isolation as an undesired lack of relationships characterized
by mutual understanding of life experiences by both parties (van Staden & Coetzee, 2010). Like
other negative affective states, loneliness by definition is an emotionally painful and undesired
experience (Laursen & Hartl, 2013). The majority of loneliness research has been guided by
cognitive theories, such as self-discrepancy theory, and psychodynamic theories, such as social
needs perspective (Marangoni & Ickes, 1989; Peplau & Perlman, 1982; Sønderby & Wagoner,
2013). Self-discrepancy theory suggests that loneliness results when a person’s social network is
not congruent with their ideal social network, and social needs perspective suggests that
8
loneliness results when a person perceives their unfulfilled needs to be related to social isolation
(Heinrich & Gullone, 2006; Laursen & Hartl, 2013; Peplau & Perlman, 1982; Thurston &
Kubzansky, 2009).
People’s perceptions of what an adequate social network is may differ due to individual
differences in quantity and quality of social stimuli considered to be ideal/ fulfilling. Therefore,
variance in loneliness is not fully explained by objective measures of social network quality or
network structure (Cacioppo et al., 2009; Laursen & Hartl, 2013). Loneliness may be
experienced as either a transient state induced by environmental or cognitive changes or as a
more permanent trait (Peplau & Perlman, 1982). Little research has focused on exploring
differences between trait and state loneliness, however longitudinal studies suggest loneliness
rates are stable for the majority of people and may fluctuate more in males (Iecovich, Jacobs, &
Stessman, 2011; Ladd & Ettekal, 2013; Schinka, van Dulmen, Mata, Bossarte, & Swahn, 2013).
Loneliness may be a multidimensional construct with separate dimensions for lack of connection
to a larger social network, lack of intimate relationships, lack of romantic relationships, or other
relationship specific experiences of loneliness (Peplau & Perlman, 1982). The experience of
loneliness and perceived social deficits that lead to loneliness may differ greatly from person to
person.
Loneliness is consistently associated with poor health outcomes in studies of health
behaviors, physiological indicators of health, and mortality (Cacioppo et al., 2002; Cacioppo &
Hawkley, 2003; Iecovich et al., 2011; Lauder, Mummery, Jones, & Caperchione, 2006; Patterson
& Veenstra, 2010). Adolescents and adults who experience high loneliness are more likely to
engage in risky behaviors such as cigarette use (Allen, Page, Moore, & Hewitt, 1994;
Christopherson & Conner, 2012; DeWall & Pond, 2011; Lauder et al., 2006; Leung, de Jong
9
Gierveld, & Lam, 2008; Page, Dennis, Lindsay, & Merrill, 2010; Peltzer, 2009; Shankar,
McMunn, Banks, & Steptoe, 2011). In a study of male medical college students in China, 33.7%
of current smokers cited loneliness as a reason for smoking their first cigarette (Xiang et al.,
1999). However, some studies failed to find a statistically significant association between
smoking and loneliness (Grunbaum, Tortolero, Weller, & Gingiss, 2000; Hays & DiMatteo,
1987; Qualter et al., 2013; Steptoe, Owen, Kunz-Ebrecht, & Brydon, 2004). Some researchers
have studied the association between loneliness and smoking in multiple nations and gender
subgroups within nations and have reported variability in findings (Page et al., 2008, 2010;
Stickley et al., 2013; Stickley, Koyanagi, Koposov, Schwab-Stone, & Ruchkin, 2014). Cigarettes
can serve as cultural and identity indicators (Pampel, 2006) and their social use and meaning
within cultures, demographic groups, and communities may modify the association between
loneliness and smoking. Furthermore, population smoking prevalence may moderate the
association between smoking and loneliness (DeWall & Pond, 2011). Additional research to
understand variability in research findings concerning smoking and loneliness is warranted.
Adolescents report feelings of loneliness at a higher rate than other age groups (Perlman
& Peplau, 1981; Qualter et al., 2013). Some 79% of persons under age 18 report experiencing
loneliness sometimes or more, in comparison to 37% of persons aged 55 and older (Heinrich &
Gullone, 2006; Perlman & Peplau, 1981). Research in adolescent loneliness trajectories suggests
that the majority of adolescents experience stable low rates of loneliness (Ladd & Ettekal, 2013;
Qualter et al., 2013; Schinka et al., 2013). However, some adolescents experience chronic high
loneliness while others experience rising or decreasing levels of loneliness throughout
adolescence (Ladd & Ettekal, 2013; Qualter et al., 2013; Schinka et al., 2013). Scholars have
suggested that loneliness is not an abnormal, pathological experience and may be experienced as
10
a normal response to changing environments, relationships, and cognitions during adolescence;
only persistent feelings of loneliness are problematic and associated with negative health
outcomes (Heinrich & Gullone, 2006). Adolescence is an important developmental period for the
experience of loneliness as well as the prevention of stable, chronic loneliness (Heinrich &
Gullone, 2006).
Roughly 10% of Hispanic adolescents experience chronic high levels of loneliness
(Asher & Paquette, 2003; Benner, 2011). A study in Los Angeles, CA of Hispanic 9
th
and 10
th
graders found that 11% of the sample experienced high chronic loneliness and an additional 11%
experienced increasing feelings of loneliness from 9
th
to 10
th
grade (Benner, 2011). Past research
in adults has found Hispanics to experience more loneliness than other ethnic groups; although
this association is explained by education status and income (Hawkley et al., 2008). However,
this does not negate that Hispanics, particularly those with lower SES, may experience higher
loneliness than other demographics. Immigration itself may not cause higher rates of loneliness,
as evidenced by research in Switzerland suggesting that adolescents who have migrated from
their home country are no different in loneliness rates in comparison to adolescents living in their
home country (Neto & Barros, 2000). Yet, immigrant Mexican American youth report higher
loneliness than US--born Mexican American youth (Polo & López, 2009). More research is
needed in Hispanic/Latino adolescents to assess if they are at higher risk for loneliness in
comparison to other demographic groups, if loneliness is associated with any specific health
outcomes for this demographic group, and what risk and/or protective factors they may have for
loneliness.
The measurement of loneliness may affect the association of loneliness with smoking.
Females report higher levels of loneliness when loneliness is measured using surveys with the
11
word “lonely” in the questions (Hawkley et al., 2008; Heinrich & Gullone, 2006; Peplau &
Perlman, 1982). In many studies, no gender differences in loneliness are noted when the word
lonely is not used in questionnaires, however there continues to be controversy regarding this
(Hawkley et al., 2008; Peplau & Perlman, 1982; Schinka et al., 2013). A review of gender
differences in loneliness in adolescents by Koenig and Abrams summarized by Heinreich &
Gullone found that males consistently report higher loneliness in comparison to females
(Heinrich & Gullone, 2006). This is supported by a meta-analysis of predictors of loneliness in
adolescence in which gender was found to have a large effect on loneliness with the majority of
studies reporting that males are higher in loneliness (Mahon, Yarcheski, Yarcheski, Cannella, &
Hanks, 2006). However, only two-thirds of the studies assessed found gender to be a significant
predictor leading the researchers to suggest the potential of methodological issues concerning the
association of loneliness and gender (Mahon et al., 2006). Thus, the association between gender
and loneliness is unclear. Improved understanding of the measurement of loneliness may elicit a
clearer understanding of potential gender differences in loneliness and potential gender
differences in the association between loneliness and smoking.
Social Networks
Exposure to and affiliation with other people in a community influences a person’s
expectations, behavior, and emotions. Social networks are the connections among a group of
bounded or unbounded people which can be used to model social processes and analyze the
effects of social processes on an individual. Network structural characteristics, position within a
network, and attributes of one’s alters all may have effects on an individual (Valente, 2010).
Research suggests a person’s own smoking behavior is associated with the smoking
behaviors of their friends through both peer influence to smoke and selection of similar peers
12
(Alexander, Piazza, Mekos, & Valente, 2001; Ennett et al., 2008; Ennett & Bauman, 1994;
Hoffman, Sussman, Unger, & Valente, 2006; Lakon, Hipp, & Timberlake, 2010; Urberg, Luo,
Pilgrim, & Degirmencioglu, 2003; Valente, Gallaher, & Mouttapa, 2004). Adolescents are more
likely to be exposed to smoking and influenced to smoke through their overall friendship groups
in comparison to their closest friends (Urberg, Değirmencioğlu, & Pilgrim, 1997). In a study of
peer influence measures, perceived friend smoking status was found to be a better predictor of
future smoking initiation in comparison to friend--reported smoking status (Valente, Fujimoto,
Soto, Ritt-Olson, & Unger, 2013). Perceiving that one’s friends smoke is a predictor of both
smoking initiation and transition to daily smoking across ethnic groups (Kandel, Kiros,
Schaffran, & Hu, 2004). Peer behavior, and particularly one’s perception of their peers’ behavior
is an important predictor of adolescent smoking.
Mutuality, or reciprocity in friendship is associated with smoking behavior--- reciprocity
has been found to be protective against smoking (Ennett et al., 2006), however, peer influence to
smoke is stronger in reciprocated friendships (Fujimoto & Valente, 2012). Direct reciprocity
occurs when two people nominate each other within a network (Valente, 2010). Reciprocal
friendships have been indicated to be more influential in smoking behavior in comparison to
friendships considered to be best friendships by an adolescent (Fujimoto & Valente, 2012).
However, having reciprocated best friendships is protective against smoking (Ennett et al.,
2006). Reciprocity has been found to interact with perceived peer smoking behavior in the
prediction of smoking behavior; reciprocity is associated with greater smoking when one
perceives that many of their peers smokes (Lakon & Valente, 2012). While most studies have
found reciprocity to be associated with smoking, in one study an aggregate measure of
reciprocity and tie strength was not found to be associated with smoking (Ennett et al., 2008).
13
Reciprocation of friendships may indicate that these relationships are particularly important in
the prediction of smoking behavior, even if tie strength or best friend status is not a correlate of
smoking behaviors.
Both high centrality and low centrality have been identified as risk factors for smoking. A
common measure of centrality is degree; a person’s in-degree centrality is the number of people
in the network who nominate them and out-degree centrality is the number of people a person
nominates (Valente, 2010). In a friendship network, high in-degree centrality is indicative of
high popularity and low in-degree centrality may indicate that a person has few or no friends and
may be considered an isolate (Valente, 2010). In-degree centrality is associated with higher
cigarette smoking (Lakon & Valente, 2012; Valente, Unger, & Johnson, 2005). One study
suggested that both isolates and popular students have higher general substance use risk in
comparison to all other students (Ennett et al., 2006). Other studies suggest that the association
between centrality and smoking may be moderated by either SES or school smoking prevalence.
In lower SES schools, isolates have higher smoking rates while in higher SES schools, popular
students have higher smoking rates (Pearson et al., 2006; Seo & Huang, 2012). A similar
association is seen with school smoking prevalence. In schools with a high prevalence of
smoking, popular students are at higher risk for smoking while within schools with low smoking
prevalence popularity was inversely associated with smoking (Alexander et al., 2001). In another
study with low school smoking rates, isolates were at higher risk for smoking (Ennett &
Bauman, 1993). Furthermore, an adolescent’s smoking behaviors may influence their centrality.
A study found that smoking in the past month was associated with an increase in centrality, yet
having friends who smoked in the past month decreased one’s centrality (Lakon et al., 2010).
14
This indicates that smoking and centrality have a complex association which may be moderated
by attributes of the network such as smoking prevalence.
Personal network transitivity may be associated with smoking as well. Transitivity refers
to the tendency for two people who share a friend to become friends with one another (Valente et
al., 2004). High personal network transitivity in a friendship network is indicative of belonging
to a social group where one’s friends are friends with each other. It has been found that the
proportion of transitive triads an adolescent belongs to is protective against smoking (Ennett et
al., 2006, 2008). A recent publication found that change in transitivity in a person’s self-reported
personal network was predictive of a composite scale of positive health behaviors (O’Malley,
Arbesman, Steiger, Fowler, & Christakis, 2012). Adolescents often form friendship groups, and
lack of a cohesive friendship group may be a risk factor for smoking. Conversely, adolescents
who smoke may do so because they are exposed to multiple friendship groups and have
increased risk for exposure to smoking. Additionally, adolescents with more friends out of their
school network are more likely to smoke (Ennett et al., 2006, 2008). These adolescents may be
on the network periphery or isolates of the school network and may be part of a friendship
network of adolescents who do not attend school (Ennett et al., 2006, 2008).
Little research has explored the association between social networks and loneliness, but
the literature suggests that social networks and loneliness are associated. In adults, people with
fewer friends and people who are isolates are more likely to experience loneliness (Cacioppo et
al., 2009). Loneliness is also found in clusters in networks; peer loneliness is associated with
one’s own loneliness (Cacioppo et al., 2009). Furthermore, greater personal network size as
measured by number of self-reported social contacts is associated with less loneliness,
independent of a person’s satisfaction with their relationships (Hawkley et al., 2008).
15
Satisfaction with relationships and frequency of social contact are also predictors of reduced
loneliness (Hawkley et al., 2008; Steptoe et al., 2004). A study of college students found that
self-reported personal network density and network size is protective against loneliness (Green,
Richardson, Lago, & Schatten-Jones, 2001; Stokes, 1985). In children, higher friendship quality,
higher friendship quantity, and friendship mutuality are protective against loneliness (Ladd,
Kochenderfer, & Coleman, 1997; Nangle, Erdley, Newman, Mason, & Carpenter, 2003; Parker
& Asher, 1993; Pedersen, Vitaro, Barker, & Borge, 2007). We do not know if these findings are
applicable to adolescents. However, the literature suggests that quantity of friendships, quality of
friendships, reciprocity, and having friends who know one another are all factors that are
protective against loneliness. The findings of studies which examined loneliness and social
networks are detailed in Table 1.
Table 1: Summary of article findings on the association between social networks and
loneliness
Study Population Findings
Classroom Peer Acceptance,
Friendship, and Victimization:
Distinct Relational Systems
that Contribute Uniquely to
Children’s School Adjustment?
(Ladd et al., 1997)
200 kindergarteners
from the Midwest
Greater number of friends and greater
peer acceptance were associated with
lower loneliness. Having a very best
friend was not associated with
loneliness, controlling for other
factors.
Friendship and Friendship
Quality in Middle Childhood:
Links with Peer Group
Acceptance and Feelings of
Loneliness and Social
Dissatisfaction (Parker &
Asher, 1993)
881 3
rd
-5
th
graders Greater peer acceptance, having a best
friend, and six measures of friendship
quality were associated with lower
loneliness.
Popularity, Friendship
Quantity, and Friendship
Quality: Interactive
Influences on Children’s
Loneliness and Depression
(Nangle et al., 2003)
193 children in 3
rd
-6
th
grade in New
England
Higher friendship quantity and quality
were associated with lower loneliness.
The Timing of Middle-
Childhood Peer Rejection and
551 children aged 6
to 13 years old
Number of friends was negatively
associated with loneliness.
16
Friendship: Linking Early
Behavior to Early-Adolescent
Adjustment (Pedersen et al.,
2007)
residing in Quebec
The Relation of Social
Network and Individual
Difference Variables to
Loneliness (Stokes, 1985)
179 undergraduate
students
Network density was associated with
lower loneliness in regression model,
number of people one feels close to
was negatively correlated with
loneliness (but not significant in
regression). Neither network size nor
percent of alters that were family was
associated with loneliness.
Network Correlates of Social
and Emotional Loneliness in
Young and Older Adults
(Green et al., 2001)
91 college students
and 110 older adults
(mean age 71)
Social loneliness was associated with
average closeness for older adults and
network size for young adults in path
model. Social loneliness was
additionally correlated with the
presence of a close other for older
adults and density, closeness, and
presence of close other for younger
adults. Emotional loneliness was
correlated with density for young
adults.
Loneliness and
neuroendocrine,
cardiovascular, and
inflammatory stress responses
in middle-aged men and
women (Steptoe et al., 2004)
240 adults aged 47-
59 years old residing
in London
Social isolation was positively
associated with loneliness.
From Social Structural Factors
to Perceptions of Relationship
Quality and Loneliness: The
Chicago Health, Aging, and
Social Relations Study
(Hawkley et al., 2008)
225 adults aged 50-
68 residing in Illinois
Network size, social group
membership, and satisfaction with
social relationships were negatively
associated with loneliness.
Alone in the Crowd: The
Structure and Spread of
Loneliness in a Large Social
Network (Cacioppo et al.,
2009)
12,067 adults from
the Framingham
Heart Study (mean
age 63.8 years old)
Being friends directly or indirectly
(connected through a shared friend
with up to three degrees of separation)
with a lonely person is associated with
higher loneliness. Being on the
periphery of the social network and
having fewer contacts was associated
with higher loneliness.
17
Overview of Studies and Theoretical Background
This dissertation investigates psychosocial and network predictors of tobacco use in a
sample of majority Hispanic/Latino adolescents residing in a low to middle class SES area of
California. It intends to clarify associations among loneliness and social connection in order to
better understand how these factors influence smoking in adolescents. Loneliness is associated
with smoking and poor health behaviors in general (Christopherson & Conner, 2012; Mahon,
Yarcheski, & Yarcheski, 1998). However, there have been mixed findings concerning this
association (Dyal & Valente, 2015). Social network characteristics are also associated with
smoking (Seo & Huang, 2012). Analyzing these variables in a combined model may lead to
clarity in previous observed associations among these variables. Figure 1 illustrates the
associations among the variables to be assessed in the three proposed studies.
18
Figure 1: Conceptual model outlining associations among constructs included in the
dissertation studies
Various theories provided guidance in hypothesizing the study outcomes. Theory of
reasoned action states that personal attitudes and subjective norms about a behavior are
predictors of behavioral intention, which predicts behavior (Glanz, Rimer, & Viswanath, 2008).
Subjective norms are determined by a combination of one’s beliefs about their peers’ opinions on
the behavior and their motivation to comply with their peers (Glanz et al., 2008). In regards to
smoking and loneliness, loneliness or social isolation may increase one’s motivation to comply
with their peers to increase their sense of social connection, causing these adolescents to rate
their motivation higher than other adolescents, and resulting in those people connected to
smokers to increase their intention to smoke and those connected to nonsmokers to decrease their
intention to smoke in comparison to non-lonely participants. Attitudes are created by a
Smoking
Loneliness
Social
Networks
Identity
Acculturation
Parental
Communication
Associations assessed in studies 1, 2, & 3
Associations assessed in studies 1 & 3
Associations assessed in study 2
Associations assessed in studies 2 & 3
Associations assessed in study 3
Associations assessed in studies 1 & 2
Note. Dotted lines indicate moderational pathways. Demographics included in all studies as covariates. Study 1
additionally assesses model fit separately by ethnicity and gender. Study 2 additionally includes an analysis
predicting accuracy of participant’s perception of peer smoking with loneliness, social networks, depression,
and identity.
# Peers
who smoke
Dep. Affect
19
combination of beliefs about the likelihood of an outcome of a behavior and desirability of the
outcome (Glanz et al., 2008). Potential outcomes of smoking relevant to loneliness may include
reduction of negative affect and increased bonding with peers. In this dissertation, we do not
model attitudes or motivation to comply with a behavior but do contend that these variables may
play a role in the association between smoking outcomes and variables such as loneliness and
social connection.
While theory of reasoned action provides guidance in understanding the social influence
to smoke and how loneliness and social networks may interact in prediction of smoking in terms
of subjective norms, this study does not contain variables which assess attitude towards smoking.
To complement the theory of reasoned action and allow exploration of variables associated with
attitude, we describe self-medication hypothesis, which suggests that drug use, such as smoking
may serve as a coping mechanism to treat feelings of loneliness (Khantzian, 1985). People
experiencing loneliness may develop positive attitudes towards and expectations of smoking due
to the temporary reduction of negative affect and/or increase in positive affect experienced due to
the psychopharmacological effects of tobacco. Combining these theories, social network
variables indicating social connection and peer smoking as well as loneliness may predict
smoking independently and interactively.
Additional theories provide guidance for this study. The dynamic theory of social impact
presents an explanation of how attitudes and behaviors spread throughout and how clusters of
people with similar attributes form within a community (Latané, 1996). This theory suggests that
the majority opinion will spread throughout a community, however, minority opinions will
continue to persevere within clusters because each person’s strength of influence on one another
is a function of distance, and some clusters may not experience enough outside influence to
20
change to the majority opinion (Latané, 1996). Applied to smoking behavior, the dynamic theory
of social impact suggests that while pro-smoking behaviors and attitudes may not be held by the
majority of the network, pro-smoking attitudes may be maintained in clusters. This theory
suggests that adolescents who smoke will likely be connected to peers who smoke and their
behavior will be correlated with their peers’ behavior.
Self-categorization and social identity theories state that social identities and group
affiliation affect behavior and attitude through social norms (Hornsey, 2008; Kobus, 2003).
Social identity theory suggests that some situations cue people to increasingly view themselves
as their category stereotype and less as an individual person (Hornsey, 2008). Social identity
theory also hypothesizes that people strive for a positive, secure self-concept, which causes
people to desire to view their category in a positive manner when compared to other groups
(Hornsey, 2008). Applied to smoking, people may identify as smokers and form social groups
based on their smoking status or change their behaviors to fit that of their social groups (Kobus,
2003). This theory may be combined with the self-control for personal harm model, which when
applied to smoking suggests that people will smoke cigarettes for social benefits despite not
wanting to smoke if they perceive smoking as a means for group acceptance (Rawn & Vohs,
2011). Therefore, someone may engage in smoking not from a personal desire to smoke but as a
product of social norms and desire for secure self-concept.
21
CHAPTER 2: ASSOCIATIONS OF SOCIAL NETWORKS AND AFFECT WITH
TOBACCO AND OTHER SUBSTANCE USE IN ADOLESCENTS
Abstract
Introduction: Past research suggests that loneliness, parental communication, social network
measures, and depression are all predictors of cigarette smoking. These variables have not been
studied concurrently in one model, despite their potential conceptual overlap. This study aims to
assess the associations of loneliness, parental communication, social network measures, and
depression with use of cigarettes, e-cigarettes, hookah, alcohol, and marijuana using structural
equation modelling.
Methods: Data from N=1279 adolescents enrolled in 12
th
grade at 5 high schools in Southern
California, USA were collected with a self-report survey administered in school in 2013.
Participants answered questions assessing psychosocial factors and network characteristics,
including demographics, egocentric and school-bounded social networks, substance use, parental
communication, loneliness, and depression. A structural equation model was calculated and
reached adequate fit. Secondary analyses include multi-group models to assess differences by
gender and ethnicity.
Results: Findings indicate that poor parental communication, high depressive affect, low
popularity as assessed by closeness centrality, and nominating many friends in the egocentric
network are risk factors for substance use. Differences by ethnicity and gender were noted.
Discussion: Loneliness may not be associated with substance use in models with depressive
affect and social network indicators, however, it was a risk factor for alcohol use in females in
multi-group analyses. Substance use prevention programs for adolescents should include
22
components providing coping skills for adolescents experiencing poor parental communication
and depressive affect, and who have few friends at their school.
Introduction
Smoking rates have dropped substantially over the past couple of decades. However,
adolescents continue to begin smoking despite notable health risks (National Center for Chronic
Disease Prevention and Health Promotion (US) Office on Smoking and Health, 2012).
Furthermore, alternative tobacco products such as e-cigarettes and hookah have increased in
popularity among adolescents and are often used concurrently with combustible cigarettes (Lee,
Hebert, Nonnemaker, & Kim, 2015). Results from the 2014 Monitoring the Future survey
suggest that hookah use is rising while cigarette use has fallen among adolescents (National
Institute on Drug Abuse. High School and Youth Trends, 2014). Additionally, marijuana and
alcohol use are higher among adolescents who use either combustible or electronic cigarettes in
comparison to adolescents who do not use tobacco products, and highest among adolescents who
use both electronic and combustible cigarettes (Kristjansson, Mann, & Sigfusdottir, 2015).
Therefore, it may be important to assess cigarette smoking in a model with additional tobacco
and other substance use outcomes. This study uses a structural equation model (SEM) to identify
social and affective factors associated with the use of cigarettes, e-cigarettes, hookah, marijuana,
and alcohol in a primarily Hispanic adolescent sample.
Studies of the association between loneliness and smoking have had varied findings
(Dyal & Valente, 2015). Loneliness is defined by cognitive theorists as a negative affective state
resulting from the perception of oneself as socially isolated (Laursen & Hartl, 2013; Peplau &
Perlman, 1982). It has been differentiated from social isolation as an undesired lack of
relationships characterized by mutual understanding of life experiences by both parties (van
23
Staden & Coetzee, 2010) and differentiated from depression as an evaluation of one’s social
network, rather than an overall feeling of negative affect (Cacioppo, Hawkley, & Thisted, 2010).
People’s perceptions of an adequate social network may differ due to individual differences in
quantity and quality of social stimuli considered to be ideal/ fulfilling. Therefore, variance in
loneliness is not fully explained by objective measures of social network quality or network
structure (Cacioppo et al., 2009; Laursen & Hartl, 2013). Loneliness is also not synonymous with
depression, and has been found to predict depression in longitudinal studies (Cacioppo et al.,
2006, 2010).
Loneliness, social network characteristics, parental communication, and depression are
all important predictors of smoking during adolescence (Chaiton et al., 2009; Cohen et al., 1994;
Dyal & Valente, 2015; Seo & Huang, 2012). However, these variables have not been studied
concurrently as predictors of cigarette smoking, despite their potential association with one
another and with smoking conjointly. The primary aim of this study, aim 1.2, is to assess the
associations of loneliness and social network indicators with substance use variables while
controlling for effects of parental communication and depression in a structural equation model
(SEM). Understanding the risk factors for tobacco and other drug use and developing preventive
interventions to address these risk factors may result in a decrease in the uptake of tobacco use
by adolescents. Differences by gender and ethnicity will be assessed as aims 1.3 and 1.4. Aim
1.1 will also be addressed by examining the covariance between the social network indicators
and loneliness.
Methods
This study used the fourth wave of data of the Social Networks Study, a cohort study on
social network dynamics and adolescent risk behavior, described in further detail elsewhere
24
(Valente, Fujimoto, Unger, Soto, & Meeker, 2013). All 12th grade students enrolled at one of
five high schools located in one school district of Southern California were eligible to participate
in the study. A total of 1787 students were enrolled and on the 12th grade rosters provided by the
five schools. Of the enrolled students, 1634 (91%) were approached to participate in the survey
and 1375 students (77%) either provided consent and assent prior to the day of data collection or
were age 18 on the day of data collection and were eligible to provide consent for themselves. Of
the 1375 students eligible to take the survey, 1258 (91%) participated in the survey. An
additional 21 students who were not included on the rosters were present in class during data
collection, age 18, and provided consent and participated in the survey, for a total of 1279
participants across the 5 schools. Total school enrollment ranged from 244 to 430, participants
by school ranged from 187 to 300, and percentage of students that participated by school ranged
from 68% to 78%.
Students completed surveys in school during a regular class period time block of roughly
50 minutes. Trained data collectors visited the classrooms in small groups to administer the
survey. Students were verbally given instructions for survey completion and provided with a
survey. Students were also provided with a roster of all students in their grade including photo,
name, and ID number for completing the social network portion of the survey. Students were
ensured that their data would be kept confidential. Students were assented and provided with
parental consent forms to return to the researchers completed by a parent prior to the data
collection session. All study procedures and materials were approved by University of Southern
California’s Institutional Review Board prior to data collection.
25
Participants were asked to list their seven best friends, regardless of where they live or go
to school, in order to collect egocentric network data. They were then asked to list their seven
best friends in 12
th
grade at their school by writing their names and roster ID numbers.
Loneliness was measured using a 10-item version of the UCLA Loneliness Scale-3
(ULS), a validated scale found to be reliable in many populations (Russell, 1996). The ULS
measures loneliness with items assessing perceptions of social isolation that do not include the
word lonely in them. Example items include “How often do you feel that you lack
companionship?”, “How often do you feel that no one really knows you well?”, and “How often
do you feel that there are people you can talk to?”. Participants responded on a 4-category likert
scale with response options “never”, “rarely”, “sometimes”, and “often.”
Depressed affect was assessed with the depressed affect subscale of the Center for
Epidemiological Studies Depression scale (CES-D: Radloff, 1977). The CES-D has been found
to adequately assess depressive symptomatology in both Hispanic and non-Hispanic adolescents
(Skriner & Chu, 2014). This scale is composed of five statements, including “I felt depressed.”
and “I had trouble shaking off sad feelings.” Participants were instructed to indicate how often in
the past 7 days they felt like the statement on a 4-category likert scale including response options
“1 day or less”, “2-3 days”, “4-5 days”, and “6-7 days.”
Parental communication was measured using a 5-item measure adapted from Cohen et
al., which assesses communication as conceptualized by frequency and openness of
communication (Cohen et al., 1994). Items are measured on a 4-response likert scale with
response options “very often”, sometimes, “hardly ever”, and “never”. Example items include:
“How often do you talk to your parents about what’s on your mind?”, “How often do you ask
your parents for advice?”, and “If you had a problem, would you be able to talk to your parents
26
about it?” (the last question has response options “yes”, “maybe yes”, “maybe no”, and “no”).
Note that large values on this scale indicate poor parental communication.
Items adapted from the Youth Risk Behavior Survey, National Youth Tobacco Survey,
and Pierce et al. (Pierce, Choi, Gilpin, Farkas, & Berry, 1998) were used to assess tobacco,
alcohol, and marijuana use. Variables were aggregated to create a scale for each drug. Cigarette
use was assessed on a scale similar to Pierce et al. where cigarette use ranged from (1) non-
susceptible, (2) susceptible, (3) ever used, (4) past year use, (5) past month use, and (6) daily
smoker. Alcohol use was aggregated to a scale with categories for (1) non-susceptible, (2)
susceptible, (3) ever used, (4) past year use, (5) past month use, and (6) past month binge
drinking. Hookah, e-cigarette, and marijuana use were measured as (1) ever and (0) never use.
Socioeconomic status (SES) was assessed using a ratio number of people who live in the
participants’ home and number of rooms in their home (people per room, PPR) (D. Myers, Baer,
& Choi, 1996). Race and ethnicity were assessed with a check all that apply list of 20 different
racial/ethnic categories. This was dummy coded into a binary variable with 1=any mention of
Hispanic/Latino ethnicity and 0=no indication of Hispanic/Latino ethnicity. Gender was assessed
with the question “What is your sex?” with response options of 1=female and 0=male. Age was
assessed with the question, “How old are you?” and response options ranging from “12 years old
or younger” to “18 years old or older”. This variable was dichotomized to 1=age 18 or older and
0=under age 18. Academic performance was assessed with the question “What grades did you
get in school last year?” with responses options ranging from “Mostly A’s” to “Mostly F’s”
(higher values indicate better performance).
A structural equation model was tested including social network indicators, loneliness,
depression, and parental communication as correlated predictors of the 5 substance use
27
outcomes: cigarette use, e-cigarette use, hookah use, alcohol use, and marijuana use. Initial data
cleaning was completed in Stata (Stata/IC 14.1, 2016) and all network measure calculations and
model fitting were conducted in R (R Core Team, 2014). Social networks were created and
network statistics were computed using the statnet package (Handcock et al., 2014). The R
package lavaan was used to fit the structural equation models (Rosseel, 2012).
Network measures included were closeness centrality, in-degree centrality, reciprocity,
personal network density, and out-degree from egocentric data. Closeness was calculated using
the closeness function in the sna package by calculating the summation of the inverse of the
distance from the ego to every other node in the network. In-degree centrality was calculated
using the degree function of the sna package and calculating the number of people who
nominated each node. Proportion of friends reciprocating was calculated using the grecip
function from the sna package by calculating the percentage of friends selected by the ego who
selected the ego as a friend. Personal network density was calculated using the gden function of
the sna package to calculate the percentage of potential ties between an ego’s friends (defined as
those alters selected by the ego as a friend and those alters who selected the ego as a friend)
which exist in the network. Out-degree was calculated from the egocentric network data as the
number of people who the participant named as a friend, regardless of where they live or go to
school.
Initially, exploratory factor analysis of the loneliness, parental communication, and
depression measures was conducted. For the loneliness scale, positively worded items and
negatively worded items formed two separate factors, likely due to common method variance, as
past studies have reported (Russell, 1996). Therefore, only positively worded items were used on
the loneliness factor to reduce model fit issues. An item from the CES-D stating “I felt lonely”
28
was removed from the scale in order to reduce conflation of loneliness with depressive affect, as
has been done in previous studies (Cacioppo et al., 2010). Network variables included in the
analysis were out-degree (number of friends nominated) from the ego-centric network and in-
degree (number of friends who nominated the participant), closeness, personal network density,
and proportion of out-going ties reciprocated from the school and grade level bounded network.
The closeness variable was rescaled by multiplying it by a constant in order to obtain a ratio of
variances between the variables acceptable for SEM.
Isolates were included in the model by imputing zeros for the missing network indices,
because many network measures could not be calculated for those participants who were not
connected to the large component of the network. Missing data was imputed with k-nearest
neighbors imputation using the R package VIM (Templ, Alfons, Kowarik, & Prantner, 2015).
Since the participants were sampled from multiple schools, we did consider that
clustering may need to be accounted for. However, ICC’s ranged from .01 to .04 for the drug use
variables using Stata’s mixed, melogit, and estat icc commands. Due to the low amount of
variance in substance use accounted for by the sampling design ( ρ<.10), the final structural
equation model does not control for the random effect of school (Kline, 2011).
A confirmatory factor analysis model for the measures of loneliness, depression, and
parental communication was calculated to assess the proposed measurement models. Model
estimation was completed using diagonally weighted least squares and scaled fit statistics.
The structural equation model was fit following guidelines suggested by Kline (Kline,
2011). The model consists of the five substance use variables regressed on the loneliness,
parental communication, and depressive affect latent variables, and the social network measures.
Covariance parameters were included among the substance use variables, among the latent
29
variables, among the network measures, and between loneliness and the network measures.
Demographic variables for ethnicity, gender, age, academic performance, and SES were included
as predictors of the substance use variables, latent variables, and network measures. The model
diagram is included as Figure 3. Diagonally weighted least squares estimation with robust
standard errors and mean and variance adjusted Χ
2
statistics were used due to the inclusion of
dichotomous and categorical endogenous variables in the model (Asparouhov & Muthén, 2010;
Bandalos, 2014). Scaled model Χ
2
, scaled RMSEA, and adjusted goodness-of-fit index were used
to assess model fit.
Multi-group models were computed to detect differences in model estimation for gender
and ethnicity. The models were fit with no constraints between the parameters. Then, the model
was fit again, with the factor loadings constrained to be equal between the groups. Next, the
models were fit with factor loadings and regression weights constrained to be invariant between
the groups. Some regression weights were found to be non-invariant for both models, and score
tests were used to indicate which constraints needed to be released in order to improve model fit.
Results
Table 2 presents demographics of the sample for cases with data and for the imputed
dataset. Descriptive statistics for the imputed data set do not appear to differ from the original
data set. The analytic sample was 75% Hispanic or Latino, 52% female, and 61% age 18 or
older. School performance was self-reported as mostly C’s or better by 90% of the students.
Regarding substance use, 62% had consumed alcohol, 39% had consumed marijuana, 34% has
smoked a cigarette, 11% had used an e-cig, and 17% had smoked hookah. Means, standard
deviations, and correlations between all variables included in the final model are provided in
Table 3.
30
Figure 2 is the confirmatory factor analysis model for the three scales included in the
model. The model obtained adequate fit after dropping two indicators each for loneliness and
parental communication and one from the depressive affect scale, Χ
2
( 24) = 28.14, p = 0.254,
RMSEA = 0.012, GFI = 1.0. The latent variables in the full SEM will use this measurement
model, with three indicators for each latent variable. The Cronbach’s alphas for parental
communication, depressive affect, and loneliness are .85, .89, and .77, respectively.
Table 4 includes standardized and unstandardized coefficients for all variables regressed
on the substance use outcomes. The model obtained adequate fit Χ
2
( 124) = 148.23, p = 0.068,
RMSEA = 0.012, GFI = .988. The results suggest that parental communication, depression, and
closeness popularity had significant associations with the majority of substance use outcomes.
Specifically, poorer parental communication was associated with increased use of alcohol,
marijuana, and cigarettes. Higher depression was associated with increased use of marijuana,
cigarettes, and e-cigarettes. Higher closeness popularity was associated with decreased use of
alcohol, marijuana, and hookah. Higher egocentric outdegree was associated with increased use
of alcohol and cigarettes and higher personal network density was associated with decreased use
of cigarettes. For aim 1.1, we also examined the covariances between loneliness and the social
network indicators. None of the social network indicators has a significant covariance parameter
with loneliness. Personal network density was marginally negatively associated with loneliness
(p = .08), suggesting that having a dense personal network may be protective against loneliness.
A multi-group model was calculated with grouping by school, with no constraints.
Heterogeneity was noted in the results. The model estimation for one school was not reliable due
to an empty cell in the association between marijuana and e-cigarette use, therefore this school
was removed from the multi-group model. No students at this school had used e-cigarettes and
31
never used marijuana. E-cigarette use was removed from the model for this school and the model
was run separately for this school. Results for some schools were similar to the results for the
total sample, with the exception of loss of significance due to smaller sample size. However,
some variables, including loneliness and network density, were significant predictors of
substance use in some schools, but not for the entire sample. However, these results may be due
to estimating a model with many parameters on a small sample.
Multi-group models were computed for gender and ethnicity. The model for ethnicity
with no parameter constraints reached adequate fit Χ
2
(236) = 253.975, p = .201, RMSEA = .011,
GFI = .982. In order to obtain the most parsimonious model, we began to constrain the
parameters between the groups. First, we constrained the factor loadings. This resulted in a
model of adequate fit as well Χ
2
(242) = 258.192, p = .227, RMSEA = .010, GFI = .982.
Univariate score tests indicated that the factor loadings were invariant between the groups (ps =
.334-.931, omnibus score test Χ
2
(6) = 3.191, p = .785). Next, we constrained the regression
parameters, which resulted in adequate fit Χ
2
(334) = 373.762, p = .066, RMSEA = .014, GFI =
.974. However, the omnibus score test was significant Χ
2
(98) = 156.139, p < .0001, indicating
that the regression weights were non-invariant. We released the parameters with significant
univariate score tests and re-ran the model. This model attained adequate fit Χ
2
(322) = 296.048, p
= .847, RMSEA = 0.000, GFI = .982. The omnibus and univariate score tests were not
significant (omnibus: Χ
2
(86) = 58.013, p < .0001, univariate: ps >.05). The parameter estimates
for this model are in Table 5 and the parameters that are non-invariant are noted in the table.
Social network measures including closeness, in-degree, out-degree, and reciprocity were non-
invariant for alcohol use, with out-degree significantly positively associated with alcohol use for
non-Hispanics only and closeness significantly negatively associated with alcohol use for
32
Hispanics only. Closeness was significantly negatively associated with marijuana use for
Hispanics only. Parental communication was significantly positively associated with e-cigarette
and hookah use for non-Hispanics only.
The multi-group model for gender with no parameter constraints reached adequate fit
Χ
2
(236) = 266.97, p = .08, RMSEA = .01, GFI = .98. We constrained the factor loadings, and
this model reached adequate fit as well Χ
2
(242) = 270.86, p = .10, RMSEA = .01, GFI = .98. The
omnibus and univariate score tests did not reach significance, indicating that the factor loadings
were invariant (omnibus: Χ
2
(6) = 2.88, p = .82, univariate: ps > .05). We then constrained the
regression weights. The model Χ
2
was significant, but the model fit measures did not exhibit lack
of fit otherwise Χ
2
(334) = 394.66, p = .01, RMSEA = .02, GFI = .97. The overall score test
indicated that some of the regression weights are non-invariant Χ
2
(98) = 163.37, p < .0001. The
parameters with significant univariate score tests were unconstrained and the model was refit
until no score tests retained significance. This final model attained adequate fit Χ
2
(318) = 310.85,
p = .60, RMSEA = .00, GFI = .98. Parameter estimates are available in Table 6 and parameters
that are non-invariant are noted in the table. Loneliness and depressive affect were positively
associated with alcohol use for females only, the parameter for loneliness reached significance
while depressive affect did not. Loneliness was positively associated with marijuana use and
closeness was negatively associated with marijuana use for females, the parameter for loneliness
did not reach significance while the parameter for closeness did. Depressive affect was
significantly positively associated with cigarette use and e-cigarette use in males only. The
regression weights for in-degree and reciprocity were non-invariant for e-cigarette use and
suggested differential associations for the genders, but did not reach significance for either
gender. In-degree was significantly positively associated with hookah use, for males only.
33
Discussion
This study explored the associations of use of alcohol, marijuana, cigarettes, hookah, and
e-cigarettes with measures of affect, parental communication, and social networks to clarify past
findings and provide novel evidence of associations of psychosocial characteristics with
substance use. We found that depressive affect, parental communication, closeness centrality,
and out-degree centrality were important correlates of substance use. Loneliness was not
associated with any of the substance use outcomes in analyses for the entire sample. These
findings suggest that in a model with social network characteristics and depressive affect,
loneliness may not be associated with substance use.
The multi-group model for ethnicity suggested that there were differences in regression
weight estimates by ethnicity. Most notably, low parental communication is a risk factor for e-
cigarette and hookah use for non-Hispanics only. This finding should be considered in light of
the fact that low parental communication was a risk factor for both Hispanic and non-Hispanic
students for alcohol, marijuana, and cigarette use. Potentially this finding is due to low use of
hookah and e-cigarettes or potentially parental communication has a much greater influence over
use of alternative tobacco products for non-Hispanics in comparison to Hispanics. Additionally,
the associations between social network indicators and alcohol use varied by ethnicity. Out-
degree was a risk factor for alcohol use for non-Hispanics and closeness was protective against
alcohol and marijuana use for Hispanics only. This suggests that popularity, defined as being in a
more central position than other adolescents, is protective against alcohol and marijuana use for
Hispanic adolescents yet nominating many people as friends is a risk factor for alcohol use for
non-Hispanics. Hispanic adolescents who are on the fringes of the school network may be most
likely to use alcohol and marijuana, yet non-Hispanic students who perceive many people to be
34
their friends are at risk for alcohol use. This finding suggests conflicting risk factors for
substance use for Hispanic and non-Hispanic adolescents.
The multi-group model for gender suggested that there were differences in regression
weight estimates by gender. The analysis suggested that loneliness was a risk factor for alcohol
use and closeness was a protective factor for marijuana use among females. Closeness was also
protective against alcohol and hookah use for males and females. This suggests that both being in
a central position and feeling positive about one’s social position is protective against substance
use, and that these factors may be of particular importance for females. Depressive affect was a
risk factor for cigarette and e-cigarette use in males only, but was also a risk factor for both
males and females for marijuana use. The findings for gender suggest many similarities for
substance use risk factors for males and females, with some findings for some factors indicating
that they may be have greater influence for one gender. Being nominated by many people was a
risk factor for hookah use among males. This was the only significant finding for in-degree and
conflicts with the more consistent finding that popularity as measured by closeness centrality is
protective against substance use.
A multi-group analysis was completed to explore differences in estimation by school.
This analysis did not prove to be useful for understanding the associations among the variables.
Estimation of the model could not be completed on one school due to the small sample size, and
regression weights for the other schools generally did not reach significance, likely due to the
small samples. We did not find this analysis helpful in interpreting the results, and did not
include these detailed results in this paper.
This study is limited by some methodological factors. The data is cross-sectional, and
only contains students in the 12
th
grade. The sample had a low prevalence of tobacco use, which
35
reduced statistical power. We were missing data from a large proportion of students enrolled in
some schools. This may have biased the findings by not including those students most at risk for
substance use and could have affected calculations of network statistics. The study population
was primarily Hispanic, and results cannot be generalized to other ethnic groups and may not
generalize to Hispanic adolescents attending schools with a small percentage of Hispanic
students. However, strengths of the study include focusing on the growing Hispanic population,
analyzing loneliness as a risk factor for substance use with measures of social networks and
general negative affect, incorporating social network indicators which do not rely solely on the
participant’s perception, and studying predictors of use of hookah and e-cigarettes, which are
increasing in popularity.
Overall, results suggest that low parental communication, high depressive affect, and low
popularity are important risk factors for substance use during adolescence. Prevention programs
which include components to improve communication skills, provide coping skills for
adolescents experiencing poor family communication and depressive affect, and which appeal to
the needs of students less central in the school friendship network may be more effective in
reducing adolescent drug use in comparison to other programs.
36
Table 2: Demographic Characteristics of sample, using complete data by variable and the
imputed data set for analysis
Pre-imputation (n
differs by item)
n (%)
Post-imputation
(n = 1279)
n (%)
Female 654 (53) 671 (52)
Hispanic (of any
race)
915 (75) 962 (75)
Age 18 747 (61) 785 (61)
Academic
performance
Mostly A’s 116 (10) 117 (9)
Mostly A’s and B’s 337 (28) 365 (29)
Mostly B’s 97 (8) 105 (8)
Mostly B’s and C’s 411 (34) 449 (35)
Mostly C’s 107 (9) 111 (9)
Mostly C’s and D’s 104 (9) 104 (8)
Mostly D’s 9 (1) 9 (1)
Mostly D’s and F’s 15 (1) 15 (1)
Mostly F’s 4 (<1) 4 (<1)
People per room
(PPR) mean (sd)
1.8 (.91) 1.8 (.89)
Eligible free/reduced
lunch
1078 (89) -
Drink
Nonsusceptible 339 (29) 339 (27)
Susceptible 152 (13) 156 (12)
Ever Drink 105 (9) 116 (9)
Past Year 170 (14) 191 (15)
Past month 201 (17) 251 (20)
Binge Drink 212 (18) 226 (18)
Marijuana 446 (38) 494 (39)
Cigarette
Nonsusceptible 692 (57) 736 (58)
Susceptible 97 (8) 105 (8)
Ever Smoke 120 (10) 126 (10)
Past Year 132 (11) 140 (11)
Past Month 135 (11) 141 (11)
Daily 31 (3) 31 (2)
E-cigarette 120 (10) 142 (11)
Hookah 195 (17) 214 (17)
37
Table 3: Means, standard deviation, and correlation matrix for variables included in SEM
µ σ 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Loneliness.1 (1) 2.60 .81
Loneliness.2 (2) 2.84 .86 .54
Loneliness.3 (3) 2.68 .85 .54 .53
Dep. Affect.1 (4) 1.77 1.00 .30 .25 .27
Dep. Affect.2 (5) 1.63 1.00 .28 .29 .30 .72
Dep. Affect.3 (6) 1.76 1.03 .36 .29 .34 .73 .72
P. Comm.1 (7) 2.23 .94 .08 .07 .06 .06 .09 .08
P. Comm.2 (8) 2.39 .95 .07 .08 .02 .05 .05 .04 .72
P. Comm.3 (9) 2.99 .96 .05 .04 .03 .02 .02 .06 .62 .61
Closeness centrality (10) 2.30 1.10 .03 -.03 -.02 .01 .02 .03 .00 -.02 -.01
Out-degree (11) 5.11 1.80 -.01 -.02 -.05 .04 .01 .02 -.06 -.02 -.04 .31
In-degree (12) 2.92 2.18 .01 -.02 -.03 .01 -.02 .01 -.03 -.03 -.06 .22 .13
Reciprocity (13) .46 .36 .00 -.02 -.02 .01 -.03 .02 -.05 -.04 -.08 .22 .03 .57
Density (14) .15 .17 -.01 -.04 -.07 -.03 -.04 -.04 .00 .02 .00 .05 .02 .11 .25
Alcohol (15) 3.42 1.88 .01 .06 .01 .07 .07 .07 .05 .08 .10 -.04 .08 -.01 .01 -.03
Marijuana (16) .39 .49 .00 .07 .01 .07 .11 .07 .06 .09 .07 -.05 .02 -.01 -.01 -.04 .51
Cigarette (17) 2.17 1.57 .01 .02 -.02 .07 .09 .07 .08 .10 .07 -.03 .04 -.01 -.02 -.06 .47 .54
E-cigarette (18) .11 .31 .00 .01 -.03 .05 .07 .04 .08 .08 .04 .01 .04 .01 -.01 -.01 .29 .31 .46
Hookah (19) .17 .37 -.04 -.02 -.03 .00 .02 .02 .01 .02 .01 -.04 .02 .02 -.01 -.06 .37 .41 .48 .45
Hispanic (20) .75 .43 -.13 -.05 -.12 .01 .01 .00 -.11 -.10 -.02 .07 .05 -.02 .04 .04 .28 .27 .18 .09 .16
Female (21) .52 .50 .06 .07 .15 .15 .10 .16 -.20 -.15 -.16 .03 .04 .03 .15 .03 .01 -.05 -.13 -.18 -.06 .08
Age (22) .61 .49 .01 .05 .03 .05 .07 .02 .00 .00 -.02 -.09 -.02 -.03 -.06 -.09 .05 .03 .08 .06 .06 .02 -.04
Academic Perf. (23) 6.59 1.56 .05 .00 .06 -.04 -.08 -.04 -.07 -.07 -.06 .01 .06 .06 .04 .06 -.15 -.20 -.20 -.07 -.12 -.30 .12 -.03
PPR (24) 1.79 .89 -.05 -.01 .00 .01 -.01 -.02 -.03 -.05 -.03 .06 .01 -.01 -.02 .00 .09 .06 .05 .03 .04 .29 .02 .02 -.10
Note. Results obtained from imputed dataset.
38
Table 4: Results from structural equation model for entire sample
Alcohol Marijuana Cigarette E-cigarette Hookah
Β b (se) Β b (se) Β b (se) Β b (se) Β b (se)
Loneliness .06 .10 (.06) .05 .09 (.08) .00 -.01 (.07) -.02 -.04 (.10) -.04 -.07 (.09)
Dep. affect .04 .05 (.05) .08* .12 (.06) .11** .14 (.05) .14** .18 (.07) .04 .06 (.06)
P. comm .11*** .14 (.04) .11** .14 (.05) .09** .11 (.04) .08 .11 (.06) .02 .03 (.06)
Closeness -.10*** -.10 (.03) -.10** -.10 (.04) -.05 -.05 (.03) .00 -.00 (.05) -.09* -.09 (.04)
Out-degree .11*** .06 (.02) .05 .03 (.02) .07* .04 (.02) .06 .04 (.03) .05 .03 (.02)
In-degree .00 .00 (.02) .02 .01 (.02) .00 .002 (.02) .02 .01 (.03) .07 .03 (.02)
Reciprocity .04 .11 (.11) .01 .03 (.13) .03 .08 (.12) .00 -.00 (.18) -.01 -.02 (.15)
Density -.05 -.28 (.19) -.05 -.33 (.22) -.07* -.43 (.20) -.01 -.05 (.28) -.09 -.57 (.30)
Age .03 .07 (.06) .00 .001 (.08) .06* .14 (.07) .06 .14 (.10) .06 .13 (.09)
Hispanic .29*** .71 (.08) .37*** .93 (.10) .21*** .51 (.08) .19*** .46 (.13) .26*** .64 (.13)
Female -.01 -.03 (.06) -.08* -.17 (.08) -.15*** -.32 (.07) -.30*** -.64 (.11) -.09* -.19 (.09)
Grade -.06* -.04 (.02) -.14*** -.10 (.02) -.14*** -.09 (.02) -.02 -.01 (.03) -.09* -.06 (.03)
PPR .02 .02 (.03) -.01 -.02 (.04) .00 .003 (.04) .01 .01 (.06) -.01 -.01 (.05)
Note. * indicates p <.05, ** indicates p <.01, *** indicates p <.001. Table includes standardized and unstandardized regression
weights for regression parameters for the five drug use outcomes. Additionally included in the model were regressions from
demographic variables to all other predictors, covariances among loneliness, depressive affect, and parental communication,
covariances between loneliness and each of the social network variables, covariances among the substance use outcomes, and
covariances among all the network variables.
39
Table 5: Results from multi-group model assessing differences by ethnicity
Alcohol Marijuana Cigarette E-cigarette Hookah
Non-
Hispanic
Β b (se) Β b (se) Β b (se) Β b (se) Β b (se)
Loneliness .05 .11 (.07) .03 .07 (.08) -.01 -.01 (.07) -.01 -.02 (.10) -.04 -.08 (.09)
Dep. affect .04 .05 (.05) .10* .13 (.06) .11** .15 (.05) .13** .17 (.07) .05 .06 (.06)
P. comm .12*** .15 (.04) .13** .16 (.05) .10** .12 (.04) .29***† .35 (.10) .25*† .31 (.12)
Closeness -.07† -.06 (.07) .04† .03 (.08) -.06 -.05 (.03) .001 .001 (.05) -.09* -.09 (.04)
Out-degree .30***† .15 (.04) .05 .03 (.02) .08* .04 (.02) .09 .05 (.02) .06 .03 (.03)
In-degree .04† .02 (.03) .05 .02 (.02) .01 .004 (.02) .01 .002 (.03) .08 .03 (.02)
Reciprocity .10† .30 (.22) -.001 -.004 (.14) .02 .07 (.12) .02 .05 (.17) -.01 -.04 (.15)
Density -.04 -.31 (.19) -.05 -.36 (.21) -.06* -.43 (.20) .01 .06 (.28) -.08 -.54 (.30)
Age .03 .07 (.06) .01 .01 (.08) .06 .13 (.07) .06 .13 (.11) .06 .13 (.09)
Female -.01 -.02 (.06) -.07 -.14 (.08) -.15*** -.30 (.07) -.06† -.13 (.27) -.10* -.21 (.10)
Grade -.06* -.04 (.02) -.13*** -.09 (.03) -.12*** -.09 (.02) -.02 -.01 (.03) -.08 -.06 (.03)
PPR .01 .02 (.03) -.01 -.02 (.04) .13† .23 (.12) .003 .01 (.06) -.01 -.01 (.05)
Hispanic
Loneliness .07 .11 (.07) .04 .07 (.08) -.01 -.01 (.07) -0.01 -.02 (.10) -.05 -.08 (.09)
Dep. affect .04 .05 (.05) .10* .13 (.06) .12** .15 (.05) .13** .17 (.07) .05 .06 (.06)
P. comm .12*** .15 (.04) .13*** .16 (.05) .09** .12 (.04) .02† .03 (.07) -.02† -.03 (.06)
Closeness -.13***† -.12 (.03) -.14***† -.13 (.04) -.06 -.05 (.03) .001 .001 (.05) -.09* -.09 (.04)
Out-degree .06† .04 (.02) .05 .03 (.02) .07* .04 (.02) .08 .05 (.02) .06 .03 (.03)
In-degree -.01† -.003 (.02) .04 .02 (.02) .01 .004 (.02) .005 .002 (.03) .07 .03 (.02)
Reciprocity .02† .05 (.12) -.001 -.004 (.14) .02 .07 (.12) .02 .05 (.17) -.01 -.04 (.15)
Density -.05 -.31 (.19) -.06 -.36 (.21) -.07* -.43 (.20) .01 .06 (.28) -.09 -.54 (.30)
Age .03 .07 (.06) .01 .01 (.08) .06 .13 (.07) .06 .13 (.11) .06 .13 (.09)
Female -.01 -.02 (.06) -.07 -.14 (.08) -.15*** -.30 (.07) -.36***† -.77 (.12) -.10* -.21 (.10)
Grade -.06* -.04 (.02) -.13*** -.09 (.03) -.13*** -.09 (.02) -0.02 -.01 (.03) -.09 -.06 (.03)
PPR .02 .02 (.03) -.02 -.02 (.04) -.03† -.03 (.04) .005 .01 (.06) -.01 -.01 (.05)
Note. * indicates p <.05, ** indicates p <.01, *** indicates p <.001, and † indicates parameter is non-invariant between the groups.
Table includes standardized and unstandardized regression weights for regression parameters for the five drug use outcomes.
40
Additionally included in the model were regressions from demographic variables to all other predictors, covariances among loneliness,
depressive affect, and parental communication, covariances between loneliness and each of the social network variables, covariances
among the substance use outcomes, and covariances among all the network variables.
41
Table 6: Results from multi-group model assessing differences by gender
Alcohol Marijuana Cigarette E-cigarette Hookah
Male Β b (se) Β b (se) Β b (se) Β b (se) Β b (se)
Loneliness .01† .02 (.09) .001† .001 (.10) -.004 -.01 (.07) -.02 -.03 (.10) -.04 -.07 (.09)
Dep. affect -.004† -.005 (.07) .09* .12 (.06) .15**† .20 (.06) .18**† .24 (.08) .04 .06 (.06)
P. comm .10*** .14 (.04) .11** .15 (.05) .08** .11 (.04) .08 .10 (.06) .01 .02 (.06)
Closeness -.10** -.09 (.03) -.03† -.03 (.05) -.06 -.06 (.03) -.001 -.001 (.05) -.09* -.09 (.04)
Out-degree .11*** .06 (.02) .05 .03 (.02) .08* .04 (.02) .07 .04 (.03) .06 .03 (.03)
In-degree -.001 .000 (.02) .03 .02 (.02) -.01 -.004 (.02) .001† .000 (.03) .13*† .06 (.03)
Reciprocity .04 .11 (.11) .005 .01 (.14) .04 .13 (.12) .11† .33 (.22) .003 .01 (.16)
Density -.04 -.27 (.19) -.06 -.34 (.22) -.07* -.45 (.20) -.02 -.09 (.28) -.09 -.57 (.30)
Age
.03 .06 (.06) .002 .004 (.08) .13**† .28 (.10) .14*† .31 (.13) .16**† .35 (.13)
Hispanic
.30*** .70 (.08) .31***† .73 (.13) .21*** .50 (.08) .20*** .46 (.13) .27*** .64 (.13)
Grade
-.06* -.04 (.02) -.14*** -.09 (.03) -.14*** -.09 (.02) -.02 -.01 (.03) -.10* -.07 (.03)
PPR .02 .02 (.03) -.01 -.02 (.04) .001 .001 (.04) .02 .03 (.06) -.01 -.01 (.05)
Female
Loneliness .12*† .21 (.09) .09† .18 (.11) -.004 -.01 (.07) -.02 -.03 (.10) -.04 -.07 (.09)
Dep. affect .08† .11 (.06) .08* .12 (.06) .07† .10 (.07) .05† .07 (.10) .04 .06 (.06)
P. comm .11*** .14 (.04) .11** .15 (.05) .09** .11 (.04) .09 .10 (.06) .01 .02 (.06)
Closeness -.10** -.09 (.03) -.15**† -.16 (.05) -.06 -.06 (.03) -.001 -.001 (.05) -.09* -.09 (.04)
Out-degree .11*** .06 (.02) .04 .03 (.02) .07* .04 (.02) .06 .04 (.03) .06 .03 (.03)
In-degree -.001 .000 (.02) .03 .02 (.02) -.01 -.004 (.02) .07† .03 (.05) -.02† -.01 (.04)
Reciprocity .04 .11 (.11) .005 .01 (.14) .04 .13 (.12) -.19† -.54 (.30) .003 .01 (.16)
Density -.04 -.27 (.19) -.05 -.34 (.22) -.07* -.45 (.20) -.01 -.09 (.28) -.09 -.57 (.30)
Age
.03 .06 (.06) .002 .004 (.08) .004† .01 (.10) -.06† -.12 (.17) -.03† -.07 (.13)
Hispanic
.28*** .70 (.08) .44***† 1.22 (.17) .20*** .50 (.08) .18*** .46 (.13) .25*** .64 (.13)
Grade
-.06* -.04 (.02) -.12*** -.09 (.03) -.13*** -.09 (.02) -.02 -.01 (.03) -.10* -.07 (.03)
PPR
.02 .02 (.03) -.01 -.02 (.04) .001 .001 (.04) .02 .03 (.06) -.01 -.01 (.05)
Note. * indicates p <.05, ** indicates p <.01, *** indicates p <.001, and † indicates parameter is non-invariant between the groups.
Table includes standardized and unstandardized regression weights for regression parameters for the five drug use outcomes.
Additionally included in the model were regressions from demographic variables to all other predictors, covariances among loneliness,
42
depressive affect, and parental communication, covariances between loneliness and each of the social network variables, covariances
among the substance use outcomes, and covariances among all the network variables.
43
Note. Model contains standardized factor loadings and correlations between latent variables.
Error terms are not included in the diagram. All parameters obtained significance at the p <.001
level with the exception of covariance between loneliness and parental communication and
depression and parental communication, which obtained significance at the <.01 and <.05 levels,
respectively.
Lone1
Lone2
Lone3
Dep.1
Dep.2
Dep.3
P. Comm3
P. Comm2
P. Comm1
Depression
Parental
Communication
Loneliness
.09
.08
.76
.71
.73
.87
.84
.72
.81
.90
.84
.48
Figure 2: Confirmatory Factor Analysis Model
44
Note. Covariances are also included among all of the substance use outcomes, among the social
network measures and loneliness, and among loneliness, depressive affect, and parental
communication. Error variances are also included in the model but not included here for
simplicity.
Loneliness
Depressive
Affect
Parental
Communication
Closeness
Out-Degree
In-Degree
Reciprocity
Cigarette
Hookah
E-Cigarette
Marijuana
Network
Density
Alcohol
Figure 3: Main Structural Equation Model
45
CHAPTER 3: ACCURACY AND INFLUENCE OF PERCEPTION OF PEER
CIGARETTE USE: MODERATION BY LONELINESS
Abstract
Introduction: Past studies have found that loneliness and peer substance use is associated with
cigarette and alcohol use. Research suggests that lonely people have greater conformity with peer
behavior in comparison to non-lonely people. This study explores loneliness as a potential
moderator for the effect of peer influence on cigarette and alcohol use.
Methods: Adolescents (n = 1279) enrolled in 12
th
grade at five high schools in Southern
California, USA completed an in-class survey including items on friendship networks,
loneliness, depression, identity, cigarette and alcohol use, and demographics. Outcomes of
accuracy of peer substance use perception, number of friends who smoke/drink, and participants’
own cigarette and alcohol use behaviors were analyzed using logistic and linear regression.
Dyadic analysis was also completed for the outcome of the participant and friend using cigarettes
or alcohol with one another.
Results: Out-degree was associated with lower accuracy of peer behavior report and
stoner/druggie identity was associated with greater accuracy of peer behavior report. Adolescents
who were lonelier, named more friends, and identified as a stoner/druggie had higher odds of
having at least one friend who smokes and at least one friend who uses alcohol. Stoner/druggie
identity and having at least one friend who uses alcohol/cigarettes was associated with higher
odds of using cigarettes and higher frequency of drinking. Depression was also associated with
higher odds of using cigarettes. No interaction was found between depression and loneliness with
other predictor variables in predicting substance use. Dyadic analyses did find significant
interaction terms for loneliness and depression with friend cigarette/alcohol use.
46
Discussion: Loneliness did not moderate the effect of peer influence on smoking or drinking
behavior. However, lonely adolescents may have greater exposure to peer substance use.
Understanding the directionality of this relationship may inform prevention program design.
Introduction
Socially isolated adolescents may be predisposed to form inaccurate perceptions of peer
smoking as they have fewer sources of peer information about social norms in comparison to
their well-connected peers. Lonely adolescents may have an increased drive to behave in
accordance with their peers in order to increase their feelings of connection to their social
network and identification with a social identity. This paper addresses how loneliness and social
networks affect the accuracy of perceptions of others’ smoking behaviors and drive to act in
accordance with peer smoking norms.
Social identity theory suggests that people will participate in activities that they believe will
make them be perceived as part of the group that they identify with and people will identify with
groups that they perceive to fit their behavior (Kobus, 2003; Stewart-Knox et al., 2005). In this
theory, peer influence occurs through a need to be in accordance with perceived social norms as
opposed to pressure from peers (Stewart-Knox et al., 2005). Adolescents who smoke may choose
social groups which fit their perceived identity as a smoker and smoking behavior (Kobus,
2003). Conversely, adolescents may smoke to enhance their feelings of identification with a
social subgroup they already belong to (Kobus, 2003). Some social identities, particularly those
considered to be deviant or elite, are associated with increased tobacco use and may increase
tobacco use through peer norms (Sussman et al., 1990; Sussman, Pokhrel, Ashmore, & Brown,
2007). Studies suggest that self-reported social identity is associated with the number of close
friends an adolescent has who smoke, and that adolescents tend to have best friends who identify
47
with the same social identity they do (La Greca, Prinstein, & Fetter, 2001). Therefore,
adolescents tend to socialize with other adolescents within their social identity crowd, which may
cause some adolescents to be exposed to higher than average peer smoking norms (Kobus,
2003). Social learning theory suggests that people may learn behaviors through exposure to their
peers’ behaviors and the social response to their behavior (Kobus, 2003). Laboratory studies of
adolescents who smoke have demonstrated that exposure to a smoking peer increases the number
of cigarettes smoked and pace of smoking during an experimental session (Kniskern, Biglan,
Lichtenstein, Ary, & Bavry, 1983). Therefore, adolescents’ choice of peer group may increase
peer influence to smoke due to the perception that smoking is in congruence with their social
identity or their increased exposure to smoking peers who may model smoking behaviors.
Marketing research studies have found that loneliness and social exclusion affect product
choice. In a laboratory study involving purchasing an item following an exclusion manipulation,
participants were more likely to choose items that increased their affiliation to the majority group
in comparison to participants who did not receive the manipulation (Mead, Baumeister, Stillman,
Rawn, & Vohs, 2011). In a subsequent study, excluded participants were more likely to rate
highly products they believe their assigned partner would like in comparison to control
participants (Mead et al., 2011). These findings are in line with research which suggests that
people who are socially rejected retain social cues more than other information (Gardner, Pickett,
& Brewer, 2000; Heinrich & Gullone, 2006). Lonely people are more likely to indicate
preference for popular products when their preferences are public, and more likely prefer
products not endorsed by the majority when their preference is kept private in comparison to
nonlonely participants (Wang, Zhu, & Shiv, 2012). People who are not lonely are more likely to
endorse the majority preference regardless of privacy in comparison to lonely participants (Wang
48
et al., 2012). These studies taken together indicate that lonely and socially excluded people are
exceptionally attentive to social cues and may alter their preferences to act in accordance with
others’ opinions, even though lonely people may be inclined to prefer products without majority
endorsement.
Substance use attitudes may be manipulated by loneliness and social exclusion in the
same way as product preference. Participants in a laboratory study who were manipulated to feel
socially excluded were more likely to indicate openness to try cocaine in a public situation with
friends in an experimental scenario in comparison to control participants (Mead et al., 2011).
Socially excluded participants did not differ from control participants in willingness to try
cocaine in a scenario where their friends would not be aware of their consumption (Mead et al.,
2011). This study in addition to those described above concerning product choice suggest that
excluded people do not use spending and drug consumption for personal identity enhancement or
coping, but instead use it as a bonding mechanism (Mead et al., 2011). Tobacco researchers have
also suggested that smoking a cigarette may be used as a bonding activity between adolescents
and may be used to strengthen or create friendships (Lakon et al., 2010). Additionally, mood was
controlled for in the analyses of the summarized laboratory studies and was not associated with
willingness to try cocaine, further suggesting that product choice/drug consumption was not due
to coping/self-medication (Mead et al., 2011). Lastly, the studies of product preference suggested
that products with minority preference are more in line with feelings of loneliness, however
lonely participants changed their preferences to majority preference products when their
preferences would be public (Wang et al., 2012). Research does support that lonely people are
less likely to take social risks (Heinrich & Gullone, 2006). Loneliness and social exclusion may
increase susceptibility to peers’ influence and use of smoking for friendship bonding, and may
49
also induce one to identify privately with the minority preference as opposed to the majority
preference.
The literature concerning how sociometric measures of social network structure may
affect susceptibility to peer influence beyond measurement of the number of friends an
adolescent has who smoke is unclear. One study found that those adolescents who have high in-
degree centrality and high friendship quality are most influenced by their friend’s smoking
behavior (Urberg et al., 2003). However, another study found that adolescents with
unreciprocated friendships are more likely to be influenced by their friend’s smoking in
comparison to those with reciprocated friendships (Aloise-Young, Graham, & Hansen, 1994). In
a study of delinquent behavior, it was found that only stable friendships exert peer influence
(Laursen, Hafen, Kerr, & Stattin, 2012). Furthermore, within stable dyadic friendships, the
person with more friends has been shown to have greater influence over the partner with less
friends in delinquent behavior (Laursen et al., 2012). In contrast, some adolescents begin to
smoke despite having few or no friends who smoke. Research suggests that adolescents who
smoke and have a majority of friends who do not smoke have high in-degree centrality (Lakon et
al., 2010). However, other researchers posit that adolescents on the periphery of a network may
feel more freedom to try behaviors not accepted by the majority such as smoking (Kobus, 2003).
Overall, it is unclear how peer influence may differentially affect adolescents with varied
network positions and characteristics.
Identity may be a particularly important concept for lonely people. Those adolescents
who conceptualize their social isolation as being caused by a personal flaw are more likely to
remain lonely (Laursen & Hartl, 2013; Peplau & Perlman, 1982). Viewing one’s loneliness as
stable may affect the identity formation processes that occur during adolescence. Evidence
50
suggests that people who experience high levels of loneliness perceive a greater discrepancy
between their actual and ideal selves in comparison to people who do not experience great
loneliness (Heinrich & Gullone, 2006). Potentially the cognitive dissonance which results from
this perceived discrepancy may motivate lonely adolescents to engage in overt behaviors which
display their identity. Adolescents are able to form identities based on more abstract contexts
than they were able to at an earlier age, including on their social relationships (Heinrich &
Gullone, 2006). Furthermore, adolescents are concerned with how they appear to others
(Sanders, 2013), and may use cigarettes as an accessory to enhance their appearance (Kobus,
2003). The identity formation process may prove particularly difficult for lonely adolescents who
may not view their social relationships as indicators of an identity and may seek other means to
express who they are socially.
Methods
This study uses cross-sectional data from wave four of the Social Network Study.
Methodology and aims for the parent study are described in greater detail elsewhere (Valente,
Fujimoto, Unger, et al., 2013). Study participants were 1279 adolescents enrolled in 12
th
grade at
one of five high schools located in Southern California, USA. All 12
th
grade students who were
enrolled at these five schools were eligible to participate. Participation and consent rates and data
collection procedures are detailed in Study 1 of this dissertation. Briefly, students’ parents
provided consent and students provided assent prior to participating. Students who were age 18
or older on the day of data collection were allowed to provide consent for themselves to
participate in the study. Students completed paper and pencil surveys during class time. Surveys
contained items on social networks, drug use, demographics, and psychosocial characteristics.
51
All study materials and methods were approved by the University of Southern California’s
Institutional Review Board.
Social network data were collected with the prompt “Please think of your seven BEST
FRIENDS regardless of where they live or go to school. Be sure to write your friends’ real
names and not their nicknames”. Subsequent questions queried additional information about each
friend including “Has this person ever smoked a cigarette?”, “Does this person drink alcohol at
least once a month?”, “Have you ever smoked a cigarette with him/her?”, and “Have you ever
drunk alcohol with him/her?”, with yes/no response options. Additional questions assessed
demographics for each friend.
Participants’ own smoking status was assessed with the question “Have you ever tried
cigarette smoking, even one or two puffs?” with response options yes or no. Participants’
drinking behavior was reported with a composite scale ranging from 1-6 with categories (1) non-
susceptible never-drinker, (2) susceptible never-drinker, (3) ever drinker, (4) past year drinker,
(5) past month drinker, and (6) binge drank in past month, created from responses to questions
“At any time in the next year (12 months) do you think you will drink alcohol?”, “During your
life, on how many days have you had at least one drink of alcohol? (please do not count drinking
alcohol for religious purposes like communion wine)”, “ During the past 12 months, how often
did you drink alcohol (beer, wine, or liquor)?”, “During the past 30 days, on how many days did
you have at least one drink of alcohol?”, and “During the past 30 days, on how many days did
you have 5 or more drinks of alcohol in a row, that is, within a couple of hours?”. Questions used
to assess alcohol and tobacco use were adapted from the Youth Risk Behavior Survey and Pierce
et al. (Pierce et al., 1998).
52
Loneliness was assessed with a 10-item version of the UCLA Loneliness Scale-3 (ULS;
Russell, 1996). The original ULS contains 20 items which assess a global loneliness factor
without using the word lonely. The 10-item version contains questions from the original, 5 which
are negatively worded, such as “How often do you feel left out?” and 5 which are positively
worded, such as “How often do you feel that there are people who really understand you?”. Item
response options are (1) Never, (2) Rarely, (3) Sometimes, and (4) Always. Positively worded
items were reverse-scored and the mean of all 10 items was computed to create a scale where
higher values indicate higher loneliness. The Cronbach’s Alpha for these items was .86,
indicating that adequate reliability was reached for this sample.
Depressive affect was assessed with the depressive affect subscale of the Center for
Epidemiological Studies Depression Scale (CES-D; Radloff, 1977) . The CES-D depressive
affect subscale is composed of five questions assessing how frequently during the past 7 days the
participant experienced depressive symptomatology with response options (1) 1 day or less, (2)
2-3 days, (3) 4-5 days, and (4) 6-7 days. Example questions include “I had trouble shaking off
sad feelings” and “I felt depressed.” The subscale also includes the item “I felt lonely” which
was removed from the scale to avoid conflating depressive affect and loneliness, as has been
done in previous studies (Cacioppo et al., 2010). The Cronbach’s Alpha for the remaining four
items was .88.
In order to assess identity, participants were provided with a list of 22 social identities
and were instructed, “Sometimes people give names to the group of people that they are friends
with. Which group or groups would you say you belong to? (You can choose more than one
group).” We retained one social identity from this list, Stoners/Druggies, coded (1) yes or (0) no.
53
Demographics variables included ethnicity, socioeconomic status, academic performance,
age, and gender. Ethnicity was collected with a check all that apply list of ethnic/racial
classifications, dummy coded as (1) any indication of Hispanic/Latino ethnicity or (0) no
indication of Hispanic/Latino ethnicity. Socioeconomic status was assessed with a measure of
overcrowding called people per room (PPR; D. Myers et al., 1996). This measure is a ratio of the
number of people who live in the participants’ homes, including themselves, and the number of
rooms in the home, not including the kitchen or bathroom (D. Myers et al., 1996). Age was
assessed with the question “How old are you?”, and was dichotomized as (1) age 18 or older or
(0) under age 18. Gender was assessed with the question “What is your sex?” with response
options (1) female or (0) male. Academic performance was assessed with the question “What
grades did you get in school last year? Please choose only ONE answer.” with response options
ranging from (9) “Mostly A’s” to (1) “Mostly F’s”.
The social network data was processed for analysis. The participant’s out-degree was the
number of friends they listed. The proportion of friends who smoke was calculated as the
proportion of the participant’s friends that the participant reported had ever smoked. The
proportion of friends who drank was calculated as the proportion of the participant’s friends that
the participant reported drank alcohol at least once a month.
In order to calculate the participant’s accuracy in their reports of their friend’s smoking
and drinking behavior, we matched the participants’ reports of their alters to data the alters’
provided on their own surveys, for those alters who completed the survey as well. To do this, we
used additional social network data provided by the participants for a friendship network of
students who are enrolled at the participants’ school in their same grade, as well as the roster of
students attending the school in the participant’s grade, either of which allowed us to pull an ID
54
number for the alter to link to the alter’s self-report data. We edited the names provided in the
egocentric friendship data, school network friendship data, and roster data, by changing all letters
to lower case and putting the first and last names of the alters into one character string, with no
space between the first and last name.
Participant listed 6543 people as friends, or roughly 5.1 friends per participant. We used
the match function in R, which identifies perfect matches for character strings. We then used the
amatch function with the Jaro-Winker distance algorithm in the R package stringdist for
approximate string matching (van der Loo, 2014). This algorithm was originally designed by the
US Bureau of the Census to match records, and performs well when matching character strings
that may have mistakes from data entry, such as misspelled names (van der Loo, 2014). The
matches which were approximate, but not exact, were checked to make sure that they were
reasonably close. Any match that did not appear to be a typo or an alternate spelling of a name
was discarded. Matching was completed separately with the participants’ nominations from their
school network and the roster. Matches from both sources were used in order to include all
possible alters in the accuracy analysis.
After removing any match for alters the participants indicated did not go to their school,
the final number of matches was 3442. Overall, participants indicated that 5018, or 77% of their
friends attended their school. We were able to match 69% of these 5018 participants. The
remaining alters that attended the school were either in other grades at the school or we failed to
match them. These 3442 alters that were matched were compared to the 3101 alters that we were
unable to match using Χ
2
tests and t-tests. The alters that we were unable to match did not differ
from the alters we matched on number of years the participant had known them (p = .91), but
were older, lived farther away from the participant, were more likely to be males, more likely to
55
use cigarettes, and more likely to use alcohol (p < .05). We used this limited data for analyses of
accuracy only, and use the full data set for all other analyses.
We had self-report smoking data on 2582 of the participants’ friends, which is 75% of the
matched sample and 39% of the total number of alters. We had both self-report and participant
perception data on smoking for 2549 records, or 39% of the total alters. For the entire sample,
1027 of the participants had smoking data from at least one friend. We had self-report alcohol
use data on 2513 of the participants’ friends, which is 73% of the matched sample and 38% of
the total alters. We had both alter self-report and participant report data on 2468 of the alters, or
38% of the total alters. For 1019 participants, there was data on at least one friend.
Four different accuracy measures were created separately for cigarette and alcohol use.
Cigarette use was assessed as ever use for both alter self-report and participant report of alter.
Alter alcohol use was reported by the participant with the question “Does this person drink
alcohol at least once a month?” and was self-reported by alters with the question “During the
past 12 months, how often did you drink alcohol (beer, wine, or liquor)?” with the response
option of “once a month or less (3-12 times in the past 12 months)” or higher considered an
affirmative response.
The friend-alter dyads were coded as either (1) Congruent report that alter uses cig/alc,
(2) Congruent report alter does not use cig/alc, (3) Participant incorrectly reports that the alter
uses substance, or (4) Participant incorrectly reports that the alter does not use the substance.
From this coding schema, we were able to calculate various measures assessing the participants’
accuracy. Proportion correct was the number of friends the participant correctly reported
smoking behavior for divided by the total number of friends with data (groups 1 + 2 / groups 1 +
2 + 3 + 4). Proportion of users correct was the number of alters the participant correctly
56
identified as smokers divided by the total number of their alters who smoke (group 1 / groups 1 +
4). Underreporting was the number of alters the participant incorrectly identified as non-users
divided by the total number of alters who they reported to be non-users (group 4 / groups 2 + 4).
Over-reporting was the number of alters the participant incorrectly identified as users divided by
the total number of alters the participant reported as users (group 3/ groups 1 and 3).
Data were imputed using k-nearest neighbors regression with the R package VIM (Templ
et al., 2015). Data were not imputed for variables measuring participants’ reports of their friends’
behaviors, such as questions “Has this person ever smoked a cigarette?” because we did not think
the data set contained enough information on the participants’ perception of their alters to
adequately predict these missing values. Therefore, data from dyads which the participant did not
have data for are not included in analyses.
The data were assessed for distribution. It was noted that the accuracy measures were
poorly distributed. Many of these measures had a score of zero or one for half of the participants
to be included in accuracy analyses. Therefore, we dichotomized the accuracy measures. The
same pattern was noted for measures of the proportion of the participant’s network that smoke
and the proportion of the participant’s network that drink. These measures were dichotomized to
be (0) no friends who use substance and (1) at least one friend who uses substance.
Mixed effects linear and logistic regression models including a random intercept for
school were computed for the outcome measures of accuracy (aim 2.1), peer behavior (aim 2.2),
and cigarette and alcohol use behavior (aim 2.3) using Stata’s mixed and melogit commands
(Stata/IC 14.1, 2016). Ethnicity, gender, SES, academic performance, and age were included as
control variables in all models. Loneliness, depressive affect, out-degree, having at least one alter
who smokes/drinks, and stoner/druggie identity were included as predictor variables. Models for
57
accuracy used only the subsample of participants who had data pertaining to the accuracy
measure. For example, when examining underreporting of smoking, only those participants who
identified at least one of their alters as a nonsmoker were included in analyses. All other analyses
used the whole dataset.
Next, moderation by loneliness and depressive affect was assessed. An interaction term
was created for combinations of loneliness and depressive affect with having at least one alter
who smokes/drinks, stoner/druggie identity, and out-degree centrality by centering loneliness and
depressive affect and multiplying the variables. Separate mixed-effects models were run for each
interaction term with all the other variables from the original model for cigarette and alcohol use
outcomes. A significant interaction term would suggest moderation (Whisman & McClelland,
2005), addressing aim 2.4.
Additionally, in order to provide additional assessment for aims 2.3 and 2.4, a dyadic
analysis was completed. The data were reshaped to long, so that mixed-effects logistic regression
models could be run including random intercepts for participant and school. Items assessing use
of alcohol and cigarettes with alter were regressed on control variables, loneliness, depressive
affect, out-degree centrality, stoner/druggie identity, having at least one alter who smokes/drinks,
the participant’s smoking/drinking behavior, and the alter’s smoking/drinking behavior.
Additional models were subsequently run for interaction terms of loneliness and depressive
affect interacted with alter’s smoking/drinking behavior, for a total of four models per substance.
Model 1 includes no interaction terms, Model 2 includes both interaction terms, Model 3
includes the loneliness and alter behavior interaction term, and model 4 includes the depressive
affect and alter behavior interaction term. Multiple models were run for the interaction terms due
58
to concerns of potential multi-collinearity. Stata and R software were used for data analysis (R
Core Team, 2014; Stata/IC 14.1, 2016).
Results
Table 7 includes sample characteristics for the imputed and original data. The sample was
predominately Hispanic, earned mostly C’s or higher in school, and was split evenly between the
genders. About a third of participants had ever used cigarettes and 58% had ever consumed
alcohol. Additionally, 11% of the participants self-identified as a stoner/druggie.
Table 8 includes the regressions for accuracy in perception of friend’s behavior. Models
for proportion of users correct for drinking, over-reporting for smoking, and over-reporting for
drinking did not have statistically significant omnibus Wald Χ
2
tests (p > .05), indicating that the
overall model was not statistically significant and no parameters were found to be associated
with these outcomes. Having higher out-degree was associated with less accurate reports of
alters’ smoking and drinking behavior using the accuracy measure proportion correct. Having a
higher out-degree was associated with higher odds of underreporting for smoking and drinking
and marginally associated with higher odds of correctly reporting smoking behavior of those
alters who smoke. Self-identifying as a stoner/druggie was associated with higher odds of
correctly reporting smoking behavior of those alters who smoke, and marginally significant
higher odds of underreporting, higher odds of correctly reporting drinking behavior of those
alters who drink, and lower odds of correctly reporting drinking behavior of at least one alter.
Neither loneliness nor depressive affect were associated with any of the accuracy measures.
Table 9 includes regression coefficients and odds ratios for the proportions of the
network that smokes and drinks and participant’s own drinking and smoking behavior.
Loneliness, out-degree, and self-identifying as a stoner/druggie were all associated with higher
59
odds of having at least one friend who smokes and having at least one friend who drinks once a
month or more. Depressive affect, self-identifying as a stoner/druggie, and having at least one
friend who smokes were associated with higher odds of having ever smoked. Self-identifying as
a stoner/druggie and having at least one friend who drinks once a month or more were positively
associated with drinking behavior.
Table 10 includes interaction coefficients testing for interactions of loneliness and
depression with having at least one friend who smokes, having at least one friend who drinks
once a month or more, stoner/druggie identity, and out-degree. These interaction terms were
included as predictors of smoking and drinking behavior one at a time in models including all
predictor variables in Table 9. No interactions were found to be significant. The interaction term
between depression and having at least one friend who drinks once a month or more was found
to be marginally associated with drinking behavior.
Table 11 presents results from analyses examining the participant—alter dyads’ drinking
and smoking behavior. From Model 1, which does not include interaction terms, participant
identifying as a stoner/druggie, having at least one smoker as a friend, participant’s own smoking
behavior, and alter’s smoking behavior were all associated with higher odds of the participant
and alter having smoked together. Loneliness was marginally associated with lower odds of the
participant and alter having smoked together. Having at least one friend who drinks once a
month or more, participant’s own drinking behavior, and alter’s drinking behavior were
associated with higher odds of the participant and alter having used alcohol together. Loneliness
and out-degree were both marginally associated with lower odds of have used alcohol together.
The interaction terms for loneliness and alter’s smoking behavior and depression and
alter’s smoking behavior were significant predictors of the dyad’s smoking behavior when
60
entered into models separately (see Models 3 and 4) but did not retain significance when entered
together. The interaction term for loneliness and alter’s drinking behavior was significantly
associated with the participant and alter having consumed alcohol together when entered into a
model without other interaction terms and when another interaction term was included. The
interaction term between depressive affect and alter’s drinking behavior did not reach
significance.
Discussion
Overall, the hypotheses for this study were not supported. Hypothesis 2.1 stated that
higher loneliness and higher social connection would be associated with greater accuracy of
perception of peer’s behavior. Loneliness was not associated with any of the accuracy measures.
Those people who scored high on the loneliness questionnaire may be chronically lonely. While
literature suggests that socially excluded people have greater recall for social behavior within a
laboratory study (Gardner et al., 2000), those adolescents who are chronically lonely may not
experience biases in their recall. Out-degree was associated with decreased accuracy in smoking
report for all alters, marginally significant increased accuracy when accuracy was examined
within the subsample of smokers and decreased accuracy when accuracy was examined within
the subsample of alters the participant perceives to be nonsmokers. This same pattern was
observed for alcohol use. Overall, out-degree was associated with less accurate reports of
friends’ behavior. While this is contrary to the hypothesis, potentially those participants who
listed more friends on the social network questionnaire included a greater number of friends that
they were less close to even though the questionnaire asks participants to list their best friends.
Adolescents may differ in their interpretation of what best friends means, and those adolescents
who listed more friends may have included some friends that they were not as close to.
61
Additionally, identifying as a stoner/druggie was a significant predictor accuracy of reporting
behavior in the subsample of alters who are smokers. Research suggests that adolescents
perceive their peers to be more similar to themselves in substance use behavior than they are in
reality (Henry, Kobus, & Schoeny, 2011), so this finding is congruent with the research in that
when only looking at the subsample of smoking alters, those participants that identified as
substance users were more likely to report that their alters smoked, that is, they identified that
their peers were similar to them.
We explored that association between loneliness, depression, out-degree, and
stoner/druggie identity with having at least one smoker/drinker as a friend, as aim 2.2. Our only
hypothesis was that those adolescents who self-identified as stoners/druggies would have higher
odds of having a friend who smokes/drinks, which was supported. Those adolescents with higher
out-degree were also more likely to have a friend who smokes and to have a friend who drinks at
least once a month, a finding which suggests that those people who consider more people to be
their friends have higher odds of being friends with someone who smokes or who uses alcohol. It
was found that loneliness was associated with higher odds of having at least one friend who
smokes and at least one friend who drinks at least once a month. This was an unexpected finding,
and suggests that lonely adolescents experience higher exposure to peer substance use in
comparison to non-lonely adolescents. It is unclear, however, the directionality of the
association, that is, do lonely adolescents become lonely due to having friends who use cigarettes
and alcohol or do lonely adolescents select peers who use alcohol and cigarettes as friends?
Hypotheses for aim 2.3 were supported. Analyses supported that self-identifying as a
stoner/druggie and having at least one friend who smokes/drinks is positively associated with
smoking and drinking behavior. Additionally, depressive affect was positively associated with
62
cigarette use, a finding supported by past studies which have found that depression and cigarette
use are associated (Luger, Suls, & Vander Weg, 2014). Neither loneliness nor out-degree was
associated with smoking. Studies of loneliness and cigarette use have had conflicting findings,
with some studies reporting an positive association while others reporting no association (Dyal &
Valente, 2015). Some studies suggest that loneliness and cigarette use may have a small
association, which this study may have been underpowered to find (DeWall & Pond, 2011). Past
studies have also reported null results for the association between out-degree and smoking
(Pearson et al., 2006). This study uses egocentric network data, and potentially perceiving many
or few people to be one’s best friend, regardless of any of the friends’ demographics, may not be
associated with smoking. Studies which examine out-degree within different contexts, such as
school, neighborhood, or extracurricular activities, may be more useful for studying cigarette
use.
Hypotheses for aim 2.4 were not supported. Loneliness and depression do not serve as
moderators for the effects of peer influence, stoner/druggie identity, or out-degree. None of the
interaction terms were significant. The dyadic analyses were conducted to explore this finding
further. The previously described study suggested that socially excluded people may be more
likely to try an illegal drug when their alters approved of drug use only when the alters would be
aware of the participant’s drug use (Mead et al., 2011). We examined if the alter’s drug use
interacted with loneliness and depression in a models with the outcomes of the alter and
participant having used cigarettes or alcohol together, a scenario where the alter would be aware
of the participant’s drug use. These models controlled for the participant’s cigarette use and
alcohol use. Analyses suggest that loneliness reduces the association between alter cigarette and
alcohol use and the participant having used these substances with the alter. This suggests that
63
loneliness attenuates the effect of peer influence on using cigarettes and alcohol with one’s peers.
Depression had a similar moderational effect for cigarettes use, but not alcohol use. This finding
does not support past laboratory-based research which suggests that loneliness should increase
compliance with peer norms and group affiliation, particularly when the participant’s behavior is
observable to their alters.
Our analyses for aim 2.4 do contradict each other--- the analyses examining participant’s
cigarette and alcohol use found no interaction effects, while the analyses examining participant’s
use of alcohol and cigarettes with an alter did find significant interaction effects. It may be that
loneliness only moderates cigarette and alcohol use when it is an activity shared with a friend.
However, since this is a cross-sectional study, and we cannot determine directionality of
associations, it may be the case that those adolescents who are not congruent with their friends
on cigarette and alcohol use may be lonelier. It may be worthwhile to examine discordance on
alcohol and cigarette use as predictors of loneliness, and furthermore, to examine discordance on
extracurricular activity participation as a predictor of loneliness. The questionnaire which
assessed loneliness included items “How often do you feel that you have a lot in common the
with people around you?” and “How often do you feel left out?,” in addition to items assessing
feelings of isolation and social support availability. Therefore, the questionnaire may have
captured how similar the participant perceived themselves to be to their peers and if they join in
with their peers for social activities. Those adolescents who are discordant with their friends on
substance use may have more feelings of loneliness or perceive themselves as having less in
common with their alters. Further research is needed to understand the multi-dimensional aspects
of loneliness.
64
This study does have limitations. The cross-sectional design of the study does not allow
longitudinal modeling and we cannot determine the directionality of the associations examined in
this study. The sample was drawn from a primarily Hispanic/Latino population and is likely not
generalizable to other demographics, however, it does allow for generalization to a growing, less
studied population. The population examined had low smoking prevalence, reducing the power
available to find results for the outcome of smoking. However, the low smoking prevalence is
due to successful public health campaigns to reduce smoking rates among adolescents.
The loneliness measure used for this study is uni-dimensional. Potentially a loneliness
measure which assessed social and emotional loneliness separately or loneliness experienced in
specific relationships such as family-related loneliness, romantic-related loneliness, and
friendship-related loneliness would have provided clearer findings. Furthermore, a measure
which separated social support, negative affect associated with perceived social isolation, and
perceiving oneself as similar to peers may have been useful in separating these related
constructs.
Alters with missing data were not included in the analyses, for both the accuracy analyses
and the dyadic analyses. This may have introduced bias, particularly when assessing accuracy,
because the participants may have left survey items blank for questions assessing friend behavior
which they did not know the answers to. We may have then, estimated higher accuracy of
participants’ reports of friends’ behavior than is occurring in reality. In addition, we were unable
to obtain self-report data for many of the participants’ alters. The alters that we were unable to
obtain self-report data from did differ from those that we could obtain self-report data for. It is
unknown how this may have affected the results.
65
When examining the participant—friend dyads, the odds ratios for the participant’s
smoking behavior and the alter’s smoking behavior were very high. These variables were
dichotomous, and likely were highly related to the outcome. If the participant has smoked with
their friend, then we would also expect that they reported that they themselves have smoked and
that they report that their friend has smoked. The same problem does not necessarily arise with
alcohol use, because participant’s alcohol use is modeled here as a continuous variable and
alter’s alcohol use is assessed as using alcohol at least once a month, and the participant may
report having used alcohol with their friend even though they do not report that their friend uses
alcohol at least once a month. We considered it important to control for participants’ smoking
behavior and their friends’ smoking behavior when examining the participants’ and friends’
behavior of smoking together, and do retain these variables in the analysis.
This study explored how loneliness may affect smoking and drinking behavior among
adolescents by assessing if loneliness may affect accuracy in perception of peer behavior,
exposure to substance using peers, directly affect substance use behavior, and serve as a
moderator for peer influence’s effect on substance use behaviors. We found that loneliness did
not affect accuracy of perception of peer’s behavior or directly affect substance use behavior, but
that it was associated with higher odds of having a friend who uses alcohol and cigarettes.
Lonely adolescents may have greater exposure to peers who are substance users. Future research
may focus on exploring if lonely adolescents choose to become friends with substance using
peers or if being friends with a peer who smokes or drinks alcohol increases an adolescent’s
loneliness. If lonely adolescents select friends who use substances, prevention efforts may be
aimed at supporting lonely adolescents to become friends with non-substance using peers to
prevent undue exposure to substance using peers. If having friends who smoke or drink increases
66
an adolescent’s loneliness, then programming may be included in substance use prevention
programs to provide coping skills to reduce psychological distress experienced by adolescents
with substance using friends.
67
Table 7: Demographics for imputed and original sample
Pre-imputation (n
differs by item)
Post-imputation (n = 1279)
n (%) n (%)
Female 654 (53) 682 (53)
Hispanic (of any race) 915 (75) 937 (73)
Asian 321 (26) 349 (27)
Age 18 747 (61) 787 (62)
Academic performance
Mostly A’s 116 (10) 116 (9)
Mostly A’s and B’s 337 (28) 353 (28)
Mostly B’s 97 (8) 102 (8)
Mostly B’s and C’s 411 (34) 457 (36)
Mostly C’s 107 (9) 116 (9)
Mostly C’s and D’s 104 (9) 107 (8)
Mostly D’s 9 (1) 9 (1)
Mostly D’s and F’s 15 (1) 15 (1)
Mostly F’s 4 (<1) 4 (<1)
People per room (PPR)* 1.79 (.91) 1.78 (.89)
Alcohol Use
Nonsusceptible 339 (29) 358 (28)
Susceptible 152 (13) 173 (14)
Ever Drink 105 (9) 115 (9)
Past Year 170 (14) 189 (15)
Past month 201 (17) 218 (17)
Binge Drink 212 (18) 226 (18)
Cigarette Use 410 (34) 448 (35)
1+ Alter who smokes 610 (48) -
1+ Alters who drinks 601 (47) -
Out-degree* 5.11 (1.80) -
Stoner/Druggie identity 136 (11) -
Loneliness 2.27 (.56) 2.27 (.53)
Depressive Affect 1.68 (.86) 1.67 (.85)
Accuracy
Cigarette use Proportion Correct 556 (54) -
Proportion of Users Correct 255 (45) -
Underreporting 363 (41) -
Over-reporting 125 (39) -
Alcohol Use Proportion Correct 589 (58) -
Proportion of Users Correct 243 (49) -
Underreporting 291 (34) -
68
Over-reporting 164 (49) -
Note. Sample size for accuracy measure percentage is the same as sample size for the regressions
in Table 8. Accuracy, identity, out-degree, and having a smoking/drinking alter were not
imputed. * indicates continuous variable, mean and standard deviation are reported.
69
Table 8: Regression results examining accuracy of participant report of friend's behavior
Proportion Correct
Proportion of Users
Correct
Underreporting Over-reporting
Cig. Use n = 1027 n = 569 n = 894 n = 320
OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Hispanic 0.55** 0.39 0.79 1.69* 1.00 2.85 2.00*** 1.36 2.95 0.57 0.28 1.15
PPR 0.93 0.8 1.09 0.79* 0.64 0.97 1.03 0.87 1.22 1.2 0.90 1.60
Sch. Perf. 1.02 0.93 1.11 0.96 0.85 1.08 0.99 0.89 1.09 0.92 0.79 1.07
Female 0.96 0.73 1.25 0.67* 0.46 0.96 0.89 0.67 1.19 1.55+ 0.94 2.56
Age 18+ 1.01 0.78 1.32 1.36+ 0.95 1.95 0.94 0.71 1.25 0.97 0.59 1.60
Loneliness 1.12 0.86 1.47 1.04 0.73 1.48 0.83 0.62 1.11 1.17 0.72 1.88
Dep. Affect 1.00 0.85 1.18 1.06 0.85 1.32 1.04 0.87 1.25 0.9 0.68 1.20
Out-Degree 0.81*** 0.75 0.88 1.10+ 0.98 1.24 1.15** 1.05 1.26 1.09 0.93 1.27
Stoner/Druggie 0.72 0.47 1.11 2.84*** 1.7 4.74 1.65+ 0.99 2.75 0.94 0.51 1.71
Alc. Use n = 1019 n = 478 n = 852 n = 337
OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Hispanic 0.34*** 0.24 0.48 1.42 0.85 2.37 3.77*** 2.51 5.66 1.06 0.57 1.96
PPR 1.19* 1.01 1.39 0.94 0.75 1.19 0.80* 0.66 0.96 0.97 0.75 1.26
Sch. Perf. 1.03 0.95 1.13 1.01 0.89 1.15 1.00 0.90 1.11 0.94 0.81 1.08
Female 1.14 0.87 1.50 0.79 0.54 1.15 0.73* 0.54 1.00 1.48+ 0.94 2.35
Age 18+ 1.00 0.76 1.30 1.00 0.68 1.46 0.90 0.66 1.21 1.05 0.67 1.65
Loneliness 1.12 0.86 1.46 0.87 0.61 1.26 0.79 0.58 1.06 0.81 0.52 1.26
Dep. Affect 0.93 0.79 1.10 1.14 0.9 1.43 1.10 0.92 1.33 0.94 0.71 1.25
Out-Degree 0.85*** 0.78 0.92 1.16* 1.02 1.31 1.15** 1.04 1.26 1.04 0.90 1.21
Stoner/Druggie 0.68+ 0.45 1.05 1.63+ 1.00 2.68 1.00 0.59 1.69 1.12 0.65 1.95
Note. + indicates p < .10, * p <.05, **p <.01, and ***p <.001. Odds ratios presented are unstandardized.
70
Table 9: Regressions for proportion of smokers/drinkers in network and participant’s drinking/smoking behavior
1+ alters who smoke
Cig. Use
1+ alters who drink
Alc. Use
OR 95% CI
OR 95% CI
OR 95% CI
b SE
Hispanic 2.91*** 2.11 4.02
1.70** 1.19 2.44
2.21*** 1.61 3.05
.70*** .11
PPR 0.88+ 0.77 1.02
1.07 0.92 1.25
1.00 0.87 1.15
.03 .05
Sch. Perf. 0.89** 0.82 0.96
0.85*** 0.78 0.93
0.91* 0.84 0.98
-.06+ .03
Female 0.82 0.64 1.05
0.70** 0.53 0.91
0.99 0.78 1.27
.14 .09
Age 18+ 1.32* 1.03 1.69
1.38* 1.05 1.82
1.19 0.93 1.51
.04 .09
Loneliness 1.38* 1.07 1.77
0.94 0.71 1.22
1.31* 1.03 1.68
.00 .09
Dep. Affect 0.91 0.78 1.06
1.19* 1.01 1.41
1.05 0.90 1.23
.03 .06
Out-Degree 1.23*** 1.14 1.32
0.95 0.88 1.02
1.20*** 1.12 1.29
.01 .03
Stoner/Druggie 4.32*** 2.72 6.86
2.51*** 1.64 3.83
4.41*** 2.80 6.96
.95*** .15
1+ alter w/ beh.
5.55*** 4.19 7.37
1.52*** .09
Note. + indicates p < .10, * p <.05, **p <.01, and ***p <.001. Odds ratios and bs presented are unstandardized.
71
Table 10: Interaction results for loneliness and depression predicting smoking and alcohol
use
Cig. Use Alc. Use
OR 95% CI b se
Loneliness X Alter 0.67 0.40 1.11 -.10 .17
Dep. Affect X Alter 0.91 0.67 1.23 -.20+ .11
Loneliness X
Stoner/Druggie
0.8 0.38 1.68 -.38 .26
Dep. Affect X
Stoner/Druggie
0.91 0.59 1.42 -.09 .15
Loneliness X Out-
Degree
0.97 0.86 1.11 -.06 .04
Dep. Affect X Out-
Degree
0.96 0.88 1.05 -.05 .03
Note. + indicates p < .10, * p <.05, **p <.01, and ***p <.001. Odds ratios and bs presented are
unstandardized. X Alter term is the interaction with having at least one friend who smokes/drinks
72
Table 11: Regression results for dyadic analyses
Model 1 Model 2 Model 3 Model 4
Smoked cig
w/ alter
OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Hispanic 0.89 0.39 2.02 0.85 0.37 1.95 0.84 0.37 1.94 0.87 0.38 2.00
PPR 1.00 0.73 1.38 0.99 0.72 1.37 0.99 0.72 1.36 1.00 0.73 1.38
Sch. Perf. 0.89 0.74 1.07 0.89 0.74 1.07 0.89 0.74 1.07 0.89 0.74 1.07
Female 0.86 0.49 1.51 0.87 0.50 1.54 0.88 0.50 1.55 0.86 0.49 1.52
Age 18+ 1.50 0.85 2.67 1.49 0.83 2.66 1.50 0.84 2.68 1.49 0.84 2.64
Loneliness 0.64+ 0.38 1.07 1.15 0.45 2.94 1.48 0.60 3.66 0.65 0.39 1.09
Dep. Affect 0.98 0.72 1.35 1.44 0.87 2.39 0.99 0.72 1.35 1.61+ 0.99 2.61
Out-Degree 1.01 0.85 1.20 1.01 0.85 1.20 1.01 0.85 1.20 1.01 0.85 1.20
Stoner/Druggie 5.88*** 3.08 11.20 6.08*** 3.17 11.67 6.04*** 3.14 11.60 6.00*** 3.15 11.45
1+ alters who
smoke
6.64** 1.74 25.36 6.15** 1.59 23.81 6.09** 1.58 23.48 6.51** 1.69 25.10
Pt. smokes 74.46*** 34.58 160.31 78.16*** 35.87 170.31 77.81*** 35.71 169.56 76.12*** 35.19 164.64
Alter smokes 96.31*** 53.98 171.83 112.59*** 60.46 209.66 103.05*** 56.48 188.00 110.55*** 59.87 204.1
Lone X Alter 0.50 0.19 1.28 0.36* 0.15 0.88
Dep. Affect X
Alter
0.62+ 0.37 1.02 0.54* 0.34 0.86
Used alc. w/
alter
OR 95% CI OR 95% CI OR 95% CI OR 95% CI
Hispanic 0.98 0.56 1.71 0.98 0.56 1.70 0.98 0.56 1.71 0.98 0.57 1.71
PPR 0.75* 0.60 0.95 0.75* 0.60 0.95 0.75* 0.60 0.95 0.75* 0.60 0.95
Sch. Perf. 1.00 0.87 1.14 0.99 0.87 1.14 1.00 0.87 1.14 1.00 0.87 1.14
Female 0.91 0.61 1.36 0.92 0.61 1.39 0.92 0.61 1.38 0.91 0.61 1.37
Age 18+ 1.13 0.76 1.68 1.13 0.76 1.69 1.13 0.75 1.68 1.13 0.76 1.69
Loneliness 0.71+ 0.47 1.05 0.99 0.62 1.58 0.91 0.58 1.44 0.70+ 0.47 1.05
Dep. Affect 1.00 0.79 1.27 0.89 0.68 1.18 1.00 0.78 1.27 0.97 0.74 1.28
73
Out-Degree 0.90+ 0.79 1.02 0.90+ 0.79 1.02 0.90+ 0.79 1.02 0.90+ 0.79 1.01
Stoner/Druggie 1.37 0.80 2.36 1.37 0.79 2.35 1.37 0.80 2.37 1.37 0.80 2.35
1+ alters who
drink
5.16*** 3.13 8.50 4.97*** 3.01 8.19 4.97*** 3.01 8.20 5.17*** 3.14 8.52
Pt. Alc. Use 3.44*** 2.95 4.00 3.46*** 2.97 4.03 3.45*** 2.96 4.02 3.44*** 2.95 4.00
Alter Alc. Use 9.81*** 7.47 12.87 10.07*** 7.64 13.27 10.21*** 7.74 13.46 9.75*** 7.41 12.81
Lone X Alter 0.48** 0.28 0.80 0.57* 0.36 0.92
Dep. Affect X
Alter
1.28 0.94 1.75 1.06 0.80 1.40
Note. + indicates p < .10, * p <.05, **p <.01, and ***p <.001. For cigarette use, there were 6386 observations nested in 1248
participants. For alcohol use, there were 6400 observations nested in 1247 participants. Odds ratios presented are unstandardized. X
Alter term is the interaction term for alter substance use.
74
CHAPTER 4: THE COEVOLUTION OF ADOLESCENT FRIENDSHIP NETWORKS,
LONELINESS, AND CIGARETTE SMOKING
Abstract
Introduction: Research suggests that loneliness, smoking, and friendship networks are associated
with one another. This study analyzes the associations among these constructs and considers a
potential interaction between loneliness and peer influence from a friendship network as a
predictor of smoking.
Methods: Data from 772 adolescents enrolled at 4 high schools in Southern California, USA,
were collected at two time points using in-class surveys including questions on friendship
networks, loneliness, parental communication, acculturation, cigarette use, and demographics. A
stochastic actor-oriented model was created with forward model selection using RSiena
including the friendship network, loneliness, and smoking as dependent variables.
Results: Selection was found for smoking for one school. Influence and selection were not found
for loneliness or smoking for any other school. Ego effects were found for smoking for two
schools and the alter effect for loneliness was significant for one school. The interaction terms
were not statistically significant.
Discussion: This study did not find influence and selection processes for smoking or loneliness.
Loneliness and peer influence did not interact in the prediction of smoking behavior.
Implications for loneliness measurement and chronicity of loneliness are discussed.
Introduction
Loneliness, social networks, and smoking may have influence on and be influenced by
each other. While the previous two chapters examined the associations among these variables
cross-sectionally, with a focus on smoking as the outcome, chapter 4 takes a longitudinal
75
approach and assesses the bi-directional associations among loneliness, a friendship network,
and smoking using a stochastic actor-oriented model (SAOM). The SAOM approach allows all
three of these variables to be modeled simultaneously as dependent variables and as predictors of
one another. This final study additionally explores parental communication and linguistic
acculturation as predictors of loneliness, friendship networks, and smoking.
Parental communication is an important predictor of smoking behavior in combination
with positive parenting practices (Cohen et al., 1994). Adolescents who experience positive
communication with their parents are less likely to smoke (Cohen et al., 1994). Importantly,
parental communication has been identified as a stronger predictor of smoking among
Hispanic/Latino adolescents in comparison to white adolescents, suggesting it may be a
particularly relevant predictor for the study population (Shakib et al., 2003).
The association between loneliness and parental communication in adolescents is unclear.
One study found that young adults in college who report being satisfied with their family
communication experience are less likely to be lonely (Segrin et al., 2012). Another study found
that positive mother--adolescent communication was a protective factor for loneliness while
father--adolescent communication had no effect on loneliness (Brage, Meredith, & Woodward,
1993). However, other researchers have reported that neighborhood connection and peer
friendship are more important in explaining older youth’s loneliness in comparison to parental
relationships (Chipuer, 2001). Normative expectations for relationships change greatly
throughout adolescence, which may affect how and if parental communication is associated with
loneliness. The ratio of time spent with parents in comparison to peers changes considerably
during adolescence (Laursen & Hartl, 2013). From childhood to adolescence, the amount of time
spent with family members decreases by about 50% and the amount of time spent with friends
76
increases for both boys and girls (Larson & Richards, 1991). Boys primarily spend an increased
amount of time alone while girls spend time with friends and alone (Larson & Richards, 1991).
These relationship changes and new norms may affect the association between loneliness and
parental communication; more research is needed to clarify the association. Despite adolescence
being a period of increasing independence in preparation for adulthood, lack of positive parent--
adolescent communication may induce loneliness.
Communication style learned through family of origin affects how adolescents interact
within other social relationships (Ledbetter, 2009). Adolescents with poor parental
communication experiences may not develop necessary skills to connect with their peers and
may experience difficulty forming and maintaining close friendships (Freitag, Belsky,
Grossmann, Grossmann, & Scheuerer-Englisch, 1996; Ledbetter, 2009). Poor parental
communication may therefore be associated with fewer close friendships, less connection to the
larger social network, and isolate status. Parental communication may therefore affect smoking
behavior through its effects on social networks.
Acculturation is the process of adopting the customs, identities, and general way of living
of a group of people which one is exposed to (Schwartz, Unger, Zamboanga, & Szapocznik,
2010). Much variation in acculturation is attributed to linguistic acculturation, the use of the
language of a group one is exposed to (Epstein, Botvin, Dusenbury, Diaz, & Kerner, 1996).
Acculturation to the US culture is an important predictor of health outcomes for Hispanic
adolescents in the US (Lara, Gamboa, Kahramanian, Morales, & Hayes Bautista, 2005). A
review of acculturation and smoking in Hispanics adults found that higher acculturation to the
dominant culture is associated with increased smoking for females and is not a predictor of
smoking in males (Bethel & Schenker, 2005). Furthermore, higher use of English is associated
77
with increased smoking behavior in Hispanic adolescents, particularly within females (Epstein et
al., 1998; Kaplan, Nápoles-Springer, Stewart, & Pérez-Stable, 2001; Unger et al., 2000).
Linguistic acculturation has been found to be associated with loneliness. Hispanic
adolescents who speak only English with friends are less likely to experience an increase in
loneliness over time (Benner, 2011). Low majority language proficiency was the best predictor
of loneliness in a study of Portuguese adolescents living in Switzerland (Neto & Barros, 2000).
This suggests that adolescents more acculturated to the dominant culture are less likely to
become lonely.
Culture is associated with the experience of loneliness and with one’s perception of the
causes of their loneliness (Heinrich & Gullone, 2006; Lykes & Kemmelmeier, 2014; Rokach,
2007; Rokach & Neto, 2005). The North American culture has been characterized as promoting
loneliness with its emphasis on individualization (Rokach, 2007; Rokach & Neto, 2005; van
Staden & Coetzee, 2010). Findings from European societies suggest that in collectivistic
societies loneliness is associated with reduced family interaction while in individualistic societies
loneliness is more associated with reduced interaction with non-family connections such as
friends (Lykes & Kemmelmeier, 2014). Loneliness may arise due to difficulties from cultural
barrier in being understood and understanding another (van Staden & Coetzee, 2010). Feelings
of personal control over loneliness may affect how loneliness is experienced and how it is
sustained, which may differ by cultural emphasis on individualization (Rokach, 2007). Cultural
norms may shape people’s views on what an ideal social life may look like and/or what their
perceived social needs are, contributing to their experience of loneliness.
Social networks may be associated with acculturation as well, either due to an adolescent
choosing friends with similar acculturation statuses (selection) or experiencing a change in
78
acculturation due to their friends (influence). Research suggests that the association between
acculturation and smoking is mediated by the number of self-reported friends who smoke/ social
peer influence (R. Myers et al., 2009; Unger et al., 2000). Increased acculturation may put
adolescents at risk for smoking through social network changes.
Parental communication may be associated with acculturation, as adolescents who
become more acculturated to the dominant culture may experience difficulty connecting and
communicating with their parents. One study found that integrated Hispanic/Latino adolescents--
-those acculturated to both the American and Hispanic/Latino way of life---had significantly
higher parental communication in comparison to adolescents acculturated to the American
culture (Sullivan et al., 2007). Parental communication and acculturation may explain some of
the same variance in smoking. Familism, a Hispanic cultural value which emphasizes connection
to family (Soto et al., 2011) could be lost through the acculturation process resulting in reduced
quality of parental communication. A study of Latina adolescents found that adolescents lower in
familism were more likely to have ever smoked (Kaplan et al., 2001).
Including parental communication and linguistic acculturation in the analyses will further
the policy and intervention implications of this study by contributing additional information in
which to understand the associations among loneliness, social networks, and smoking as well as
to identify additional constructs which may be addressed in prevention programs. Modeling
social network structure and factors related to communication and identity with loneliness and
smoking will contribute to our knowledge of which adolescents are at most risk to smoke and
how we can intervene to prevent them from smoking and help them quit smoking if they have
already started.
79
Methods
This study uses a stochastic actor-oriented model (SAOM) for network dynamics in the
RSiena (Simulation Investigation for Empirical Network Analysis) package in R (Ripley,
Boitmanis, & Snijders, 2013). This SAOM was developed specifically for analyzing longitudinal
network data and is an improvement to past statistical models which assumed static network
structure, an often unrealistic assumption (Snijders, van de Bunt, & Steglich, 2010; Valente,
2010). This modeling approach also allows for modeling the co-evolution of networks and
behavior by simultaneously simulating network change and behavior change (Snijders et al.,
2010).
RSiena uses a Markov Chain process to model the change from one network state to the
next by changing ties and actors’ behavior in a series of microsteps, using the network state at
first network observation as the starting state (Snijders, Steglich, & Schweinberger, 2007;
Snijders et al., 2010). At each microstep, an actor is chosen either with equal probability or with
probabilities dependent on attributes/network position (Snijders et al., 2010). The selected actor
will then select to change a tie or behavior based on some probability, and will then make or not
make this change with probability based on a function describing the likelihood of an actor to
make the selected change calculated using the current network structure and actor attributes
(Snijders et al., 2010). Behavior change can only happen with a one unit increase or decrease
(Snijders et al., 2007). These probabilities are defined by the rate and objective functions. There
are separate rate functions for network change and behavior change which are used to calculate
when an actor changes a tie and changes a behavior (Snijders et al., 2007). The probability of if
the selected change is made or not is determined by the objective functions, also referred to as
the evaluation functions (Snijders et al., 2007). Creation and endowment functions can also be
80
specified, which include parameters for creating ties and increasing behaviors (creation) and
terminating ties and decreasing behaviors (endowment) (Ripley, Snijders, Boda, Vörös, &
Preciado, 2016). A separate evaluation function is included for each dependent network or
behavior variable in the simulation.
The parameters in these functions are updated throughout the modeling process by trying
to estimate parameters that simulate a network which best matches the end network observation.
This is done by simulating networks based off of currently estimated parameter values and the
initially observed network, calculating the difference between the simulated network and
observed network at the next time point, and updating the parameter values such that the next
simulated network using the updated parameter values is expected to be slightly closer to that
which was observed (Ripley et al., 2016). In terms of the data used in this study (described
below) the initial network observation is the 11
th
grade data and the final network observation is
the 12
th
grade data. The algorithm will use the 11
th
grade data as a starting point to simulate
networks at 12
th
grade, based on the rate and evaluation functions. Each time a simulated
network is created, it is compared to the true 12
th
grade data, and the parameters of the rate and
evaluation functions are updated. After parameter estimation has taken place, standard errors are
estimated. From this, statistical significance of the parameters can be determined.
This study uses longitudinal data from waves three and four of the Social Network Study,
collected in the 11
th
and 12
th
grade years of the participants. The Social Networks Study is a
longitudinal cohort study of adolescents’ social networks and risk behavior encompassing four
waves of data collected from students attending five schools in Southern California, USA.
Additional details of the study methodology are available elsewhere (Valente, Fujimoto, Unger,
et al., 2013). One school did not participate in wave three of the study, therefore we are only
81
using data from four schools. A total of 772 participants completed the survey in both waves. For
the SAOM, at least 20 participants need to be sampled from each school, with no more than a
few hundred participants in each school sample, so the data meets the requirements for sample
size (Snijders et al., 2010). Data from all 772 participants will be included in the model, even if
they are missing data on some variables. RSiena includes a data imputation feature which allows
for the use of data from all participants in the dataset, even if they have missing data on some
variables.
Data were collected using a paper survey in school during normal class periods. The
participants were provided with a photo book which included photos with ID numbers of all
students attending the participants’ school and enrolled in their grade in order to complete the
social network portion of the survey. Trained data collectors verbally provided instructions and
ensured participants that the survey responses would remain confidential. The survey included
questions assessing social networks, school activity participation, affect, family life, and
demographics. All participants provided assent and parental consent prior to participating in the
study. All study procedures and materials were approved by the Institutional Review Board of
the University of Southern California prior to data collection.
The friendship network data was collected at both waves with the questions “Please think
of your seven BEST FRIENDS in XX grade. […]” and “Are there other people in the XX grade
who you consider a close friend? […]”, with space to list an additional 12 friends, for a total of
19 possible nominations. The participants wrote down the photo ID numbers for all friends they
nominated and provided additional information for their first seven nominations. The participants
were also asked to indicate their photo ID number on their survey. In order for the SAOM to be
estimated, enough change must have occurred within the network between the waves as assessed
82
by the Jaccard index (Snijders et al., 2010). The Jaccard index for the friendship network ranged
from .236 to .284. This is adequate for modeling the network evolution (Ripley et al., 2016).
Loneliness was assessed at both waves using a one-item measure of loneliness from the
Center for Epidemiological Studies Depression scale (CES-D: Radloff, 1977). The CES-D is a
20-item measurement of depressive symptomatology which includes the item “I felt lonely”,
which is used in this study to assess loneliness. Participants were asked to “Think about how you
felt during the past 7 days. For each statement, please indicate how often you felt this way during
the past 7 days,” and provided with the response options (1) 1 day or less, (2) 2 - 3 days, (3) 4 - 5
days, and (4) 6 – 7 days. This item, while not a comprehensive measure of loneliness, does have
face validity and has been used in past studies examining the associations of loneliness and
networks (Cacioppo et al., 2009). Loneliness was assessed in both 11
th
and 12
th
grades.
Cigarette use was assessed with the item “During the past 12 months, how often did you
smoke cigarettes?” with response options ranging from (1) I have never smoked cigarettes to (7)
Everyday or almost everyday. Due to skewness, this variable was dichotomized to (1) Participant
has smoked in past 12 months and (0) Participant has not smoked in the past 12 months.
Cigarette use was assessed in both 11
th
and 12
th
grades.
Parental communication was assessed in 11
th
grade with a 5-item scale adapted from
Cohen et al. (1994). This scale included questions such as “How often do you talk to your
parents about what’s on your mind?” and “How often do you ask your parents for advice?” and
response options ranging from (1) Very Often to (4) Never. The scale was created by taking the
mean of all five items and rounding it to the nearest integer, as RSiena cannot handle variables
with decimals. For this scale, low values indicate good parental communication and high values
indicate poor parental communication.
83
Linguistic acculturation was assessed in 11
th
grade using a 5-item scale which assesses
what language participants speak in their home, with friends, at school, with siblings, and with
parents. Participants given prompts such as, “In your home, do you speak…”, followed by 5
answer choices including (1) Only English, (3) English and another language equally, and (5)
Only another language. On this scale, low numbers indicate use of primarily English in most
social contexts and high numbers indicate primarily the use of another language in most social
contexts. The scale was created by taking the mean of the five items and rounding the mean to
the nearest integer.
Demographic and control variables included in this study were eligibility for free lunch,
ethnicity, gender, age, academic performance, use of cigarettes by siblings, and use of cigarettes
by the two adults the participant spends the most time with. All control variables except for age
were assessed in 11
th
grade. Eligibility for free lunch was coded (1) Eligible and (0) Not Eligible.
Ethnicity was coded as (1) Hispanic of any race and (0) Not Hispanic. Gender was coded as (1)
Female and (0) Male. Age was coded as (1) Age 18 or older at 12
th
grade survey or (0) Under
age 18. Academic performance was assessed on scale from (9) Mostly A’s to (1) Mostly F’s. Use
of cigarettes by siblings was assessed with the question “Do any of your siblings (brother or
sister) smoke cigarettes once a month or more?” Various response options were provided for this
question and were dichotomized as (1) Yes or (0) No. Use of cigarettes by adults you spend time
with was assessed with the question “Think of the two (2) adults that you spend the most time
with. How many of them smoke cigarettes every day or most days?,” with response options (0)
None, (1) One of them, and (2) Two of them.
Multiple SAOM were estimated using RSiena throughout the process of model
specification. The modeling process involved specifying a network evolution model, a network
84
and smoking coevolution model, a network and loneliness coevolution model, and lastly
specifying a network, smoking, and loneliness coevolution model. Forward selection was used to
specify effects necessary for inclusion in the models to create a model which obtained adequate
fit to the data. The friendship network evolution model was estimated first. Effects for network
structure and covariate effects on network evolution were entered in groups using a forward
selection procedure, and score tested to determine if they contributed to the model (Ripley et al.,
2016; Snijders et al., 2007). The endogenous network effects were tested first. Different effects
for transitivity and degree activity were score tested and the effects which were the most
significant for the majority of the schools were entered into the model. Covariate effects were
grouped by covariate--- that is, all effects for gender were entered simultaneously, etc. Order of
effects entered into the model was decided by considering which effects were most important
based on theory and past research. Those effects which the score test suggested were important
for three or more networks, or which the score test was significant for two networks and
marginally significant for another network were retained for the final model.
The same procedure was then used to fit co-evolution models for loneliness and smoking,
separately, using the base model obtained for network evolution. However, as few covariates had
significant effects for the behavior variables, covariates with low p-values for the majority of the
schools were retained in the model, even if they did not reach significance. The effects of the
network on the behavior and of the behavior on the network were additionally included in the
models. Different effects for influence were score tested, and the effect that had the lowest p-
values for each behavior was added to the model. Lastly, smoking, loneliness, and the friendship
network were included in a co-evolution model simultaneously. The effects of loneliness on
smoking and of smoking on loneliness were included in the final model in addition to an
85
interaction between influence from the friendship network and loneliness as a predictor of
smoking. A variety of resources were consulted to determine appropriate model effects (Burk,
Steglich, & Snijders, 2007; Ripley et al., 2016; Snijders, Lomi, & Torló, 2013; Snijders et al.,
2010; Steglich, Snijders, & West, 2006). The parental communication, linguistic acculturation,
and all demographic variables described in this section were assessed as potential predictor
variables in for friendship, loneliness, and smoking. The variables listed in Table 13 were
included in the final model.
Model fit was assessed for each school using goodness-of-fit statistics for in-degree
distribution, out-degree distribution, loneliness behavior distribution, geodesic distribution, triad
census, and constraint distribution. Currently, in-degree, out-degree, and behavior distributions
are recommended to meet adequate fit (Snijders, 2015). The overall maximum convergence ratio
was also assessed to make sure that the model converged. Maximum convergence ratios of .25 or
less indicated convergence (Ripley et al., 2016).
The results are presented by school as well as using a meta-analysis, using RSiena’s
internal meta-analysis function siena08 and reporting results using both the Snijders and
Baerveldt (2003) method and Fisher’s method (Ripley et al., 2016). The sample size may be too
low to use the Snijders and Baerveldt method, but it allows us to estimate mean parameters and
standard deviations for the parameters, whereas Fisher’s method only allows for testing if there
are networks which exist with significant positive or negative parameters.
Results
Demographic data is presented by school in Table 12. Additional network descriptives
are included in Table 14. Aim 3.1 includes assessing the rate of loneliness in the sample and
predictors of loneliness. Roughly half, or 45.3% of the sample indicated they experienced
86
consistent low levels of loneliness at both waves, they felt lonely one day or less in the past
seven days, while 4.7% indicated consistent high levels of loneliness---they indicated feeling
lonely six to seven days out of the past seven days at both waves. The loneliness measure was
dichotomized by making a low loneliness category for those participants who felt lonely on three
or less days out of the last seven days and a high loneliness category for participants who felt
lonely four or more days out of the last seven days. The dichotomized measure indicated 70.5%
of participants experienced consistent low loneliness, 10.2% of participants increased from low
to high loneliness between the waves, 11.4% of participants decreased from high to low
loneliness, and 8% of participants experienced consistent high loneliness.
Aims 3.1 and 3.2 were addressed with the coevolution model for loneliness, smoking,
and friendship network. Table 13 is a list of all effects which were included in this model. Note
that the interaction term for assessing aim 3.3 was not included in this model. Table 14 details
information about the model convergence and fit. The overall maximum convergence ratios were
all less than or equal to .16, indicating that convergence was obtained for all four schools.
Goodness-of-fit indices indicated adequate fit, as evidenced by out-degree distribution, in-degree
distribution, and loneliness distribution tests having p-values > .05. The out-degree distribution
for school three did not obtain adequate fit. The triad census and constraint distribution did not
reach adequate fit, overall. Different parameterizations of endogenous network characteristics
and out-degree activity were tried to improve model fit, but we were unable to improve the fit.
Results by school are presented in Table 15. The ego effect of smoking was negatively
associated with friendship tie formation for school one and positively associated with friendship
tie formation for school two. For school two, the selection effect for smoking—having the same
value for smoking status was significant. For school one, there was a significant and positive
87
effect for alter loneliness on friendship tie formation. No influence effect was noted for any of
the schools for smoking or loneliness. Parental communication was a marginally significant
predictor of loneliness for school four, such that poorer parental communication was associated
with higher loneliness. Table 16 includes the results from the meta-analysis. The meta-analysis
did not find support for any of the proposed hypotheses. The associations among loneliness,
smoking, and the friendship network were not statistically significant, using either meta-analysis
method.
Aim 3.3 was addressed by computing the co-evolution model with an interaction term for
the effect of influence from the friendship network and the effect of loneliness on smoking.
Convergence for school one was 1.77 and for school four was 2.49. We were unable to obtain
convergence for these schools, therefore, we will not report results from them. For school two,
convergence was .18 and for school three it was .13, indicating that these models converged and
we can interpret the results. For school two, the parameter estimate was -.68 (se = 1.98) and for
school three it was .36 (se = 1.61). The interaction effect did not reach statistical significance for
either school (ps > .05).
Discussion
Overall, the hypotheses for this study were not supported. For aim 3.1, we did find
similar rates of loneliness to those reported in other studies of Latino adolescents residing in the
Los Angeles area (Benner, 2011). Categorizations for low and high loneliness in the present
study were consistent with rates reported by Benner (2011), despite the two studies using
different surveys to assess loneliness. Aim 3.1 included assessing the effects of the network on
loneliness (influence) and effects of loneliness on the network (selection). No influence effects
were found. Using different methodology, past studies have found that loneliness may be
88
contagious and spread throughout the network (Cacioppo et al., 2009), however, we did not find
that a participant’s alters levels of loneliness affected the participant’s own loneliness. No
selection effects were found for loneliness either. However, for school one there was a
significant positive alter effect for loneliness, which indicates that lonely adolescents receive
more friendships. This would indicate that lonely adolescents are desirable as friends, which is
not in line with past studies which suggest that lonely people are pushed towards the periphery of
a network (Cacioppo et al., 2006). Recent research suggests that young adults are able to
recognize when their friends are lonely, although they tend to rate their friend’s loneliness as
slightly lower than the adolescent would rate their own loneliness (Luhmann, Bohn, Holtmann,
Koch, & Eid, 2016). We would expect that adolescents, similarly, would be able to recognize
that their friends are lonely. It is unexpected that adolescents would select lonely peers as
friends, and further research may be needed to clarify this. Although, we do note that this
phenomena was not observed for the other three schools.
Aim 3.2 was overall, not supported. Selection effects were noted for school two, such that
adolescents selected friends who had their same smoking status. This indicates that nonsmokers
choose nonsmokers as friends and smokers choose smokers as friends. Influence effects were not
found for any school. The results of the meta-analysis supported findings of a previous study
examining selection and influence in smoking behavior in the Social Network Study dataset for
10
th
grade data (Huang et al., 2014). For both the present study and Huang et al.’s (2014) study,
no effects were found for influence and a marginally significant effect was found for selection
using Fisher’s method for the right side test, indicating that at least one of the networks has a
positive effect for smoking similarity. Other studies have also found more evidence for selection
mechanisms vs. influence mechanisms in smoking dynamics (Mercken, Snijders, Steglich, & de
89
Vries, 2009). Additionally, the ego effect for smoking was significant for schools one and two,
but in opposite directions. For school one, smokers sent out fewer ties in comparison to
nonsmokers, and for school two, smokers sent out more ties in comparison to nonsmokers.
Smoking rates differed for these two schools. Smoking rates were higher for school one and
increased more between 11
th
and 12
th
grade in comparison to school two. Potentially for school
one, with a higher and growing percentage of smokers, smokers did not feel like they were
marginalized and were not motivated to send out ties whereas for school two, smokers may have
viewed themselves and/or their behavior as marginalized and may have tried to increase their
connection to the network.
Neither loneliness nor smoking had significant effects on each other for any of the
schools. Hypothesis 3.3 was not supported, the interaction term for friendship influence and
loneliness was not a significant predictor of smoking. This hypothesis could only be tested for
two schools, because the model with the interaction term would not converge for two of the
schools. Neither of the interacted variables had main effects on smoking, and the interaction term
further supports that for these schools, loneliness and friendship influence do not have effects on
smoking behavior. The data for this study was collected over a year apart. Potentially, influence
and loneliness interact within a shorter time period. In laboratory studies, evidence suggests that
socially excluded people have increased compliance with peer behavior (Mead et al., 2011).
Potentially, adolescents who experience transient loneliness may react by increasing their
affiliation to peers by changing their behavior to be congruent with their friends’ behavior.
However, those adolescents who are chronically lonely may not be capable of altering their
behavior to reduce their loneliness. This interpretation is in line with recent theory concerning
the evolutionary basis of loneliness which suggests that loneliness evolved as a motivating
90
emotion to keep people connected for increased survival (Cacioppo, Cacioppo, & Boomsma,
2014). Although loneliness was assessed as having occurred within the past 7 days, those
adolescents who experienced high loneliness within those last 7 days may be more likely to be
those who experience chronic high loneliness as opposed to those who experienced a transient
loneliness event, and there is no evidence that those adolescents who experienced loneliness
transiently experienced it simultaneously with exposure to peer smoking behavior. Loneliness
and peer influence may not interact within longitudinal studies, although loneliness may
moderate compliance with peer behavior within a short time span.
This study does have limitations. One major limitation is that we only included data from
those adolescents who participated in both the 11
th
and 12
th
grade data collection. Adolescents
who missed data collection may also be those more likely to smoke, be isolated in the friendship
network, and/or experience loneliness. Unfortunately, given our modeling strategy, it would have
been difficult to include adolescents who missed data collection. Additionally, the data came
from four schools located in one school district. The dynamics which operate on loneliness,
smoking, and friendship within these four schools may be unique to the school district or due to
larger community factors, and may not be generalizable to schools in other parts of the country.
The loneliness measure for this study was a one-item measure. It may have not captured
some of the variability in loneliness. Some adolescents may have avoided self-identifying
themselves as lonely due to stigma associated with loneliness. Additionally, we were unable to
control for depression in analyses because the loneliness item came from a depressive affect
scale, and including the rest of the scale would likely have caused multi-collinearity in the
model. Loneliness and depression are associated, yet distinct constructs (Cacioppo et al., 2010),
91
and it may be important to control for the effects of depression when examining the evolution of
loneliness and the effect of loneliness on smoking and friendship.
Few variables were entered into the model as predictors of loneliness and smoking. When
covariate effects were score tested, few effects reached low enough p-values to support adding
the effects to the model. Many effects which would likely have been significant predictors of
loneliness and smoking in multiple linear regression models were not significant predictors
within the SAOM. We do not suggest interpreting that these covariates were not associated with
the outcome variables, and instead consider that the SAOM may not have been appropriately
powered to find these effects.
92
Table 12: Descriptive Statistics by School
S1 (n = 192) S2 (n = 181) S3 (n = 251) S4 (n = 148)
Wave 11
th
12
th
11
th
12
th
11
th
12
th
11
th
12
th
Cig Use 41 (22) 49 (27) 30 (17) 32 (18) 40 (16) 43 (19) 36 (25) 44 (30)
Loneliness* 1.73 (1.07) 1.79 (1.10) 1.77 (1.09) 1.63 (.99) 1.70 (1.03) 1.61 (.97) 1.62 (.98) 1.65 (.97)
Free lunch 157 (82) 167 (92) 200 (82) 129 (90)
Hispanic (of any race) 141 (77) 165 (93) 111 (46) 134 (92)
Asian 39 (21) 11 (6) 137 (57) 11 (8)
Female 103 (54) 103 (57) 127 (51) 74 (50)
Age 18 112 (61) 79 (44) 129 (54) 74 (51)
Academic Performance
Mostly A’s 9 (5) 8 (4) 42 (17) 12 (8)
Mostly A’s and B’s 34 (18) 37 (21) 70 (28) 43 (29)
Mostly B’s 17 (9) 16 (9) 22 (9) 12 (8)
Mostly B’s and C’s 72 (38) 67 (38) 71 (29) 49 (33)
Mostly C’s 24 (13) 15 (8) 18 (7) 5 (3)
Mostly C’s and D’s 28 (15) 24 (13) 22 (9) 21 (14)
Mostly D’s 1 (1) 4 (2) 2 (1) 1 (1)
Mostly D’s and F’s 7 (4) 7 (4) 2 (1) 5 (3)
Parental Communication* 2.33 (.79) 2.28 (.84) 2.42 (.77) 2.25 (.80)
Linguistic Acculturation* 1.98 (.82) 2.43 (.70) 2.03(.73) 2.20 (.84)
Sibling Cig Use 36 (19) 18 (10) 31 (13) 20 (14)
Adult Cig Use* .32 (.59) .37 (.61) .36 (.57) .23 (.51)
Note. For categorical variables, n (%) are presented. For continuous variables, marked with an *, mean (SD) are presented. This data is
the unimputed data, therefore percentages are not necessarily calculated using the number of observations with data available as the
denominator. RSiena imputes for missing values internally.
93
Table 13: Effects included in Coevolution models
Effect Name Description
Endogenous Network Effects
rate: basic rate parameter net Change rate for network
eval: outdegree (density) Number of out-going ties
eval: reciprocity Number of reciprocated ties
eval: transitive ties Number of alters directly and indirectly connected to
eval: GWESP I -> K -> J (69) Geometrically Weighted Edgewise shared partners
eval: indegree - popularity (sqrt) Sum of the square roots of the in-degrees of i's alters
eval: outdegree - popularity (sqrt) Sum of the square roots of the out-degrees of i's alters
Covariate Network Effects
eval: female alter Number of female alters
eval: female ego Number of out-going ties weighted by female
eval: same female Number of i's alters that are i's gender
eval: age18 ego Number of out-going ties weighted by age 18
eval: hispanic alter Number of Hispanic alters
eval: hispanic ego Number of out-going ties weighted by Hispanic
eval: same hispanic Number of i's alters that are i's ethnicity (Hispanic vs. non-
Hispanic)
eval: perf alter Sum of school performance for i's alters
eval: perf ego Number of out-going ties weighted by school performance
eval: perf similarity Sum of similarity on school performance for i and i's alters
eval: accult alter Sum of acculturation for i's alters
eval: accult ego Number of out-going ties weighted by acculturation
eval: accult similarity Sum of similarity on acculturation for i and i's alters
eval: sibsmk ego Number of out-going ties weighted by sibling smoking
eval: adsmk alter Sum of adults who smoke for i's alters
eval: adsmk ego Number of out-going ties weighted by adult smoking
eval: adsmk similarity Sum of similarity on adult smoking for i and i's alters
Hypothesized Network Effects
eval: smk alter Number of alters who smoke
eval: smk ego Number of out-going ties weighted by smoking
eval: same smk Number of i's alters that have i's smoking status (smoke in
past 12 months or not)
eval: loneb alter Sum of loneliness for i's alters
eval: loneb ego Number of out-going ties weighted by loneliness
eval: loneb similarity Sum of similarity on loneliness for i and i's alters
Smoking Behavior Effects
rate: rate period 1 Change rate for smoking behavior
eval: linear shape Linear trend for smoking behavior
eval: total similarity Sum of centered similarity scores for i and i's alters
eval: effect from female Effect of i's gender
eval: effect from lunch Effect of i's SES
94
eval: effect from hispanic Effect of i's ethnicity
eval: effect from perf Effect of i's academic performance
eval: effect from sibsmk Effect of i's siblings' smoking behavior
eval: effect from adsmk Effect of i's closest adults' smoking behavior
eval: effect from loneb Effect of i's loneliness
eval: tot. sim. (net) x ego's
loneb*
Sum of centered similarity scores for i and i's alters
interacted with i's loneliness
Loneliness Behavior Effects
rate: rate period 1 Change rate for loneliness
eval: linear shape Linear trend for loneliness
eval: quadratic shape Effect of loneliness on itself
eval: average alter i's loneliness multiplied by average loneliness of i's alters
eval: dense triads i's loneliness multiplied by the number of dense triads which
i is in
eval: effect from female Effect of i's gender
eval: effect from parcomm Effect of i's parental communication
eval: effect from smk Effect of i's smoking behavior
Note. These definitions come from the Manual for RSiena (Ripley et al., 2016). *The interaction
effect was not included in the model used to assess aims 3.1 and 3.2 due to convergence issues.
95
Table 14: Model Fit and Descriptives
S1 S2 S3 S4
Network Statistics
Jaccard's Index 0.236 0.284 0.279 0.263
Avg Deg 11
th
grade 3.63 5.901 5.534 3.953
Avg Deg 12
th
grade 2.365 4.42 3.797 3.318
Number of ties 11
th
697 1068 1389 585
Number of ties 12
th
454 800 953 491
Density 11
th
0.019 0.033 0.022 0.027
Density 12
th
0.012 0.025 0.015 0.023
GOF and
Convergence
Convergence 0.1609 0.1344 0.1559 0.1366
In-degree 0.829 0.156 0.326 0.318
Out-degree 0.397 0.299 0.008 0.323
Loneliness Distribution >.99 0.173 0.959 0.264
Geodesic Distribution 0.649 0.479 0.099 0.813
Triad Census <.0001 <.0001 <.0001 0.001
Constraint Distribution 0.027 0.005 0.025 0.572
Note. GOF and convergence information is provided for the model used to assess aims 3.1 and
3.2, which did not include the interaction term between influence and loneliness in the prediction
of smoking behavior.
96
Table 15: Results from Smoking, Loneliness, and Friendship Network Coevolution model
by school
S1 S2 S3 S4
Par se Par se Par se Par se
Network Evolution
basic rate parameter net 10.09 .70 15.61 .88 13.69 .74 11.79 .98
outdegree (density) -2.64*** .39 -2.72*** .29 -2.63*** .21 -2.77*** .26
reciprocity 2.71*** .17 2.06*** .11 2.20*** .12 2.50*** .14
transitive ties .76** .26 .08 .12 .20 .13 .47** .18
GWESP I -> K -> J (69) .68† .36 1.21*** .12 1.25*** .15 .77*** .22
indegree - popularity (sqrt) -.40 .28 .02 .08 -.01 .07 .15 .12
outdegree - popularity (sqrt) -.47** .18 -.40*** .09 -.55*** .08 -.60*** .12
female alter .01 .12 -.15* .07 .05 .07 -.11 .09
female ego .03 .12 .14† .07 -.21** .07 -.04 .11
same female .46*** .10 .23*** .07 .14* .06 .30*** .08
age18 ego -.26** .10 .17* .07 .00 .07 -.22* .09
hispanic alter .16 .15 .14 .20 .08 .08 -.39* .18
hispanic ego .04 .15 .13 .22 .07 .09 -.01 .21
same hispanic .30** .11 .01 .16 .53*** .07 .44** .15
perf alter .03 .04 .01 .02 .04 .02 .03 .03
perf ego -.03 .04 -.01 .02 -.04† .02 .06* .03
perf similarity .11 .22 .76*** .17 .49** .17 .47* .20
accult alter .00 .06 -.05 .05 .00 .04 .02 .05
accult ego .14* .06 .04 .05 .00 .05 .06 .06
accult similarity -.17 .18 -.27† .14 .18 .14 .42** .16
sibsmk ego .11 .13 .07 .12 -.17 .11 .21 .13
adsmk alter -.38** .13 .08 .06 .11† .06 -.04 .13
adsmk ego -.04 .12 .12† .06 .03 .06 -.23† .14
adsmk similarity -.46† .24 .00 .12 -.34* .13 -.22 .27
smk alter .33 .25 -.01 .18 -.11 .15 .12 .18
smk ego -.48* .23 .67** .23 -.22 .15 -.03 .18
same smk .27 .23 .51* .25 .10 .18 -.06 .20
loneb alter .25* .12 .00 .08 -.06 .10 .01 .14
loneb ego -.13 .09 -.07 .08 .10 .10 -.05 .15
loneb similarity .17 .40 -.41 .32 -.51 .43 -.22 .62
Smoking Evolution
rate smk period 1 .55 .15 .86 .24 .47 .18 .63 .17
linear shape -1.02 .63 -1.75* .79 -1.19 .77 -.31 .57
total similarity -.20 .39 .01 .18 .21 .27 .31 .37
effect from female -.39 .78 -1.24 .89 -.13 .96 -1.13 .83
effect from lunch .96 1.10 -1.26 1.09 -1.69† .99 -.45 1.46
effect from hispanic .30 .93 1.24 1.75 0 fixed -1.76 1.89
effect from perf -.31 .24 .09 .30 -.40 .30 .14 .27
97
effect from sibsmk .40 .92 2.76* 1.29 2.40* 1.12 .43 1.26
effect from adsmk .63 .61 .57 .60 .28 .71 -1.36 1.26
effect from loneb -.42 .59 .49 .54 .33 .77 -.07 .67
Loneliness Evolution
rate loneb period 1 5.16 1.68 4.69 1.23 6.33 1.28 3.79 1.16
linear shape -1.19*** .15 -1.20*** .17 -1.13*** .15 -1.00*** .18
quadratic shape .48*** .09 .48*** .10 .49*** .09 .38*** .11
average alter -.28 .36 .50 .44 .34 .27 .16 .47
dense triads .27† .14 -.01 .02 -.03 .03 -.06 .05
effect from female .27 .18 .13 .21 .10 .14 .31 .21
effect from parcomm -.08 .10 .01 .11 -.05 .09 .28† .15
effect from smk -.29 .31 .15 .34 -.26 .26 -1.03 .69
Note. The model for S3 did not include an effect from ethnicity (Hispanic) on smoking because
the model would not reach adequate convergence when this parameter was added, likely due to
multi-colinearity.
98
Table 16: Meta-analysis results for SAOM of loneliness, smoking, and the friendship
network
Snijder’s and Baerveldt’s Method Fisher’s Method
Est.
Par se p
Est.
SD p
Right- sided
p
Left-sided
p
Network Evolution
Basic rate parameter 12.78
1.20
.002 2.39 .001 < .001 1.000
outdegree (density) -2.69
.04
< .001 .07 .977 1.000 < .001
reciprocity 2.35
.14
< .001 .29 .006 < .001 1.000
transitive ties .32
.14
.106 .28 .054 < .001 .999
GWESP I -> K -> J (69) 1.07
.14
.004 .27 .156 < .001 1.000
indegree - popularity (sqrt) .01
.08
.874 .15 .285 .443 .430
outdegree - popularity
(sqrt) -.50 .05 .002 .10 .474 1.000 .001
female alter -.05
.05
.369 .10 .174 .788 .068
female ego -.02
.08
.780 .16 .007 .265 .041
same female .27
.07
.026 .13 .041 < .001 1.000
age18 ego -.07
.10
.534 .20 < .001 .215 .006
hispanic alter .01
.12
.927 .24 .088 .243 .313
hispanic ego .06
.02
.071 .04 .972 .368 .887
same hispanic .34
.11
.052 .22 .015 < .001 .996
perf alter .02
.01
.058 .02 .796 .074 .981
perf ego -.01
.02
.824 .05 .041 .313 .142
perf similarity .48
.13
.034 .26 .138 < .001 .999
accult alter -.01
.01
.682 .03 .805 .722 .527
accult ego .05
.03
.199 .06 .314 .033 .966
accult similarity .04
.16
.811 .32 .005 .043 .217
sibsmk ego .05
.08
.621 .17 .122 .158 .553
adsmk alter -.04
.11
.746 .22 .006 .160 .058
adsmk ego .00
.07
.958 .14 .108 .229 .344
adsmk similarity -.22
.11
.138 .21 .168 .983 .004
smk alter .03
.09
.739 .17 .435 .306 .735
smk ego -.02
.24
.942 .48 .002 .070 .064
same smk .17
.12
.238 .23 .326 .052 .943
loneb alter .04
.07
.579 .13 .220 .162 .736
loneb ego -.04
.05
.434 .10 .426 .745 .236
loneb similarity -.26
.16
.197 .32 .636 .898 .162
Smoking Evolution
Rate period 1 .60
.07
.003 .14 .607 < .001 1.000
linear shape -.94
.30
.053 .61 .494 .998 .004
total similarity .07
.09
.518 .18 .739 .398 .795
99
effect from female -.73
.26
.070 .53 .768 .972 .085
effect from lunch -.68
.62
.351 1.23 .310 .797 .119
effect from hispanic .14
.73
.861 1.26 .487 .517 .548
effect from perf -.13
.14
.409 .27 .410 .792 .195
effect from sibsmk 1.38
.63
.115 1.26 .296 .008 .987
effect from adsmk .37
.31
.321 .62 .529 .283 .703
effect from loneb .08
.22
.736 .44 .694 .502 .683
Loneliness Evolution
Rate period 1 4.90
.57
.003 1.13 .528 < .001 1.000
linear shape -1.14
.04
< .001 .09 .835 1.000 < .001
quadratic shape .47
.02
< .001 .05 .844 < .001 1.000
average alter .19
.17
.360 .35 .456 .189 .811
dense triads .00
.06
.981 .12 .116 .374 .189
effect from female .18
.05
.041 .11 .793 .040 .993
effect from parcomm .01
.07
.855 .15 .189 .266 .543
effect from smk -.22
.16
.274 .32 .459 .931 .100
Note. Significant results are in bold type.
100
CHAPTER 5: CONCLUSION
This dissertation examined the associations of loneliness, parental communication, school
social networks, and depression with substance use, with a focus on cigarette use. It includes
three studies which examine data from adolescents enrolled at five high schools in one school
district in Southern California, USA. The studies all used different methodologies including
structural equation modeling, linear and logistic regression, and stochastic actor-oriented
modeling. The data were examined cross-sectionally and longitudinally, and as a whole,
separately by school, and controlling for nesting within schools.
Loneliness, overall, was not associated with the outcomes analyzed in this study. Among
females, loneliness was found to be positively associated with alcohol use. Loneliness was
associated with higher odds of having at least one friend who smokes and at least one friend who
uses alcohol at least once a month. Additionally, a significant interaction was found between
loneliness and a friend’s smoking/drinking behavior in logistic regressions with the outcome of
the participant and their friend having used alcohol or cigarettes together, suggesting that
loneliness decreases the effect of the friend’s influence on the participant having used alcohol or
cigarettes with the friend. Lonely adolescents may have more friends who use cigarettes and
alcohol, but they may also be less likely to participate in cigarette and alcohol use with their
friends who use alcohol and cigarettes in comparison to non-lonely adolescents.
Further research might explore the directionality of these associations. Do lonely
adolescents make friends with peers who use alcohol and tobacco or do adolescents who are
friends with peers who use alcohol and tobacco become lonely? Potentially an adolescent’s lack
of participation in informal social activities with friends, such as those involving cigarette or
alcohol use, can make them feel lonely. Loneliness theories suggest that loneliness should be a
101
motivating affective state (Cacioppo et al., 2014). Potentially loneliness experienced when
friends begin participating in activities which the adolescent does not want to be involved in is a
stimulus to make new friends who have similar interests to the adolescent.
Low closeness centrality and high out-degree were associated with many of the substance
use outcomes in study one. However, out-degree was not associated with tobacco or alcohol use
in study two, although it was associated with higher odds of having at least one friend who
smokes and at least one friend who drinks at least once a month, and lower accuracy in
perception of friends’ smoking and drinking behaviors. Out-degree was assessed in study one as
number of friends named within the 12
th
grade school network, whereas study two assessed out-
degree as number of friends named regardless of where they live or go to school. Furthermore,
study two controlled for having at least one friend who smokes/drinks. Potentially out-degree
alone is not associated with substance use but instead increases the likelihood of having a friend
who uses substances which increases the likelihood of the adolescent engaging in substance use
themselves.
Furthermore, while study two found that having a friend who smokes is associated with
higher odds of smoking, study three did not find selection or influence effects for smoking, with
the exception of selection effects for one school. These results seem to contradict each other, in
that study two suggests that adolescents who smoke have friends who smoke also, and study
three indicated that this phenomenon does not occur due to selection or influence. Studies two
and three did have many methodological differences. Study two used ever vs. never smoking as
the outcome, while study three used cigarette use within the last 12 months. Study two used
egocentric network data and friend’s smoking behavior was assessed as perceived friend’s
smoking behavior while study three used sociometric data and friend’s smoking behavior was
102
self-reported by the adolescent’s friend. Furthermore, study two was cross-sectional while study
three was longitudinal and used a model specifically designed for network data. These
methodological differences may account for the different conclusions drawn from results of
studies two and three. Research suggests that perceived peer smoking is a better predictor of
smoking in comparison to actual peer smoking behavior and that peer smoking is a better
predictor of smoking in cross-sectional studies in comparison to longitudinal studies (Valente,
Fujimoto, Soto, et al., 2013). Therefore, these contradictory findings are in line with past studies.
It is still surprising, though, that we did not find selection or influence effects in study three,
given that many SAOMs have found selection and influence for smoking. These mechanisms
may not control smoking or network evolution for this sample.
This dissertation additionally examined parental communication, depressive affect,
linguistic acculturation, and identity. Linguistic acculturation was not assessed in 12
th
grade, and
could only be included in study three. It did not enter the model as a predictor of smoking or
loneliness. Similarity on acculturation was a significant predictor of network evolution for one
school. Future research might use a different methodology to examine associations among
linguistic acculturation, loneliness, smoking, and friendship networks. The SAOM was
developed for analyzing network data and may not be the best technique for analyzing the
associations of linguistic acculturation with loneliness and smoking.
Poor parental communication was associated with substance use in study one. In study
three it was marginally positively associated with loneliness for one school. Parental
communication did not enter the model as a predictor of smoking in study 3. Overall, it appears
that poor parental communication is associated with substance use, particularly for use of
alcohol, marijuana, and cigarettes, and may be of particular importance for adolescents who did
103
not identify as Hispanic or Latino, which for this sample are predominately Asian American
adolescents.
High depressive affect was consistently associated with use of cigarettes, e-cigarettes,
and marijuana throughout study one. It was also associated with higher odds of cigarette use in
study two. However, depressive affect was not associated with alcohol use in either study, or
hookah use in study one. Alcohol use was prevalent in this sample, with 58% of the sample
having used alcohol at least once. For these adolescents, alcohol use may be a social activity
which is unrelated to general feelings of depressive affect. Depressive affect was associated with
cigarette use across studies one and two. The consistency of these findings, considering that
different methodologies were used in the studies, does suggest that depressive affect is an
important factor to consider when studying cigarette use etiology and consequences in
adolescents.
Stoner/druggie identity was associated with greater accuracy of peer behavior report,
higher odds of having at least one friend who smokes and at least one friend who uses alcohol,
higher odds of using cigarettes, and higher frequency of drinking. An adolescent’s social identity
as someone who uses drugs is associated with their drug use behavior. This cross-sectional study
cannot provide more information as to if perceiving oneself socially as a drug user leads to
substance use or if an adolescent’s behavior precedes defining themselves socially as a
stoner/druggie, but it does suggest that people who perceive themselves socially as drug users do
have higher drug use and are friends with more drug using peers in comparison to their peers.
Studies two and three did not find significant interactions between loneliness and peer
smoking, or any other significant interactions for the outcomes alcohol and cigarette use. While
laboratory studies have found that lonely people are more likely to comply with their peer’s
104
behavior and preferences in comparison to nonlonely people, this survey study which examined
real world dynamics of smoking behavior over a long period of time did not find the same
behavior. Potentially, this mechanism may operate in the real world, but on a much shorter time
scale. Given that loneliness is a normal human experience, adolescents may feel loneliness
transiently, and when in a transient lonely state, they may be more likely to comply with peer
behavior. Those participants who scored higher than their peers on the loneliness questionnaires
may be adolescents who are chronically lonely. Chronically lonely adolescents may be
chronically lonely because they do not respond to their loneliness with coping mechanisms that
are effective at reducing loneliness. Reconnecting to peers is likely an effective coping method
for reducing loneliness, and may be accomplished by joining in with peers for social activities,
even if the activities are aversive or harmful to the adolescent’s health (Rawn & Vohs, 2011).
Therefore, this potential dynamic may not be observed in a longitudinal study with data collected
over a year apart.
Furthermore, laboratory studies of peer influence are generally conducted with peer
influence occurring through a confederate, as opposed to a friend of the participant. Compliance
with a novel person’s attitude and behavior may not be the same as compliance to a friend’s
behavior and attitudes. Furthermore, adolescents may be more influenced to comply with
smoking behavior by new friendships in comparison to old friendships (Aloise-Young et al.,
1994). Lonely adolescents may not be influenced by their current friend’s behavior. Potentially
those adolescents who are lonely do seek out new friendships, and are influenced by the behavior
of these new friends. Adolescence is a time for exploring identity through friendships (Laursen
& Hartl, 2013). Transient loneliness may instigate or be the result of a change in friendships, and
105
exposure to new friends may influence an adolescent to change their behaviors, or they may
select new friends who engage in a behavior that they already do or want to engage in.
This dissertation’s findings suggest that smoking prevention programming for
adolescents should focus on reducing depression and improving parental communication to
reduce adolescent cigarette use. Programming could include coping skills for those adolescents
experiencing depressive affect or highlight community and school-based adults who can serve as
role models and confidants for those adolescents lacking parents who they can discuss personal
matters with and receive support from. Programming should be appealing to adolescents who are
less central in the network, based on findings from study one, but not necessarily aimed at
adolescents who have no or few friends, as study one also found that out-degree was positively
associated with smoking. Loneliness may not be an important target for prevention programs.
Prevention programs with limited funds may focus on reaching adolescents who have friends
who smoke, as they are more likely to use cigarettes themselves, but from this study, we do not
have evidence that influence or selection operates to result in this association.
106
REFERENCES
Alexander, C., Piazza, M., Mekos, D., & Valente, T. (2001). Peers, schools, and adolescent
cigarette smoking. The Journal of Adolescent Health: Official Publication of the Society
for Adolescent Medicine, 29(1), 22–30.
Allen, O., Page, R. M., Moore, L., & Hewitt, C. (1994). Gender differences in selected
psychosocial characteristics of adolescent smokers and nonsmokers. Health Values: The
Journal of Health Behavior, Education & Promotion, 18(2), 34–39.
Aloise-Young, P. A., Graham, J. W., & Hansen, W. B. (1994). Peer influence on smoking
initiation during early adolescence: a comparison of group members and group outsiders.
The Journal of Applied Psychology, 79(2), 281–287.
Asher, S. R., & Paquette, J. A. (2003). Loneliness and peer relations in childhood. Current
Directions in Psychological Science, 12(3), 75–78. http://doi.org/10.1111/1467-
8721.01233
Asparouhov, T., & Muthén, B. (2010). Simple Second Order Chi-Square Correction (Mplus
Technical Appendices). Retrieved from
http://www.statmodel.com/download/WLSMV_new_chi21.pdf
Bandalos, D. L. (2014). Relative Performance of Categorical Diagonally Weighted Least
Squares and Robust Maximum Likelihood Estimation. Structural Equation Modeling: A
Multidisciplinary Journal, 21(1), 102–116.
http://doi.org/10.1080/10705511.2014.859510
Benner, A. D. (2011). Latino Adolescents’ Loneliness, Academic Performance, and the
Buffering Nature of Friendships. Journal of Youth and Adolescence, 40(5), 556–567.
http://doi.org/10.1007/s10964-010-9561-2
107
Bethel, J. W., & Schenker, M. B. (2005). Acculturation and Smoking Patterns Among Hispanics.
American Journal of Preventive Medicine, 29(2), 143–148.
http://doi.org/10.1016/j.amepre.2005.04.014
Brage, D., Meredith, W., & Woodward, J. (1993). Correlates of loneliness among midwestern
adolescents. Adolescence, 28(111), 685–693.
Burk, W. J., Steglich, C. E. G., & Snijders, T. A. B. (2007). Beyond dyadic interdependence:
Actor-oriented models for co-evolving social networks and individual behaviors.
International Journal of Behavioral Development, 31(4), 397–404.
http://doi.org/10.1177/0165025407077762
Cacioppo, J. T., Cacioppo, S., & Boomsma, D. I. (2014). Evolutionary mechanisms for
loneliness. Cognition and Emotion, 28(1), 3–21.
http://doi.org/10.1080/02699931.2013.837379
Cacioppo, J. T., Fowler, J. H., & Christakis, N. A. (2009). Alone in the crowd: The structure and
spread of loneliness in a large social network. Journal of Personality and Social
Psychology, 97(6), 977–991. http://doi.org/10.1037/a0016076
Cacioppo, J. T., & Hawkley, L. C. (2003). Social isolation and health, with an emphasis on
underlying mechanisms. Perspectives in Biology and Medicine, 46(3 Suppl), S39-52.
Cacioppo, J. T., Hawkley, L. C., Crawford, L. E., Ernst, J. M., Burleson, M. H., Kowalewski, R.
B., … Berntson, G. G. (2002). Loneliness and health: potential mechanisms.
Psychosomatic Medicine, 64(3), 407–417.
Cacioppo, J. T., Hawkley, L. C., Ernst, J. M., Burleson, M., Berntson, G. G., Nouriani, B., &
Spiegel, D. (2006). Loneliness within a nomological net: An evolutionary perspective.
108
Journal of Research in Personality, 40(6), 1054–1085.
http://doi.org/10.1016/j.jrp.2005.11.007
Cacioppo, J. T., Hawkley, L. C., & Thisted, R. A. (2010). Perceived social isolation makes me
sad: 5-year cross-lagged analyses of loneliness and depressive symptomatology in the
Chicago Health, Aging, and Social Relations Study. Psychology and Aging, 25(2), 453–
463. http://doi.org/10.1037/a0017216
Chaiton, M. O., Cohen, J. E., O’Loughlin, J., & Rehm, J. (2009). A systematic review of
longitudinal studies on the association between depression and smoking in adolescents.
BMC Public Health, 9, 356. http://doi.org/10.1186/1471-2458-9-356
Chipuer, H. M. (2001). Dyadic attachments and community connectedness: Links with youths’
loneliness experiences. Journal of Community Psychology, 29(4), 429–446.
http://doi.org/10.1002/jcop.1027
Christopherson, T. M., & Conner, B. T. (2012). Mediation of Late Adolescent Health-Risk
Behaviors and Gender Influences. Public Health Nursing, 29(6), 510–524.
http://doi.org/10.1111/j.1525-1446.2012.01007.x
Cohen, D. A., Richardson, J., & LaBree, L. (1994). Parenting behaviors and the onset of
smoking and alcohol use: a longitudinal study. Pediatrics, 94(3), 368–375.
DeWall, C. N., & Pond, R. S. (2011). Loneliness and smoking: The costs of the desire to
reconnect. Self and Identity, 10(3), 375–385.
http://doi.org/10.1080/15298868.2010.524404
Dyal, S. R., & Valente, T. W. (2015). A Systematic Review of Loneliness and Smoking: Small
Effects, Big Implications. Substance Use & Misuse, 50(13), 1697–1716.
http://doi.org/10.3109/10826084.2015.1027933
109
Ennett, S. T., & Bauman, K. E. (1993). Peer group structure and adolescent cigarette smoking: a
social network analysis. Journal of Health and Social Behavior, 34(3), 226–236.
Ennett, S. T., & Bauman, K. E. (1994). The contribution of influence and selection to adolescent
peer group homogeneity: the case of adolescent cigarette smoking. Journal of Personality
and Social Psychology, 67(4), 653–663.
Ennett, S. T., Bauman, K. E., Hussong, A., Faris, R., Foshee, V. A., Cai, L., & DuRant, R. H.
(2006). The Peer Context of Adolescent Substance Use: Findings from Social Network
Analysis. Journal of Research on Adolescence, 16(2), 159–186.
http://doi.org/10.1111/j.1532-7795.2006.00127.x
Ennett, S. T., Faris, R., Hipp, J., Foshee, V. A., Bauman, K. E., Hussong, A., & Cai, L. (2008).
Peer Smoking, Other Peer Attributes, and Adolescent Cigarette Smoking: A Social
Network Analysis. Prevention Science, 9(2), 88–98. http://doi.org/10.1007/s11121-008-
0087-8
Epstein, J. A., Botvin, G. J., & Diaz, T. (1998). Linguistic acculturation and gender effects on
smoking among Hispanic youth. Preventive Medicine, 27(4), 583–589.
http://doi.org/10.1006/pmed.1998.0329
Epstein, J. A., Botvin, G. J., Dusenbury, L., Diaz, T., & Kerner, J. (1996). Validation of an
Acculturation Measure for Hispanic Adolescents. Psychological Reports, 79(3), 1075–
1079. http://doi.org/10.2466/pr0.1996.79.3.1075
Freitag, M. K., Belsky, J., Grossmann, K., Grossmann, K. E., & Scheuerer-Englisch, H. (1996).
Continuity in Parent-Child Relationships from Infancy to Middle Childhood and
Relations with Friendship Competence. Child Development, 67(4), 1437.
http://doi.org/10.2307/1131710
110
Fujimoto, K., & Valente, T. W. (2012). Decomposing the Components of Friendship and
Friends’ Influence on Adolescent Drinking and Smoking. Journal of Adolescent Health,
51(2), 136–143. http://doi.org/10.1016/j.jadohealth.2011.11.013
Gardner, W. L., Pickett, C. L., & Brewer, M. B. (2000). Social Exclusion and Selective Memory:
How the Need to belong Influences Memory for Social Events. Personality and Social
Psychology Bulletin, 26(4), 486–496. http://doi.org/10.1177/0146167200266007
Glanz, K., Rimer, B. K., & Viswanath, K. (Eds.). (2008). Health behavior and health education:
theory, research, and practice (4th ed). San Francisco, CA: Jossey-Bass.
Green, L. R., Richardson, D. S., Lago, T., & Schatten-Jones, E. C. (2001). Network Correlates of
Social and Emotional Loneliness in Young and Older Adults. Personality and Social
Psychology Bulletin, 27(3), 281–288. http://doi.org/10.1177/0146167201273002
Grunbaum, J. A., Tortolero, S., Weller, N., & Gingiss, P. (2000). Cultural, social, and
intrapersonal factors associated with substance use among alternative high school
students. Addictive Behaviors, 25(1), 145–151.
Handcock, M. S., Hunter, D. R., Butts, C. T., Goodreau, S. M., Krivitsky, P., Bender-deMoll, S.,
& Morris, M. (2014). statnet: Software Tools for the Representation, Visualization,
Analysis and Simulation of Network Data (Version 2014.2.0). The Statnet Project.
Retrieved from CRAN.R-project.org/package=statnet
Hawkley, L. C., Hughes, M. E., Waite, L. J., Masi, C. M., Thisted, R. A., & Cacioppo, J. T.
(2008). From social structural factors to perceptions of relationship quality and
loneliness: the Chicago health, aging, and social relations study. The Journals of
Gerontology. Series B, Psychological Sciences and Social Sciences, 63(6), S375-384.
111
Hays, R. D., & DiMatteo, M. R. (1987). A short-form measure of loneliness. Journal of
Personality Assessment, 51(1), 69–81. http://doi.org/10.1207/s15327752jpa5101_6
Heinrich, L. M., & Gullone, E. (2006). The clinical significance of loneliness: A literature
review. Clinical Psychology Review, 26(6), 695–718.
http://doi.org/10.1016/j.cpr.2006.04.002
Henry, D. B., Kobus, K., & Schoeny, M. E. (2011). Accuracy and bias in adolescents’
perceptions of friends’ substance use. Psychology of Addictive Behaviors, 25(1), 80–89.
http://doi.org/10.1037/a0021874
Hoffman, B. R., Sussman, S., Unger, J. B., & Valente, T. W. (2006). Peer influences on
adolescent cigarette smoking: a theoretical review of the literature. Substance Use &
Misuse, 41(1), 103–155. http://doi.org/10.1080/10826080500368892
Hornsey, M. J. (2008). Social Identity Theory and Self-categorization Theory: A Historical
Review. Social and Personality Psychology Compass, 2(1), 204–222.
http://doi.org/10.1111/j.1751-9004.2007.00066.x
Huang, G. C., Unger, J. B., Soto, D., Fujimoto, K., Pentz, M. A., Jordan-Marsh, M., & Valente,
T. W. (2014). Peer Influences: The Impact of Online and Offline Friendship Networks on
Adolescent Smoking and Alcohol Use. Journal of Adolescent Health, 54(5), 508–514.
http://doi.org/10.1016/j.jadohealth.2013.07.001
Iecovich, E., Jacobs, J. M., & Stessman, J. (2011). Loneliness, Social Networks, and Mortality:
18 Years of Follow-Up. The International Journal of Aging and Human Development,
72(3), 243–263. http://doi.org/10.2190/AG.72.3.e
112
Jones, S. E., Kann, L., & Pechacek, T. F. (2011). Cigarettes Smoked per Day Among High
School Students in the U.S., 1991–2009. American Journal of Preventive Medicine,
41(3), 297–299. http://doi.org/10.1016/j.amepre.2011.05.018
Kandel, D. B., Kiros, G.-E., Schaffran, C., & Hu, M.-C. (2004). Racial/ethnic differences in
cigarette smoking initiation and progression to daily smoking: a multilevel analysis.
American Journal of Public Health, 94(1), 128–135.
Kaplan, C. P., Nápoles-Springer, A., Stewart, S. L., & Pérez-Stable, E. J. (2001). Smoking
acquisition among adolescents and young Latinas: the role of socioenvironmental and
personal factors. Addictive Behaviors, 26(4), 531–550.
Khantzian, E. J. (1985). The self-medication hypothesis of addictive disorders: focus on heroin
and cocaine dependence. The American Journal of Psychiatry, 142(11), 1259–1264.
Kline, R. B. (2011). Principles and practice of structural equation modeling (3rd ed). New York:
Guilford Press.
Kniskern, J., Biglan, A., Lichtenstein, E., Ary, D., & Bavry, J. (1983). Peer modeling effects in
the smoking behavior of teenagers. Addictive Behaviors, 8(2), 129–132.
http://doi.org/10.1016/0306-4603(83)90006-0
Kobus, K. (2003). Peers and adolescent smoking. Addiction (Abingdon, England), 98 Suppl 1,
37–55.
Kristjansson, A. L., Mann, M. J., & Sigfusdottir, I. D. (2015). Licit and Illicit Substance Use by
Adolescent E-Cigarette Users Compared with Conventional Cigarette Smokers, Dual
Users, and Nonusers. The Journal of Adolescent Health: Official Publication of the
Society for Adolescent Medicine, 57(5), 562–564.
http://doi.org/10.1016/j.jadohealth.2015.07.014
113
La Greca, A. M., Prinstein, M. J., & Fetter, M. D. (2001). Adolescent peer crowd affiliation:
linkages with health-risk behaviors and close friendships. Journal of Pediatric
Psychology, 26(3), 131–143.
Ladd, G. W., & Ettekal, I. (2013). Peer-related loneliness across early to late adolescence:
normative trends, intra-individual trajectories, and links with depressive symptoms.
Journal of Adolescence, 36(6), 1269–1282.
http://doi.org/10.1016/j.adolescence.2013.05.004
Ladd, G. W., Kochenderfer, B. J., & Coleman, C. C. (1997). Classroom peer acceptance,
friendship, and victimization: distinct relational systems that contribute uniquely to
children’s school adjustment? Child Development, 68(6), 1181–1197.
Lakon, C. M., Hipp, J. R., & Timberlake, D. S. (2010). The Social Context of Adolescent
Smoking: A Systems Perspective. American Journal of Public Health, 100(7), 1218–
1228. http://doi.org/10.2105/AJPH.2009.167973
Lakon, C. M., & Valente, T. W. (2012). Social integration in friendship networks: The synergy
of network structure and peer influence in relation to cigarette smoking among high risk
adolescents. Social Science & Medicine, 74(9), 1407–1417.
http://doi.org/10.1016/j.socscimed.2012.01.011
Lara, M., Gamboa, C., Kahramanian, M. I., Morales, L. S., & Hayes Bautista, D. E. (2005).
Acculturation and Latino health in the United States: A review of the literature and its
sociopolitical context. Annual Review of Public Health, 26(1), 367–397.
http://doi.org/10.1146/annurev.publhealth.26.021304.144615
Larson, R., & Richards, M. H. (1991). Daily companionship in late childhood and early
adolescence: changing developmental contexts. Child Development, 62(2), 284–300.
114
Latané, B. (1996). Dynamic Social Impact: The Creation of Culture by Communication. Journal
of Communication, 46(4), 13–25. http://doi.org/10.1111/j.1460-2466.1996.tb01501.x
Lauder, W., Mummery, K., Jones, M., & Caperchione, C. (2006). A comparison of health
behaviours in lonely and non-lonely populations. Psychology, Health & Medicine, 11(2),
233–245. http://doi.org/10.1080/13548500500266607
Laursen, B., Hafen, C. A., Kerr, M., & Stattin, H. (2012). Friend influence over adolescent
problem behaviors as a function of relative peer acceptance: to be liked is to be emulated.
Journal of Abnormal Psychology, 121(1), 88–94. http://doi.org/10.1037/a0024707
Laursen, B., & Hartl, A. C. (2013). Understanding loneliness during adolescence: Developmental
changes that increase the risk of perceived social isolation. Journal of Adolescence,
36(6), 1261–1268. http://doi.org/10.1016/j.adolescence.2013.06.003
Ledbetter, A. M. (2009). Family Communication Patterns and Relational Maintenance Behavior:
Direct and Mediated Associations with Friendship Closeness. Human Communication
Research, 35(1), 130–147. http://doi.org/10.1111/j.1468-2958.2008.01341.x
Lee, Y. O., Hebert, C. J., Nonnemaker, J. M., & Kim, A. E. (2015). Youth tobacco product use in
the United States. Pediatrics, 135(3), 409–415. http://doi.org/10.1542/peds.2014-3202
Leung, G. T. Y., de Jong Gierveld, J., & Lam, L. C. W. (2008). Validation of the Chinese
translation of the 6-item De Jong Gierveld Loneliness Scale in elderly Chinese.
International Psychogeriatrics, 20(6), 1262. http://doi.org/10.1017/S1041610208007552
Luger, T. M., Suls, J., & Vander Weg, M. W. (2014). How robust is the association between
smoking and depression in adults? A meta-analysis using linear mixed-effects models.
Addictive Behaviors, 39(10), 1418–1429. http://doi.org/10.1016/j.addbeh.2014.05.011
115
Luhmann, M., Bohn, J., Holtmann, J., Koch, T., & Eid, M. (2016). I’m lonely, can’t you tell?
Convergent validity of self- and informant ratings of loneliness. Journal of Research in
Personality, 61, 50–60. http://doi.org/10.1016/j.jrp.2016.02.002
Lykes, V. A., & Kemmelmeier, M. (2014). What Predicts Loneliness? Cultural Difference
Between Individualistic and Collectivistic Societies in Europe. Journal of Cross-Cultural
Psychology, 45(3), 468–490. http://doi.org/10.1177/0022022113509881
Mahon, N. E., Yarcheski, A., & Yarcheski, T. J. (1998). Social Support and Positive Health
Practices in Young Adults: Loneliness as a Mediating Variable. Clinical Nursing
Research, 7(3), 292–308. http://doi.org/10.1177/105477389800700306
Mahon, N. E., Yarcheski, A., Yarcheski, T. J., Cannella, B. L., & Hanks, M. M. (2006). A meta-
analytic study of predictors for loneliness during adolescence. Nursing Research, 55(5),
308–315.
Marangoni, C., & Ickes, W. (1989). Loneliness: A Theoretical Review with Implications for
Measurement. Journal of Social and Personal Relationships, 6(1), 93–128.
http://doi.org/10.1177/026540758900600107
Matthews, T., Danese, A., Wertz, J., Odgers, C. L., Ambler, A., Moffitt, T. E., & Arseneault, L.
(2016). Social isolation, loneliness and depression in young adulthood: a behavioural
genetic analysis. Social Psychiatry and Psychiatric Epidemiology, 51(3), 339–348.
http://doi.org/10.1007/s00127-016-1178-7
Mead, N. L., Baumeister, R. F., Stillman, T. F., Rawn, C. D., & Vohs, K. D. (2011). Social
Exclusion Causes People to Spend and Consume Strategically in the Service of
Affiliation. The Journal of Consumer Research, 37(5), 902–919.
http://doi.org/10.1086/656667
116
Mercken, L., Snijders, T. A. B., Steglich, C., & de Vries, H. (2009). Dynamics of adolescent
friendship networks and smoking behavior: Social network analyses in six European
countries. Social Science & Medicine, 69(10), 1506–1514.
http://doi.org/10.1016/j.socscimed.2009.08.003
Myers, D., Baer, W. C., & Choi, S.-Y. (1996). The Changing Problem of Overcrowded Housing.
Journal of the American Planning Association, 62(1), 66–84.
http://doi.org/10.1080/01944369608975671
Myers, R., Chou, C.-P., Sussman, S., Baezconde-Garbanati, L., Pachon, H., & Valente, T. W.
(2009). Acculturation and Substance Use: Social Influence as a Mediator among Hispanic
Alternative High School Youth. Journal of Health and Social Behavior, 50(2), 164–179.
http://doi.org/10.1177/002214650905000204
Nangle, D. W., Erdley, C. A., Newman, J. E., Mason, C. A., & Carpenter, E. M. (2003).
Popularity, Friendship Quantity, and Friendship Quality: Interactive Influences on
Children’s Loneliness and Depression. Journal of Clinical Child & Adolescent
Psychology, 32(4), 546–555. http://doi.org/10.1207/S15374424JCCP3204_7
National Center for Chronic Disease Prevention and Health Promotion (US) Office on Smoking
and Health. (2012). Preventing Tobacco Use Among Youth and Young Adults: A Report
of the Surgeon General. Atlanta (GA): Centers for Disease Control and Prevention (US).
Retrieved from http://www.ncbi.nlm.nih.gov/books/NBK99237/
National Institute on Drug Abuse. High School and Youth Trends. (2014). Retrieved from
http://www.drugabuse.gov/publications/drugfacts/high-school-youth-trends
117
Neto, F., & Barros, J. (2000). Predictors of Loneliness among Adolescents from Portuguese
Immigrant Families in Switzerland. Social Behavior and Personality: An International
Journal, 28(2), 193–205. http://doi.org/10.2224/sbp.2000.28.2.193
Office of the Surgeon General (US), & Office on Smoking and Health (US). (2004). The Health
Consequences of Smoking: A Report of the Surgeon General. Atlanta (GA): Centers for
Disease Control and Prevention (US). Retrieved from
http://www.ncbi.nlm.nih.gov/books/NBK44695/
O’Malley, A. J., Arbesman, S., Steiger, D. M., Fowler, J. H., & Christakis, N. A. (2012).
Egocentric Social Network Structure, Health, and Pro-Social Behaviors in a National
Panel Study of Americans. PLoS ONE, 7(5), e36250.
http://doi.org/10.1371/journal.pone.0036250
Page, R. M., Dennis, M., Lindsay, G. B., & Merrill, R. M. (2010). Psychosocial Distress and
Substance Use Among Adolescents in Four Countries: Philippines, China, Chile, and
Namibia. Youth & Society, 43(3), 900–930. http://doi.org/10.1177/0044118X10368932
Page, R. M., Zarco, E. P. T., Ihasz, F., Suwanteerangkul, J., Uvacsek, M., Mei-Lee, C., …
Kalabiska, I. (2008). Cigarette Smoking and Indicators of Psychosocial Distress in
Southeast Asian and Central-Eastern European Adolescents. Journal of Drug Education,
38(4), 307–328. http://doi.org/10.2190/DE.38.4.a
Pampel, F. C. (2006). Socioeconomic Distinction, Cultural Tastes, and Cigarette Smoking*.
Social Science Quarterly, 87(1), 19–35. http://doi.org/10.1111/j.0038-4941.2006.00366.x
Parker, J. G., & Asher, S. R. (1993). Friendship and friendship quality in middle childhood:
Links with peer group acceptance and feelings of loneliness and social dissatisfaction.
Developmental Psychology, 29(4), 611–621. http://doi.org/10.1037/0012-1649.29.4.611
118
Patterson, A. C., & Veenstra, G. (2010). Loneliness and risk of mortality: A longitudinal
investigation in Alameda County, California. Social Science & Medicine, 71(1), 181–
186. http://doi.org/10.1016/j.socscimed.2010.03.024
Pearson, M., Sweeting, H., West, P., Young, R., Gordon, J., & Turner, K. (2006). Adolescent
substance use in different social and peer contexts: A social network analysis. Drugs:
Education, Prevention, and Policy, 13(6), 519–536.
http://doi.org/10.1080/09687630600828912
Pedersen, S., Vitaro, F., Barker, E. D., & Borge, A. I. H. (2007). The timing of middle-childhood
peer rejection and friendship: linking early behavior to early-adolescent adjustment.
Child Development, 78(4), 1037–1051. http://doi.org/10.1111/j.1467-8624.2007.01051.x
Peltzer, K. (2009). Prevalence and correlates of substance use among school children in six
African countries. International Journal of Psychology, 44(5), 378–386.
http://doi.org/10.1080/00207590802511742
Peplau, L. A., & Perlman, D. (Eds.). (1982). Loneliness: a sourcebook of current theory,
research, and therapy. New York: Wiley.
Perlman, D., & Peplau, L. A. (1981). Toward a social psychology of loneliness. In S. Duck & R.
Gilmour (Eds.), Personal Relationships in Disorder (Vol. 3, pp. 31–56). Academic Press.
Pierce, J. P., Choi, W. S., Gilpin, E. A., Farkas, A. J., & Berry, C. C. (1998). Tobacco industry
promotion of cigarettes and adolescent smoking. JAMA: The Journal of the American
Medical Association, 279(7), 511–515.
Pierce, J. P., Messer, K., White, M. M., Cowling, D. W., & Thomas, D. P. (2011). Prevalence of
heavy smoking in California and the United States, 1965-2007. JAMA: The Journal of the
119
American Medical Association, 305(11), 1106–1112.
http://doi.org/10.1001/jama.2011.334
Polo, A. J., & López, S. R. (2009). Culture, Context, and the Internalizing Distress of Mexican
American Youth. Journal of Clinical Child & Adolescent Psychology, 38(2), 273–285.
http://doi.org/10.1080/15374410802698370
Qualter, P., Brown, S. L., Rotenberg, K. J., Vanhalst, J., Harris, R. A., Goossens, L., … Munn, P.
(2013). Trajectories of loneliness during childhood and adolescence: Predictors and
health outcomes. Journal of Adolescence, 36(6), 1283–1293.
http://doi.org/10.1016/j.adolescence.2013.01.005
R Core Team. (2014). R: A language and environment for statistical computing. Vienna, Austria:
R Foundation for Statistical Computing. Retrieved from http://www.R-project.org/
Radloff, L. S. (1977). The CES-D Scale: A Self-Report Depression Scale for Research in the
General Population. Applied Psychological Measurement, 1(3), 385–401.
http://doi.org/10.1177/014662167700100306
Ramirez, R. R., & de la Cruz, G. P. (2002). The Hispanic population in the United States: March
2002 (Current Population Reports No. P20-545). Washington, D.C.: US Census Bureau.
Rawn, C. D., & Vohs, K. D. (2011). People Use Self-Control to Risk Personal Harm: An Intra-
Interpersonal Dilemma. Personality and Social Psychology Review, 15(3), 267–289.
http://doi.org/10.1177/1088868310381084
Ripley, R. M., Boitmanis, K., & Snijders, T. A. B. (2013). RSiena: Siena - Simulation
Investigation for Empirical Network Analysis (Version 1.1-232). Retrieved from
http://CRAN.R-project.org/package=RSiena
120
Ripley, R. M., Snijders, T. A. B., Boda, Z., Vörös, A., & Preciado, P. (2016, February 25).
Manual for RSiena. University of Oxford: Department of Statistics; Nuffield College.
Rokach, A. (2007). The effect of age and culture on the causes of loneliness. Social Behavior
and Personality: An International Journal, 35(2), 169–186.
http://doi.org/10.2224/sbp.2007.35.2.169
Rokach, A., & Neto, F. (2005). Age, culture, and the antecedents of loneliness. Social Behavior
and Personality: An International Journal, 33(5), 477–494.
http://doi.org/10.2224/sbp.2005.33.5.477
Rosseel, Y. (2012). lavaan: An R Package for Structural Equation Modeling. Journal of
Statistical Software, 48(2). http://doi.org/10.18637/jss.v048.i02
Russell, D. (1996). UCLA Loneliness Scale (Version 3): reliability, validity, and factor structure.
Journal of Personality Assessment, 66(1), 20–40.
http://doi.org/10.1207/s15327752jpa6601_2
Samet, J. M. (2013). Tobacco smoking: the leading cause of preventable disease worldwide.
Thoracic Surgery Clinics, 23(2), 103–112. http://doi.org/10.1016/j.thorsurg.2013.01.009
Sanders, R. A. (2013). Adolescent psychosocial, social, and cognitive development. Pediatrics in
Review / American Academy of Pediatrics, 34(8), 354-358-359.
http://doi.org/10.1542/pir.34-8-354
Schinka, K. C., van Dulmen, M. H. M., Mata, A. D., Bossarte, R., & Swahn, M. (2013).
Psychosocial predictors and outcomes of loneliness trajectories from childhood to early
adolescence. Journal of Adolescence, 36(6), 1251–1260.
http://doi.org/10.1016/j.adolescence.2013.08.002
121
Schwartz, S. J., Unger, J. B., Zamboanga, B. L., & Szapocznik, J. (2010). Rethinking the concept
of acculturation: Implications for theory and research. American Psychologist, 65(4),
237–251. http://doi.org/10.1037/a0019330
Segrin, C., Nevarez, N., Arroyo, A., & Harwood, J. (2012). Family of Origin Environment and
Adolescent Bullying Predict Young Adult Loneliness. The Journal of Psychology, 146(1–
2), 119–134. http://doi.org/10.1080/00223980.2011.555791
Seo, D.-C., & Huang, Y. (2012). Systematic review of social network analysis in adolescent
cigarette smoking behavior. The Journal of School Health, 82(1), 21–27.
http://doi.org/10.1111/j.1746-1561.2011.00663.x
Shakib, S., Mouttapa, M., Johnson, C. A., Ritt-Olson, A., Trinidad, D. R., Gallaher, P. E., &
Unger, J. B. (2003). Ethnic variation in parenting characteristics and adolescent smoking.
Journal of Adolescent Health, 33(2), 88–97. http://doi.org/10.1016/S1054-
139X(03)00140-X
Shankar, A., McMunn, A., Banks, J., & Steptoe, A. (2011). Loneliness, social isolation, and
behavioral and biological health indicators in older adults. Health Psychology, 30(4),
377–385. http://doi.org/10.1037/a0022826
Skriner, L. C., & Chu, B. C. (2014). Cross-ethnic measurement invariance of the SCARED and
CES-D in a youth sample. Psychological Assessment, 26(1), 332–337.
http://doi.org/10.1037/a0035092
Snijders, T. A. B. (2015, January). Goodness of fit testing in RSiena. Retrieved March 25, 2016,
from http://www.stats.ox.ac.uk/~snijders/siena/SienaGOF_s.pdf
122
Snijders, T. A. B., & Baerveldt, C. (2003). A multilevel network study of the effects of
delinquent behavior on friendship evolution. The Journal of Mathematical Sociology,
27(2–3), 123–151. http://doi.org/10.1080/00222500305892
Snijders, T. A. B., Lomi, A., & Torló, V. J. (2013). A model for the multiplex dynamics of two-
mode and one-mode networks, with an application to employment preference, friendship,
and advice. Social Networks, 35(2), 265–276. http://doi.org/10.1016/j.socnet.2012.05.005
Snijders, T. A. B., Steglich, C. E. G., & Schweinberger, M. (2007). Modeling the co-evolution of
networks and behavior. In K. V. Montfort, H. Oud, & A. Satorra (Eds.), Longitudinal
models in the behavioral and related sciences (pp. 41–71). Lawrence Erlbaum.
Snijders, T. A. B., van de Bunt, G. G., & Steglich, C. E. G. (2010). Introduction to stochastic
actor-based models for network dynamics. Social Networks, 32(1), 44–60.
http://doi.org/10.1016/j.socnet.2009.02.004
Sønderby, L. C., & Wagoner, B. (2013). Loneliness: An integrative approach. The Journal of
Integrated Social Sciences, 3(1), 1–29.
Soto, C., Unger, J. B., Ritt-Olson, A., Soto, D. W., Black, D. S., & Baezconde-Garbanati, L.
(2011). Cultural values associated with substance use among Hispanic adolescents in
southern California. Substance Use & Misuse, 46(10), 1223–1233.
http://doi.org/10.3109/10826084.2011.567366
Stata/IC 14.1. (2016). (Version 12.1) [Windows 64-bit]. College Station, TX: StataCorp LP.
Steglich, C., Snijders, T. A. B., & West, P. (2006). Applying SIENA. Methodology: European
Journal of Research Methods for the Behavioral and Social Sciences, 2(1), 48–56.
http://doi.org/10.1027/1614-2241.2.1.48
123
Steptoe, A., Owen, N., Kunz-Ebrecht, S. R., & Brydon, L. (2004). Loneliness and
neuroendocrine, cardiovascular, and inflammatory stress responses in middle-aged men
and women. Psychoneuroendocrinology, 29(5), 593–611. http://doi.org/10.1016/S0306-
4530(03)00086-6
Stewart-Knox, B. J., Sittlington, J., Rugkåsa, J., Harrisson, S., Treacy, M., & Abaunza, P. S.
(2005). Smoking and peer groups: Results from a longitudinal qualitative study of young
people in Northern Ireland. British Journal of Social Psychology, 44(3), 397–414.
http://doi.org/10.1348/014466604X18073
Stickley, A., Koyanagi, A., Koposov, R., Schwab-Stone, M., & Ruchkin, V. (2014). Loneliness
and health risk behaviours among Russian and U.S. adolescents: a cross-sectional study.
BMC Public Health, 14(1), 366. http://doi.org/10.1186/1471-2458-14-366
Stickley, A., Koyanagi, A., Roberts, B., Richardson, E., Abbott, P., Tumanov, S., & McKee, M.
(2013). Loneliness: its correlates and association with health behaviours and outcomes in
nine countries of the former Soviet Union. PloS One, 8(7), e67978.
http://doi.org/10.1371/journal.pone.0067978
Stokes, J. P. (1985). The relation of social network and individual difference variables to
loneliness. Journal of Personality and Social Psychology, 48(4), 981–990.
http://doi.org/10.1037/0022-3514.48.4.981
Sullivan, S., Schwartz, S. J., Prado, G., Shi Huang, Pantin, H., & Szapocznik, J. (2007). A
Bidimensional Model of Acculturation for Examining Differences in Family Functioning
and Behavior Problems in Hispanic Immigrant Adolescents. The Journal of Early
Adolescence, 27(4), 405–430. http://doi.org/10.1177/0272431607302939
124
Sussman, S., Dent, C. W., Stacy, A. W., Burciaga, C., Raynor, A., Turner, G. E., … Burton, D.
(1990). Peer-group association and adolescent tobacco use. Journal of Abnormal
Psychology, 99(4), 349–352.
Sussman, S., Pokhrel, P., Ashmore, R. D., & Brown, B. B. (2007). Adolescent peer group
identification and characteristics: A review of the literature. Addictive Behaviors, 32(8),
1602–1627. http://doi.org/10.1016/j.addbeh.2006.11.018
Templ, M., Alfons, A., Kowarik, A., & Prantner, B. (2015). VIM: Visualization and Imputation
of Missing Values (Version R package version 4.4.1). Retrieved from http://CRAN.R-
project.org/package=VIM
Thurston, R. C., & Kubzansky, L. D. (2009). Women, loneliness, and incident coronary heart
disease. Psychosomatic Medicine, 71(8), 836–842.
http://doi.org/10.1097/PSY.0b013e3181b40efc
Unger, J. B., Cruz, T. B., Rohrbach, L. A., Ribisl, K. M., Baezconde-Garbanti, L., Chen, X., …
Johnson, C. A. (2000). English language use as a risk factor for smoking initiation among
Hispanic and Asian American adolescents: Evidence for mediation by tobacco-related
beliefs and social norms. Health Psychology, 19(5), 403–410.
http://doi.org/10.1037/0278-6133.19.5.403
Urberg, K. A., Değirmencioğlu, S. M., & Pilgrim, C. (1997). Close friend and group influence on
adolescent cigarette smoking and alcohol use. Developmental Psychology, 33(5), 834–
844.
Urberg, K. A., Luo, Q., Pilgrim, C., & Degirmencioglu, S. M. (2003). A two-stage model of peer
influence in adolescent substance use: individual and relationship-specific differences in
125
susceptibility to influence. Addictive Behaviors, 28(7), 1243–1256.
http://doi.org/10.1016/S0306-4603(02)00256-3
US Census Bureau. (n.d.). Quickfacts. Retrieved November 21, 2013, from
http://quickfacts.census.gov/qfd/index.html
Valente, T. W. (2010). Social networks and health: models, methods, and applications. Oxford ;
New York: Oxford University Press.
Valente, T. W., Fujimoto, K., Soto, D., Ritt-Olson, A., & Unger, J. B. (2013). A Comparison of
Peer Influence Measures as Predictors of Smoking Among Predominately
Hispanic/Latino High School Adolescents. Journal of Adolescent Health, 52(3), 358–
364. http://doi.org/10.1016/j.jadohealth.2012.06.014
Valente, T. W., Fujimoto, K., Unger, J. B., Soto, D. W., & Meeker, D. (2013). Variations in
network boundary and type: A study of adolescent peer influences. Social Networks,
35(3), 309–316. http://doi.org/10.1016/j.socnet.2013.02.008
Valente, T. W., Gallaher, P., & Mouttapa, M. (2004). Using Social Networks to Understand and
Prevent Substance Use: A Transdisciplinary Perspective. Substance Use & Misuse,
39(10–12), 1685–1712. http://doi.org/10.1081/LSUM-200033210
Valente, T. W., Unger, J. B., & Johnson, C. A. (2005). Do popular students smoke? The
association between popularity and smoking among middle school students. The Journal
of Adolescent Health: Official Publication of the Society for Adolescent Medicine, 37(4),
323–329. http://doi.org/10.1016/j.jadohealth.2004.10.016
van der Loo, M. P. J. (2014). The stringdist package for approximate string matching. The R
Journal, 6(1), 111–122.
126
van Staden, W. (CW), & Coetzee, K. (2010). Conceptual relations between loneliness and
culture: Current Opinion in Psychiatry, 23(6), 524–529.
http://doi.org/10.1097/YCO.0b013e32833f2ff9
VanderWeele, T. J., Hawkley, L. C., Thisted, R. A., & Cacioppo, J. T. (2011). A marginal
structural model analysis for loneliness: Implications for intervention trials and clinical
practice. Journal of Consulting and Clinical Psychology, 79(2), 225–235.
http://doi.org/10.1037/a0022610
Wang, J., Zhu, R. (Juliet), & Shiv, B. (2012). The Lonely Consumer: Loner or Conformer?
Journal of Consumer Research, 38(6), 1116–1128. http://doi.org/10.1086/661552
Whisman, M. A., & McClelland, G. H. (2005). Designing, Testing, and Interpreting Interactions
and Moderator Effects in Family Research. Journal of Family Psychology, 19(1), 111–
120. http://doi.org/10.1037/0893-3200.19.1.111
Xiang, H., Wang, Z., Stallones, L., Yu, S., Gimbel, H. W., & Yang, P. (1999). Cigarette
Smoking among Medical College Students in Wuhan, People’s Republic of China.
Preventive Medicine, 29(3), 210–215. http://doi.org/10.1006/pmed.1999.0525
Abstract (if available)
Abstract
Past research has found that loneliness and social networks are both associated with cigarette smoking. However, social factors and loneliness have not been studied concurrently as correlates of smoking behavior, despite their potential association with each other. Furthermore, past studies have often conflated loneliness and depression or did not include depression as a covariate in analyses, although research suggests that loneliness and depression are correlated and are associated with smoking. Examining social networks, loneliness, and depression together as correlates of smoking may provide clarity into how they contribute to smoking behavior individually and through interaction. This dissertation examines the associations among loneliness, social networks, depression, and smoking in a sample of predominantly Hispanic/Latino adolescents enrolled in five schools located in one school district in Southern California, USA. ❧ Study one is a structural equation model of social networks, parental communication, loneliness, and depression as predictors of use of cigarettes, e-cigarettes, hookah, marijuana and alcohol. Study two uses logistic and linear regression to assess the associations of social networks, loneliness, and depression with accuracy in assessment of peer behavior, exposure to peer smoking and alcohol use, participant’s own use of cigarettes and alcohol, and participant’s use of cigarettes and alcohol with friends. Study three uses a stochastic actor-oriented model to analyze the dynamics among social networks, loneliness, and smoking. Studies two and three additionally consider potential interaction terms between peer influence and loneliness in the prediction of smoking. ❧ Contrary to past studies, we did not find that loneliness was associated with smoking and other substance use. Exposure from peers to cigarette and alcohol use was associated with adolescents’ own cigarette and alcohol use, but loneliness was not a moderator of this association. Furthermore, selection and influence processes were not found for cigarette use, despite findings that having a friend who smokes is associated with an adolescent’s smoking behavior. We found that centrality, parental communication, depression, and self-identifying as a stoner/druggie were associated with smoking and other substance use. Future research directions and implications of the findings are discussed.
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 network influences on depressive symptoms among Chinese adolescents
PDF
Social self-control and adolescent substance use
PDF
Normative and network influences on electronic cigarette use among adolescents
PDF
Contextualizing social network influences on substance use among high risk adolescents
PDF
Sociocultural stress, coping and substance use among Hispanic/Latino adolescents
PDF
Friendship network position on adolescent behaviors: an examination of a broker position and the likelihood of alcohol and cigarette use
PDF
Cultural risk and protective factors for tobacco use behaviors and depressive symptoms among American Indian adolescents in California
PDF
The role of social support in the relationship between adverse childhood experiences and addictive behaviors across adolescence and young adulthood
PDF
Evaluating social networks and impact of micro-influencers who promote e-cigarettes on social media
PDF
The role of depression symptoms on social information processing and tobacco use among adolescents
PDF
Exploring the role of peer influence, linguistic acculturation, and social networks in substance use
PDF
Energy drink consumption, substance use and attention-deficit/hyperactivity disorder among adolescents
PDF
Genetic variants and smoking progression in Chinese adolescents
PDF
The Internet activities, gratifications, and health consequences related to compulsive Internet use
PDF
Investigating factors that influence peer relationships and obesity during middle childhood
PDF
Psychosocial and behavioral ractors associated with emotional eating in adolescents
PDF
Relationship formation and information sharing to promote risky health behavior on social media
PDF
A theoretical framework and mixed-methods investigation of document status as a social determinant of emergency department utilization…
PDF
Anxiety symptoms and nicotine use among adolescents and young adults
Asset Metadata
Creator
Dyal, Stephanie Raye
(author)
Core Title
Adolescent social networks, smoking, and loneliness
School
Keck School of Medicine
Degree
Doctor of Philosophy
Degree Program
Preventive Medicine (Health Behavior Research)
Publication Date
06/20/2016
Defense Date
05/09/2016
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Adolescent,Cigarette,Loneliness,OAI-PMH Harvest,Smoking,social network,tobacco
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Valente, Thomas W. (
committee chair
), Bluthenthal, Ricky N. (
committee member
), Rohrbach, Louise Ann (
committee member
), Unger, Jennifer B. (
committee member
)
Creator Email
stepharp@usc.edu,stevieraye@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-254799
Unique identifier
UC11280321
Identifier
etd-DyalStepha-4454.pdf (filename),usctheses-c40-254799 (legacy record id)
Legacy Identifier
etd-DyalStepha-4454-0.pdf
Dmrecord
254799
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Dyal, Stephanie Raye
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
social network