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
/
Diffusion of a peer‐led suicide preventive intervention in secondary schools: strategies to increase effectiveness of peer‐led interventions
(USC Thesis Other)
Diffusion of a peer‐led suicide preventive intervention in secondary schools: strategies to increase effectiveness of peer‐led interventions
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
1 | Pickering
Diffusion of a Peer-Led Suicide Preventive Intervention in Secondary Schools:
Strategies to Increase Effectiveness of Peer-Led Interventions
A Dissertation Presented to the
Faculty of the Graduate School of the
University of Southern California
In Partial Fulfillment of the
Requirement for the Degree
Doctor of Philosophy
(Preventive Medicine – Health Behavior Research)
Keck School of Medicine
Department of Preventive Medicine
Institute for Health Promotion and Disease Prevention Research
Trevor A. Pickering, M.S.
May 2019
2 | Pickering
Acknowledgments
I would like to thank: Tom Valente for his mentorship and for inspiring me to pursue study in the field of
social network analysis; Peter Wyman for his immense guidance and support during my doctoral
program; the contributions of my committee including Kayla de la Haye, Jimi Huh, and Genevieve
Dunton; Daniel Chu for providing technical assistance and helping me become a better programmer;
George Vega Yon who helped me brainstorm ideas and created the network visualization package
“netplot” that I used; Tim Yuen for his motivation through the years; Jennifer Tsai for providing much
needed emotional support and constructive feedback; Jeff Yeh for just being there for me; and the friends,
staff, and faculty I’ve met at USC who have all influenced me for the better. I also thank my parents for
their love and for always expecting the best from me while giving me the freedom to pursue my passions.
3 | Pickering
Table of Contents
List of Figures ............................................................................................................................................... 5
List of Tables ................................................................................................................................................ 6
Introduction ................................................................................................................................................... 7
Research Aims ........................................................................................................................................... 7
Study 1 ................................................................................................................................................... 9
Study 2 ................................................................................................................................................. 11
Study 3 ................................................................................................................................................. 12
Background and Significance ................................................................................................................. 14
Adolescent Suicidality: A Direction for Public Health Research ....................................................... 14
Social Connectedness: A New Emphasis ............................................................................................ 16
Peer Leader Interventions ................................................................................................................... 17
The Sources of Strength Intervention .................................................................................................. 20
Overview of Proposed Studies ................................................................................................................ 22
Research Design and Methods ................................................................................................................ 25
Parent Study ........................................................................................................................................ 25
Data Collection ................................................................................................................................... 26
Sample ................................................................................................................................................. 27
Study 1 A Comparison of Peer Leader Selection Methods for Optimal Network Positioning .................. 30
Background ............................................................................................................................................. 30
Methods .................................................................................................................................................. 34
Results ..................................................................................................................................................... 42
Discussion ............................................................................................................................................... 48
Study 2 Training and Social Support Influence Diffusion in A Peer-Led School-Based Intervention ...... 54
Background ............................................................................................................................................. 54
Methods .................................................................................................................................................. 57
Results ..................................................................................................................................................... 62
Discussion ............................................................................................................................................... 73
Study 3 Network Structure and Peer Leader Seeds Affect Diffusion of a Suicide Preventive Intervention
in High Schools: A Simulation Study ......................................................................................................... 79
4 | Pickering
Background ............................................................................................................................................. 79
Methods .................................................................................................................................................. 83
Results ..................................................................................................................................................... 92
Discussion ............................................................................................................................................. 104
Conclusions ............................................................................................................................................... 110
Literature Cited ......................................................................................................................................... 117
Appendix A. A list of commonly-used terms in the manuscript ............................................................... 126
5 | Pickering
List of Figures
Figure 1. Sample school network from Sources of Strength. ..................................................................... 17
Figure 2. Conceptual model of the determinants of intervention diffusion. ............................................... 23
Figure 3. Peer leader characteristics hypothesized to promote diffusion. ................................................... 24
Figure 4. Sources of Strength intervention evaluation study design and measurement times. ................... 26
Figure 5. Conceptaul model of exposure modalities in Sources of Strength. ............................................. 29
Figure 6. Percent of peer leaders by school size, with the total number of peer leaders reflected next to
data points. .................................................................................................................................................. 35
Figure 7. Distribution of at-risk students in one sample network ............................................................... 40
Figure 8. Distribution of the proportion of at-risk students across schools, by risk type............................ 41
Figure 9. Network graph of peer leaders selected using various methods in a sample school ................... 53
Figure 10. Conceptual model for Study 2. .................................................................................................. 56
Figure 11. Hypothetical 1-step friendship network around a peer leader ................................................... 61
Figure 12. Distribution of the sizes of peer leaders' 1- and 2-step egocentric networks. ............................ 63
Figure 13. Relationship between having additional peer leader friends and exposure to four Sources of
Strength modalities ..................................................................................................................................... 72
Figure 14. Graphs of two friendship networks from the Sources of Strength intervention. ....................... 81
Figure 15. Graph of one iteration of diffusion simulation in one school. .................................................. 90
Figure 16. Correlation between observed exposure at T9 and average simulated exposure at T9 in 20
schools. ....................................................................................................................................................... 92
Figure 17. Simulated intervention diffusion over 9 time points in 20 schools, depending on peer leader
selection method. ........................................................................................................................................ 93
Figure 18. Simulated exposure to the intervention at time 9 and time 4 for several selection methods. .... 95
Figure 19. Comparion of two schools where sociometric selection had the least and greatest impact on
diffusion .................................................................................................................................................... 103
6 | Pickering
List of Tables
Table 1. Demographic characteristics of all students at T1 assessment,..................................................... 27
Table 2. Methods of peer leader identification with theoretical rationale and identification techniques. .. 32
Table 3. Concordance among peer leader selection methods. .................................................................... 43
Table 4. Metrics of the 459 peer leaders chosen by various methods ......................................................... 46
Table 5. Relationship between selection method concordance and exposure to Sources of Strength across
four modalities in 20 schools ...................................................................................................................... 47
Table 6. Intervention diffusion to peer leaders’ local social networks as a function of demographic
characteristics and social support. ............................................................................................................... 65
Table 7. Relationship between number of peer leader friends and 1- and 2-step egocentric network
exposure to intervention modalities. ........................................................................................................... 73
Table 8. Steps involved in each iteration of the diffusion simulation. ........................................................ 89
Table 9. Regression of selection method and network variables on intervention exposure after 9 months,
in four subpopulations ................................................................................................................................. 97
7 | Pickering
Introduction
Research Aims
Despite efforts at the individual and population level, adolescent rates of suicide in the
United States have continued to rise over the past decade (Centers for Disease Control and
Prevention, 2017). Public health experts have recommended several population-focused areas,
including increasing social connectedness, as future strategic directions to reduce rates of suicide
(U.S. Department of Health and Human Services, 2012). While several mechanisms may be
involved in improving mental health outcomes, suicide preventive interventions that increase
social connectedness at schools have already shown promise in ameliorating correlates of
suicidality—improving perceptions of adult support and acceptability of help-seeking behaviors
(Wyman et al., 2010).
The Sources of Strength intervention is the first suicide preventive intervention to utilize
students to diffuse intervention messaging throughout schools. In this intervention “peer leaders”
from diverse social groups and differing demographic backgrounds, including students already
at-risk of suicidality, are nominated by adults at school to receive comprehensive training. These
leaders then act as champions of the intervention and lead messaging activities under adult
supervision with the intent of promoting eight protective “strengths” that may result in
connecting students with adults at school and modifying norms around seeking help for mental
illness like suicide (LoMurray, 2005).
The proposed effectiveness of this intervention stems from diffusion of innovations
theory, which posits that new information within a network spreads through interpersonal
communication (Rogers, 2003), and that the use of opinion leaders to spread information results
in faster diffusion (Rogers, 2010; Valente & Davis, 1999). Aside from being a more cost-
effective way of delivering the intervention (e.g., only training a subset of the population vs.
8 | Pickering
delivering the training to the entire school), the use of peer leaders and peer-to-peer
communication is essential to the successful dissemination and adoption of the intervention. The
Sources of Strength program has already seen success in raising awareness of the program
through presentations, posters, videos, and activities. Research shows that interventions like
these may spread through peer-to-peer communication, and that closer proximity to a peer leader
in one’s social network has a stark effect on likelihood of intervention exposure (Christakis &
Fowler, 2007, 2008; Li, Weeks, Borgatti, Clair, & Dickson-Gomez, 2012; Pickering et al., 2018;
Valente, Hoffman, Ritt-Olson, Lichtman, & Johnson, 2003).
While there is evidence of social influence on the spread of the intervention through these
schools, the nature this diffusion has not yet been studied in detail. As such, optimization
methods, such as methods to select peer leaders and factors that contribute to successful
diffusion, have not been examined. As part of data collection in Sources of Strength, students at
school were asked to name students that they consider their friends, and students they consider
leaders in the school. This information (the network census, Figure 1) can be used to objectively
identify individuals in particularly connected or popular positions—who could possibly diffuse
the intervention more effectively than those chosen with the current selection method. Diffusion
of innovations theory suggests additional characteristics of students and their friends, such as
individuals who are more homophilous (i.e., similar) to their peers, may also influence diffusion
of the intervention to peers (Rogers, 2010).
This dissertation includes three studies that use social network information to explore
improving diffusion through these schools. Broadly, this is examined through 1) optimal
selection of peer leaders, 2) an examination of the effect of individual- and school-level peer
leader support, and 3) simulating diffusion through the network to explore which network-level
9 | Pickering
characteristics affect overall diffusion and diffusion to at-risk individuals. Addressing these
research questions will aid the development of protocol that can produce tailored, effective
interventions that can be translated easily across health behaviors. These study aims and
hypotheses are detailed as follows.
Study 1
Peer leaders in the Sources of Strength are currently chosen after careful consideration by
adults (e.g., teachers, counselors) at school. However, this method does not take into account the
rich sociometric information provided by the network census, collected as part of the Sources of
Strength evaluation study (Wyman et al., 2010). Successful intervention diffusion is expected to
be a function of having peer leader “seeds” in the network who are 1) respected and influential
opinion leaders, 2) in strategic network positions to diffuse the intervention to the most people
possible, and to at-risk students, and 3) representative of the school population, as adoption of
new information spreads more easily between individuals who are more alike.
We will compare sociometric and demographic characteristics among current adult-
selected peer leaders and hypothetical peer leaders chosen by various selection methods to
examine which qualities the current selection methods are capturing. Then, we will propose a
selection method that incorporates each of these characteristics to maximize diffusion potential.
The peer leader selection methods are as follows:
Adult selection. Peer leaders chosen by adults at school (empirical comparison group).
Opinion leaders. Individuals with the most nominations on the student survey question
asking who students think the respected leaders are at school.
High-centrality students. Individuals with the highest values of centrality for: degree,
closeness, betweenness, and coreness.
10 | Pickering
Key players. Individuals selected by the Key Players algorithm, a procedure designed to
select an optimum small set of the population to use as seeds for the diffusion of information,
attitudes, or practices (Borgatti, 2006).
Representativeness. Individuals selected to match the demographic distribution of the
school.
The research aims of this study are:
1.1. Assess which selection method produces peer leaders with the highest social network
centrality metrics. We hypothesize that, by design, peer leaders chosen by maximizing
network metrics will have the highest value of these metrics. We also hypothesized
that, as adults do not have access to information about the network census, this
selection method will systematically produce peer leaders with the lowest centrality
measures.
1.2. Assess which selection method produces peer leaders who are closest to those at-risk
in the network (i.e., suicidal students, peripheral students, and students isolated from
adults). We hypothesize that peer leaders chosen solely by maximizing centrality
measures will be more central in the network and, therefore, further from at-risk
individuals.
1.3. Assess which selection method produces peer leaders who with the least amount of
clustering. We hypothesize that peer leaders chosen by maximizing centrality
measures will have the most clustering, and by design peer leaders chosen by the Key
Players algorithm will have the least amount of clustering.
1.4. Assess which selection method produces peer leaders who best represent the
demographics of the school. Previous studies have found that opinion leaders are
11 | Pickering
more likely to be male, but that adult-selected peer educators are more likely to be
female; thus, this aim will be exploratory.
1.5. Identify which selection method best aligns with the following criteria: 1) individuals
who are considered opinion leaders in the school, 2) individuals in positions with the
greatest potential reach to at-risk students, 3) individuals with the greatest amount of
demographic representation, and 4) individuals with the least amount of clustering,
which leads to redundancy in the individuals able to be reached. If no such method
clearly meets all criteria, we propose a hybrid selection method that will take these
into account.
Study 2
It is currently unclear how peer leaders’ diffusion potential is affected by their support
systems during the implementation of the intervention. This study will examine school-level and
individual-level characteristics to determine the factors that contribute most to peer leaders’
diffusion after one school year.
2.1. Examine the effect of formal support and training on intervention diffusion. We
hypothesize that peer leaders who attend more meetings will have greater
intervention diffusion in their ego networks (i.e., in those to whom they are
connected).
2.2. Examine the effect of adult support on intervention diffusion. We hypothesize that
peer leaders who name more trusted adults and who feel supported by adults at
school will have greater intervention diffusion in their ego networks.
12 | Pickering
2.3. Examine the effect of peer support on intervention diffusion. Though there is some
support for this effect, is not clear whether undergoing peer leader training with a
friend increases intervention diffusion in the peer leader’s ego network.
2.4. Examine the additional effect of school-level support on intervention diffusion. We
include aggregated school-level measures of support, and hypothesize that students in
schools with more support will be more have greater intervention diffusion, above
and beyond the effect of their individual training and support factors.
Study 3
While there is some evidence that school social network structure influences diffusion
(Rulison, Feinberg, Gest, & Osgood, 2015), the effect of empirical school network structures on
intervention diffusion has not been studied. It is also unknown whether these network structures
affect diffusion similarly with different peer leader seeds, or the extent to which these structures
facilitate diffusion to at-risk students (i.e., those with suidical ideation or attempt, those who are
more peripheral, and those who are isolated from adults) in the network. This study will use a
novel network diffusion simulation approach to simulate diffusion of the intervention within
each of these school networks.
3.1. Examine how network structure affects simulated network diffusion in all schools.
We predict that networks that have more cliques and are more modular will have
slower diffusion (Choi, Kim, & Lee, 2010, p. 20), that density will generally increase
diffusion of the intervention messaging activities (Brown et al., 2012), and that more
centralized networks will have slower diffusion (Valente, 2010)
3.2. Examine how peer leader selection affects simulated network diffusion overall in all
schools. We hypothesize that the effect of peer leader selection on diffusion will
13 | Pickering
depend on network structure. For example, selecting central nodes in a centralized
network will have a greater effect on diffusion than selecting central nodes in a
decentralized network (Valente, 2012).
3.3. Examine how peer leader selection affects simulated network diffusion to at-risk
(i.e., peripheral, suicidal, isolated from adults) nodes in all schools.
14 | Pickering
Background and Significance
Adolescent Suicidality: A Direction for Public Health Research
Traditional school-based approaches to reducing adolescent suicide have focused on
treating students at the individual level. This includes programs attempting to expand the ability
to recognize and/or screen suicidal individuals, refer students to appropriate treatment, and raise
awareness through school-based educational programs. The rate of suicide in adolescents has
risen starkly over the past decade, though, and suicide is the second leading cause of death
among individuals 15 to 34 years of age. Between 2007 and 2015, the suicide rate for girls 15 to
19 years old doubled and the rate for boys increased by 30% (Centers for Disease Control and
Prevention, 2017). This implies that new strategies are warranted as traditional ways of
addressing suicide and mental illness may not be effective alone.
Screening and referral. Several studies have focused on screening adolescents for
suicidal thoughts and behaviors and subsequently referring them to appropriate treatment
(Aseltine et al., 2007; Gould et al., 2005; Kaess et al., 2014; McGuire & Flynn, 2003; Valla,
Bergeron, & Smolla, 2000). Adolescent screenings typically take place in a school (Gould et al.,
2009) or primary care setting (Gardner et al., 2010). Treatment options after referral include but
are not limited to medication (March et al., 2007), cognitive behavioral therapy (Robinson,
Hetrick, & Martin, 2011; Taylor et al., 2011), and dialectical behavioral therapy (Burns et al.,
2005).
There is limited evidence that screening programs are effective in isolation, with their
success primarily being measured through identification and referral rates (Gould et al., 2009;
Pena & Caine, 2006). Other studies show that these programs have low predictive validity
(Shaffer et al., 2004), do not change attitudes about seeking help (Gould et al., 2009), and may
15 | Pickering
lead to prescribing unnecessary medication (Lenzer, 2012). Subsequently, some programs
focusing on screening in schools have since been terminated.
Gatekeeper training. Gatekeeper training seeks to provide individuals in the community
the skills to identify students at risk for suicide, question students about suicide, and provide
referrals to treatment. Recent gatekeeper interventions for adolescents have focused on training
school staff (Wyman et al., 2008) and peer counselors (Reis & Cornell, 2008), and are
hypothesized to work through identification of suicidal risk factors in youth and timely referrals
(Goldston et al., 2010). While these programs may be effective in increasing gatekeepers’
knowledge about suicide and number of potential referrals, and lead to a modest decrease in
population-level suicide rates (Walrath et al., 2015), there is no evidence that these programs
change the social stigma of mental illness or improve help-seeking attitudes of suicidal students
(Tompkins, Witt, & Abraibesh, 2010).
Educational programs. Educational programs at the school level have generally focused
on individual-level correlates of suicidality. Some studies attempt to foster abilities to cope with
depression, anxiety, and suicidality (Brent & Brown, 2015); improving knowledge, attitudes
toward suicide, and self-identification skills (Cusimano & Sameem, 2011); increase help-seeking
behaviors (Freedenthal, 2010); or a combination of these methods (King, Strunk, & Sorter,
2011). The effectiveness of these programs has been mixed. Some randomized controlled trials
that focused on mental health literacy, suicide risk awareness, and skills training showed a
reduction in rates of suicide attempt and ideation (Aseltine et al., 2007; Wasserman et al., 2015),
while some prospective cohort studies have shown mixed results (Ciffone, 2007; Freedenthal,
2010; Hooven, Herting, & Snedker, 2010). Although these programs generally teach students
how to recognize signs of suicidality in peers and connect at-risk individuals to adults who can
16 | Pickering
help, none specifically include a component to increase social cohesion at schools to adults and
other students.
Social Connectedness: A New Emphasis
The CDC has identified social connectedness—the feeling of closeness to others and
fulfilling relationships—as an effective way to prevent suicidal behavior (Centers for Disease
Control and Prevention, 2012). This is because some of the tallest barriers to treating mental
illness occur at the community level; for instance, the systematic stigmatization of mental health
problems and lack of a social support network that promotes seeking care (Corrigan, Druss, &
Perlick, 2014). Interventions that increase social connectedness may be successful for a variety
of reasons: increasing social integration and support (Centers for Disease Control and
Prevention, 2009), increasing behavioral monitoring by peers (Bearman & Moody, 2004),
increasing exposure to social influences that encourage healthy coping and (Centers for Disease
Control and Prevention, 2009), and fostering an environment with shared understanding that
mental health wellness and recovery are priorities (U.S. Department of Health and Human
Services, 2012). Improving the amount and quality of social ties also addresses empirical
disparities seen in at-risk students. Previous studies have found that suicidal adolescents reported
more friendship ties to other suicidal youth (Bearman & Moody, 2004), but that they generally
have fewer friends and are located in less cohesive friendship groups (Wyman et al., 2018). This
effect is illustrated in Figure 1.
17 | Pickering
Figure 1. Sample school network from Sources of Strength with individuals (circles) connected
by friendship nominations (lines). Students in less cohesive friendship groups and who cluster
with other suicidal students (A) are more likely to experience suicidal ideation (yellow) and
attempts (red) than individuals in more cohesive friendship groups (B).
Peer Leader Interventions
Several public health interventions have involved a peer-led component when trying to
deliver health education. Peer leaders are a central component of these programs—agents in the
community who receive intervention training and implement a range of instructional or behavior
change interventions in the community. This approach has shown to be more effective than
education training alone in adolescents across a range of health behaviors such as HIV/AIDS
18 | Pickering
prevention (Menacho, Galea, & Young, 2015; Pearlman, Camberg, Wallace, Symons, & Finison,
2002), nutrition promotion (Story, Lytle, Birnbaum, & Perry, 2002), substance use prevention
(Black, Tobler, & Sciacca, 1998; Bloor et al., 1999; Komro et al., 1996; Komro, Perry, Veblen-
Mortenson, Williams, & Roel, 1999; Perry et al., 2002), violence prevention (Orpinas et al.,
2000; Wiist, Jackson, & Jackson, 1996), smoking cessation (Starkey, Audrey, Holliday, Moore,
& Campbell, 2009), bicycle helmet wearing (Hall, Cross, Howat, Stevenson, & Shaw, 2004) and,
recently, to suicide prevention (Wyman et al., 2010).
It is hypothesized that peer leaders are more effective messengers of educational
interventions in schools (compared to traditional intervention approaches) for a variety of
reasons (Holliday, 2006), several of which align with the goals of the Sources of Strength
intervention.
Credibility. Peers are more credible sources of information, they can communicate with
peers in the manner they speak and media they use, and they are empathetic to those educated.
(Forrest, Strange, & Oakley, 2002; Frankham, 1998; Milburn, 1995).
Acceptability. Adolescents prefer peers to deliver health education (Hamdan, Story,
French, Fulkerson, & Nelson, 2005) as they feel the peer leaders are more understanding
(Harden, Oakley, & Oliver, 2001) and believe their conversations are more confidential
(Backett-Milburn & Wilson, 2000), leading to a more relaxed environment.
Efficacy. Interventions that employee peer leaders and student-to-student exercises are
generally more effective than interventions led by adults (Black et al., 1998; Cuijpers, 2002;
Tobler & Stratton, 1997), but peer leaders may be better suited at changing social determinants
of health behavior (e.g., behavioral norms, attitudes) than at delivering information (Mellanby,
Rees, & Tripp, 2000).
19 | Pickering
Alternative benefits. Peer leader educators have reported an increase in their own
leadership skills, self-esteem, and confidence (Pearlman et al., 2002), which additionally helps
these students in their peer leader role (Cowie & Olafsson, 2000).
Role modeling and sustainability. Peer educators can act as role models for others’
behavior (Valente & Davis, 1999) and reinforce the intervention message through ongoing
informal contact in the school setting (Turner & Shepherd, 1999). Since these educators come
from the community, they are more sustainable (Kelly et al., 2006).
Accessibility. Peer-led interventions can reach a larger set of the population compared to
traditional structured interventions as they can use informal routes of communication (Hunter,
Ward, & Power, 1997; Power, Jones, Kearns, Ward, & Perera, 1995). This allows information to
reach students who may be less engaged at school.
Cost-effectiveness. Peer educators generally act as unpaid volunteers in the community
and improve the reach of the intervention (Wilton, Keeble, Doyal, & Walsh, 1995), decreasing
the amount of financial resources necessary to achieve the intervention goals.
The use of peer leaders to deliver messaging activities in Sources of Strength is especially
beneficial as they are cost-effective, are better at changing norms and attitudes around suicide
than general education programs, and are able to reach students that are most at-risk for suicidal
thoughts and behaviors.
Nonetheless, research on the mechanisms of how peer-led programs work is limited. A
growing body of evidence shows that proximity to trained peer leaders increases the likelihood
of intervention diffusion. For example, in a study on family planning in Madagascar, individuals
who had a link to a community-based distributor had increased family planning contraceptive
knowledge (Stoebenau & Valente, 2003). Individuals in a community of drug users who were
20 | Pickering
closer to peer health advocates were more likely to be exposed to and adopt strategies in a drug
avoidance intervention (Li et al., 2012). Students in the Sources of Strength study were more
likely to be exposed (across a range of modalities from peer communication to structured
activities) to the intervention the closer they were to trained peer leaders (Pickering et al., 2018).
Though a body of literature describes the way information diffuses through networks, research is
still lacking on how to optimize diffusion through various intervention schools in order to reach
the most students—and the most vulnerable students.
The Sources of Strength Intervention
Sources of Strength (LoMurray, 2005) is a suicide prevention intervention designed to
utilize peer networks to change unhealthy coping norms, improve help-seeking, and promote
connections to caring peers and adults at school. The intervention is based on a prevention
approach designed to build socioecological protective influences across the whole student body
at each school. By modifying the norms propagated through communication within peer group,
peer leaders can change perceptions of what may be considered typical behavior (i.e., descriptive
norms) and of the social consequences for positive coping behaviors (i.e., injunctive norms)
(Petty & Cacioppo, 1986; Pisani et al., 2012).
Mechanisms of change. All hypothesized mechanisms of change come from diffusion of
innovations theory (Rogers, 2010; Valente, 1996), which emphasizes the importance of one’s
personal social network in being exposed to new information and adopting new norms and
behaviors. These mechanisms include:
Increased ties to caring adults at school: Peer leaders encourage others to name and
engage with trusted adults (whom they feel they can go to for help) at school. Suicidal
21 | Pickering
thoughts and behavior are lower, and psychological well-being is higher, among
individuals with more, stronger, bonds to others (Whitlock, Wyman, & Moore, 2014).
Increased help-seeking attitudes and behavior: Peer leaders reinforce an expectation that
friends ask adults for help for themselves and for suicidal friends. Adolescents with more
positive help-seeking attitudes are less likely to experience suicidal thoughts and
behaviors (Pisani et al., 2012).
Increased coping: Peer leaders encourage others to identify and use interpersonal and
formal coping resources. Adolescents with better, more adaptive coping strategies are
less likely to experience suicidal thoughts and behaviors (Wyman et al., 2010).
Peer leader training. Sources of Strength recruits and trains key students at school along
with school staff who serve as advisors. These “peer leaders” are selected by adults at school to
represent demographically-diverse social cliques, including at-risk adolescents. Through their
training they learn a model of health and resilience that emphasizes developing healthy social
bonds (e.g., adults at school, family support, positive friends) and resources to manage adversity
(e.g., healthy activities, medical and mental health resources for suicidal individuals). Through
the mentorship of adults at school, these peer leaders conduct messaging activities to disseminate
the “sources of strength” to the school population. Peer leaders are also trained to conduct well-
defined messaging activities at school which may change the norms and behaviors of their peers.
Sources of Strength evaluation. Prior assessments of Sources of Strength have shown it to
be effective. In a cluster randomized trial of 18 schools implementing Sources of Strength, norms
about help-seeking and perceptions of adult support were improved among students in the three
months after peer leaders began to implement prevention messaging activities (Wyman et al.,
2010). Proximity to peer leaders, and connections to trusted adults, were shown to be vital to the
22 | Pickering
dissemination of several prevention messaging activities (i.e., presentations, posters, peer
communication, and activities), thus reinforcing the hypothesized effect of peer-to-peer
communication and adult support (Pickering et al., 2018).
Overview of Proposed Studies
The use of peer leaders may be ideal for the dissemination of the Sources of Strength
intervention, but several factors can influence the diffusion of this program’s messaging and
activities through the schools. While the diffusion of information and adoption of behavior has
been previously studied in communities, Sources of Strength provides a unique setting in which
to study diffusion. This is because 1) there is little information about how to comprehensively
use sociometric and demographic information to inform the peer leader selection process in an
ecological setting, 2) the effect of social support (via formal training and informal interactions)
on the effectiveness peer leaders not been studied, and 3) counterfactual simulations have not
been performed to study program efficacy had these other selection methods been employed
instead, nor has there been an evaluation on how each of these methods might reach at-risk
students in these schools. Figure 2 provides a conceptual diagram of the contribution of each of
these dissertation studies.
23 | Pickering
Figure 2. Conceptual model of the determinants of intervention diffusion. Contribution of each
study is indicated with the corresponding study number.
Peer leader selection. In the first study I evaluate the potential for social network
information to inform peer leader selection. I examine various alternative ways of selecting peer
leaders and compare the demographic and sociometric characteristics of these peer leaders
among selection methods. Many interventions have previously used trained students to deliver
intervention programs and messaging (“peer educators”, or “opinion leaders”). These leaders are
typically chosen haphazardly, by self-selection, or by trained intervention staff. It has been
shown, however, that diffusion of information and adoption of behaviors are greatest when
opinion leaders—highly regarded individuals in the community with the potential to influence
others—deliver the intervention (Valente & Davis, 1999). Other studies have examined network
24 | Pickering
position alone and found that optimal network positions may exist to promote diffusion (Kempe,
Kleinberg, & Tardos, 2003; Valente, 2012). Finally, diffusion of innovations theory suggests that
individuals who are similar to others (i.e., homophilous) are more likely to spread information to
their ilk (Rogers, 2010), a finding that has been replicated in several subsequent studies (Aral,
Muchnik, & Sundararajan, 2009; De Choudhury, Sundaram, John, Seligmann, & Kelliher, 2010).
This study comprehensively identifies the differences in peer leader selection sets produced
among selection methods and discuss how this selection may be related to further diffusion
(Figure 3).
Figure 3. Peer leader characteristics hypothesized to promote diffusion. It is unknown whether
these characteristics will co-occur among various selection methods.
Peer leader support. The second study examines the effect of peer leader support at the
individual and school level. While studies have examined the effect of student leaders on
25 | Pickering
intervention effectiveness, there is little information on the effect of support structures on their
effectiveness. These support systems include 1) formal training through the intervention
program, 2) adult mentorship, and 3) the effect of undergoing peer leader training with a friend. I
use multi-level models to examine the diffusion of the intervention through peer leaders’ ego
networks, using individual and school-level support structures as predictors. School-level support
is included to determine if there is an effect of being in a supportive school environment above
and beyond the impact of individual-level support.
Network effects. The last study draws upon information from studies 1 and 2 to model
intervention diffusion through these school networks. Previous studies have shown that network
structure can affect diffusion, but it is unknown how these network structures will affect
diffusion given the various peer leader selection methods identified in Study 1. This study will
use a custom-written program in R that simulates diffusion through networks. I use the various
peer selection methods and examine: 1) how differing network structures (e.g., size, density,
clustering) moderate the effect of peer leader selection on diffusion, 2) which peer leader
selection method provides the greatest reach to the entire student population, and 3) which peer
leader selection method allows for diffusion to at-risk (i.e., peripheral students, suicidal students,
or adult isolates).
Research Design and Methods
Parent Study
Data for these studies comes from a parent study on 40 high schools enrolled in a
randomized controlled trial testing a universal suicide prevention program, Sources of Strength
(LoMurray, 2005; Wyman et al., 2010). Schools were located in primarily rural, small town, or
micropolitan communities of New York (n=31) and North Dakota (n=9), based on Rural Urban
26 | Pickering
Commuting Area scores (WWAMI Rural Health Resource Center, 2004). Schools were selected
for enrollment in the Sources of Strength intervention based on two criteria: being underserved
by mental health services according to county mental health departments and/or school
administrators’ reports of barriers to accessing services, and self-identified need for suicide
prevention programming.
A school-based randomized “wait list” design was used to test the effectiveness of the
Sources of Strength program. Schools were randomly assigned to 1) the immediate condition
with training to begin in the fall of the first year over a two-year span and 2) the wait list
condition, where assessments were performed but Sources of Strength training began after a two
year wait period. Four longitudinal assessments were performed in the first two study years.
Figure 4. Sources of Strength intervention evaluation study design and measurement times.
Data Collection
Students were invited to enroll with parent permission to complete an assessment in the
fall following the initial enrollment of schools in the Sources of Strength trial. Subsequent
assessments were completed, regardless of experimental condition, in spring at the end of the
first school year, in fall of the second school year, and in spring of the second school year
(Figure 4). These assessments were performed on computers at school using a web-based format,
27 | Pickering
monitored by project staff. Research staff collected student assent/consent and the University of
Rochester IRB approved the study protocol.
Sample
The total student populations across the 40 schools ranged from 72 – 1,600. The number
of students participating per school ranged from 65 to 1,168, reflecting 62.8 – 98.6% of the
student population (M=84.0%). In the two largest schools a representative group was selected
using stratified (by grade) sampling to invite approximately one-half of the student body; 63.9%
and 75.1% of those invited participated leading to final school-wide representation of 26.5% and
44.5% participation, respectively. These schools were not used in our analysis. Demographic
characteristics of the study sample are presented in Table 1.
Table 1. Demographic characteristics of all students at T1 assessment, and characteristics of
students with suicide attempt and suicide ideation. *p<.05. +p<.10, for difference in proportions
between/among groups. †Reference group. Categories may not add to 100% due to missing data.
N (%)
Suicide Attempt
N (%)
Suicidal Ideation
N (%)
Total 5677 397 (7.0) 502 (8.8)
Sex Male
†
2874 (49.9) 139 (4.8) 172 (6.0)
Female 2803 (48.6) 278 (9.9)* 325 (11.7)*
Grade 9
th†
1486 (25.8) 121 (8.4)* 118 (8.0)
10
th
1501 (26.0) 115 (8.0) 135 (9.1)
11
th
1315 (22.8) 93 (7.3) 125 (9.6)
12
th
1306 (22.7) 81 (6.4) 112 (8.7)
Race Asian 133 (2.3) 9 (7.2) 14 (10.6)
Black/AA 588 (10.2) 47 (8.3) 34 (5.9)*
Am. Indian 270 (4.7) 26 (9.9)+ 17 (6.3)
White
†
4248 (73.7) 285 (6.9) 390 (9.5)
Other 408 (7.1) 40 (10.4)* 37 (9.1)
Ethnicity Hispanic
†
503 (8.7) 61 (12.9) 54 (10.9)
Non-Hisp. 5147 (89.3) 357 (7.2)* 443 (8.7)+
28 | Pickering
Network measures. Students were asked to name 1) up to seven of their closest friends at
school, 2) up to three students at school who are leaders and others listen to, and 3) up to seven
adults at school who they can trust and talk to about personal things. Students made these
nominations by typing in names, a method that produces fewer, closer relationships than
selecting friends from a roster of names (Valente et al., 2009). A network census of the school
was constructed using the friendship nominations (e.g., Figure 1). This allowed network metrics
to be computed for each student at school. Network metrics included: in-degree (number of
friendship nominations received), out-degree (number of friendship nominations made), in-
closeness (closeness to all other individuals based on incoming ties), and betweenness (the
number of times a path between any two individuals passes through an individual).
Intervention exposure. Intervention exposure was measured through several questions
that were later grouped into four exposure modalities across a continuum of peer engagement.
These modalities included:
(1) Exposure to a presentation consisted of answering “yes” to either: Have you seen a
presentation or assembly about… (a) strengths that help teens get through hard times?, or (b)
helping suicidal teens by getting adults involved?
(2) Exposure to posters or videos was assessed by answering “yes” to: Have you seen
posters or videos at school about strengths?
(3) Direct peer communication was measured by answering “yes” to either: Has a friend
or other student… (a) told you about Sources of Strength?, or (b) talked to you about using
strengths?
29 | Pickering
(4) Participation in an activity consisted of answering “yes” to either: (a) Have you
participated in a Sources of Strength activity such as adding your trusted adult to a poster?, or (b)
Has a friend or other student asked you to name adults you can go to for help?
While these exposures reflect actual activities performed by peer leaders in the
intervention, studying exposures along the continuum of peer engagement provides an
opportunity to see which messaging activities are most affected by peer leader selection, training,
and support.
This continuum of activities also corresponds to the Knowledge, Attitudes, and Practices
model (Coleman, Katz, & Menzel, 1966). In this model, originally designed to describe the
spread of a medical innovation among physicians, knowledge of the innovation spread quickest,
followed by attitudes toward using the innovation, then adoption of the innovation. This has an
analog in Sources of Strength, as peer-led activities range from simple knowledge of the
intervention to actual participation in intervention activities (Figure 5).
Figure 5. Exposure modalities in Sources of Strength represent increasing levels of peer
engagement and represent both knowledge of, and participation in, the intervention.
30 | Pickering
Study 1
A Comparison of Peer Leader Selection Methods for
Optimal Network Positioning
Background
It is well-accepted in network theory that selecting influential nodes (i.e. individuals) in a
network results in superior diffusion of information through that network (Valente, 2012). When
interventions attempt to use individuals within a network as change agents, the diffusion of the
intervention messaging and behavior can be accelerated by using respected, influential “opinion
leaders” within the community (Valente & Davis, 1999). It has been shown that, in peer-led
interventions, diffusion of an intervention was moderated by the recipients’ perception of
opinion leadership from the messenger (Iyengar, Van den Bulte, & Valente, 2011) and that
opinion leaders proved to be better able to diffuse innovations for reasons extending beyond just
their connections to multiple individuals (Iyengar, Van den Bulte, Eichert, & West, 2011).
Simulation studies have confirmed the powerful effect of opinion leaders on product adoption
not just because of their place in the network, because of their respected role within their
networks (Van Eck, Jager, & Leeflang, 2011).
Aside from contributing to fast intervention diffusion, there may be other specialized
reasons to use peers to deliver an adolescent intervention. Peer leaders/educators have been
found to be more credible at communicating the message of the intervention compared to adults
(Forrest et al., 2002; Frankham, 1998; Milburn, 1995) and can act as role models who persist
within the community after the intervention ends (Kelly et al., 2006).
31 | Pickering
From a functional perspective, the use of peer leaders can lead to a more cost-effective
intervention that has greater reach than traditional interventions. Peer leaders have access to
informal routes of communication (Hunter et al., 1997; Power et al., 1995) which can be
especially beneficial at reaching less connected and less engaged students at school. The use of
peer leaders allows a smaller set of the population to be trained, thus decreasing the financial
resources necessary to achieve program goals. This allows information to reach students who
may be less socially engaged at school.
A study by Valente and Pumpuang (Valente & Pumpuang, 2007) identified several ways
of selecting respected opinion leaders in networks. The most powerful of these, in the context of
a school intervention, is, when feasible, to collect sociometric information on the entire school
network. The ability to question students about who they consider to be friends, leaders, admired,
or respected (to name a few) in their school can provide valuable information for making
informed peer leader selection.
Peer-led interventions without access to this type of sociometric information have relied
on methods such as self-selection (Stephenson et al., 2004), staff selection (Phelps, Mellanby,
Crichton, & Tripp, 1994), or a combination of both (Miller & MacGilchrist, 1996). Recent work,
though, has focused on developing methods for network-informed selection methods in
situations where network information is limited. For example, an effective way of recruiting
community leaders is to select the friends of randomly selected seeds within the community and
use these friends instead as the community leaders. While this set of individuals will tend to have
more connections than randomly selected seeds and is useful when sociometric information is
not available, this method does not take advantage of information regarding the entire network
(Chami, Ahnert, Kabatereine, & Tukahebwa, 2017; Kim et al., 2015).
32 | Pickering
In a study such as Sources of Strength, though, the use of sociometric information to
inform peer leader selection is considered superior to uninformed selection as it provides more
power (Valente, Gallaher, & Mouttapa, 2004). The use of one sole algorithm for peer leader
identification, though, may ignore several facets that might operate on differing levels; a one-
size-fits-all approach may not be appropriate, and effectiveness may depend on network
structure. For example, selecting individuals with the most opinion leader nominations might
result in one particularly popular clique being identified, but this clique may not be able to reach
students on the social periphery. Similarly, using network information to identify individuals in
strategic network positions may result in peer leaders who are not as well-respected or perceived
as credible within their school.
Table 2. Methods of peer leader identification with theoretical rationale and identification
techniques.
Method Rationale Technique
Opinion Leaders Individuals are more likely to adopt
from others who are admired or
popular in the network
In-degree of opinion leader
network
Network Position Individuals who are strategically
located in the network can spread
information to those who might not
otherwise be reached
Degree
Closeness
Betweenness
Key Players
Representativeness Individuals are more likely to adopt
from others who are similar to them
Gender
Race/ethnicity
Based on characteristics that have identified to promote successful diffusion in the
literature, we identified three qualities that are expected to promote effective diffusion in the
school networks (Table 2): 1) peer leaders who are respected and influential opinion leaders, 2)
33 | Pickering
peer leaders in strategic network positions to diffuse the intervention to the most people possible,
and to at-risk students, and 3) peer leaders who are representative of the school population.
The specific aims of this study are as follows:
1.1.We compute the sociometric qualities of peer leaders under various selection
methods: opinion leaders, sociometric (i.e., position in the friendship network)
characteristics (degree, coreness, closeness, and betweenness) and key players. The
social network characteristics of leaders produced within each method are computed
to see the sociometric qualities of leaders produced by each method.
1.2. We evaluate the distance of peer leaders to at-risk students by determining the
distance of each at-risk student (stratified into 3 groups: suicidal students, peripheral
students, and students isolated from adults) to the nearest peer leader. Then the mean
and standard error of these distances are computed.
1.3.We address the clustering produced by each selection method by calculating the
number of peer leaders each individual peer leader is connected to, then take the
mean and standard error of these values.
1.4. We evaluate whether peer leaders represent school demographics by comparing the
demographic characteristics of peer leaders to the demographic characteristics of the
student body at each school.
1.5. For each school we compute the concordance of the adult-selected peer leaders with
hypothetical peer leaders chosen by each method. We use this information to see if
concordance with a particular sociometric peer leader selection method is related to
actual rates of diffusion in the program.
34 | Pickering
Methods
The 20 networks used in this study were a part of an evaluation trial of the Sources of
Strength program, involving a total of 40 schools. Schools in this study are located in rural, small
town, and micropolitan areas of New York and North Dakota and were recruited from regions
with past five-year youth suicide rates above the state averages. Participants were followed for
four waves of data collection spanning two years, but as this study addresses initial peer leader
selection methods, only baseline network data was used. Of the 40 schools in the parent study,
the 20 randomized to receive the Sources of Strength immediately are included in this study.
This includes 16 schools in New York and 4 in North Dakota, ranging in size from 63-1,207
students (M=366). Two schools served American Indian reservations.
Students were recruited in two phases: 1) at the beginning of the school year for the
school-wide Sources of Strength evaluation and 2) immediately following the first phase for the
selection of peer leaders. All students at school were invited to enroll in the evaluation study by
completing web-based assessments during the fall and spring over two years. Information letters
were sent to parents that allowed the option to decline their child’s participation. The University
of Rochester IRB approved the study protocol, and research personnel collected opt-out forms
and verbal assent with eligible students. Students were given information about how to access
help and support for themselves and other students if necessary.
Peer Leader Selection
Peer leader selection was preceded by the training of adult advisors at school and baseline
assessments. Recruitment was performed following standardized procedures by distributing
nomination forms to staff members at school which asked for nominations of up to 6 students
whose “voices are heard” by other students. Nominations were reviewed with an attempt to
35 | Pickering
select individuals who reflected diverse groups within their school, with a target of 5-10% of the
school population selected as a peer leader. The size of peer leader teams, therefore, was
dependent on school size and staff selection. A total of 959 students were invited (19 to 86 per
school), with 798 (83.2%) enrolling with parent permission and youth assent. Of these, 459 (9-45
per school) were retained as active peer leaders.
Figure 6. Percent of peer leaders by school size, with the total number of peer leaders reflected
next to data points.
Alternate Peer Leader Selection Criteria
The number of adult-selected peer leaders (APL) varied by school (Figure 6). For any
given school with n peer leaders, n alternative peer leaders were chosen by the following
36 | Pickering
selection methods. Any time a ranked method produced a tie, students were selected randomly to
break the tie.
Opinion leaders. Students were asked to name up to three students in school who they
considered to be student leaders who others listen to. These nominations were summed to
produce the total number of nominations per student (opinion leader in-degree), information that
was later used to identify opinion leaders at school. The top n opinion leaders at each school
were selected as opinion peer leaders (OPL).
Friendship network. Students were asked to name up to seven students in school who are
their closest friends. These nominations were used to produce several individual-level network
variables. This information included: (a) In-degree: the number of friendship nominations
received; (b) Coreness: for each student, the k-core is the maximal subgraph in which each vertex
has degree k, with larger values indicating a position in a more highly interconnected group of
friends; (c) Closeness: the reciprocal sum of distances to each other student in the network,
indicating a central proximity to all other students; and (d) Betweenness: the number of times an
individual is in the shortest path connecting any two nodes. The iGraph package in R (Csardi &
Nepusz, 2006) was used to compute all individual-level friendship network variables. The top n
students in each metric were selected as sociometric peer leaders (SPL).
Key players. The key players algorithm (KPP-POS) is a technique that identifies key
players for the purpose of optimally diffusing information through a network (Borgatti, 2006).
Borgatti notes one practical implementation of this algorithm would be to select a small set of a
population as seeds to diffuse practices or attitudes that promote health, such as in a public health
context—an application that is analogous to using peer leader seeds in the Sources of Strength
intervention. The approach selects a set of maximally connected individuals who will tend to be
37 | Pickering
spaced equally throughout the network. An advantage of this approach is that it addresses the
“redundancy problem,” the tendency of highly central nodes to be structurally equivalent and
therefore be connected to the same individuals. The key players algorithm was performed using
the InfluenceR package in R (Jacobs, Khanna, & Madduri, 2016) to identify n individuals in each
school as key players (KPL).
Hybrid methods. Three hybrid methods of peer leader selection were created. In all cases,
representative samples of the population were taken by stratifying the school population by
ethnicity, gender, and grade level and choosing a proportional number of peer leaders within that
stratum, rounded down. Then, the key players algorithm was used to bring the total number of
peer leaders up to n per school. These custom peer leaders (CPL) were selected by the following
algorithms:
1) Influence-weighted: the students with the highest opinion leader in-degree and
friendship in-degree were chosen within each stratum. (CPL1)
2) Centrally-weighted: the students with the highest closeness and betweenness were
chosen within each stratum. (CPL3)
3) Structurally-weighted: peer leaders were initially chosen by highest opinion leader
and friendship in-degree but were constrained to a maximum of 2 individuals chosen
per stratum. This resulted in a greater proportion of peer leaders being chosen through
the key players algorithm. (CPL5)
Survey Variables
Demographics. The baseline survey administered to all students collected information on
student sex, ethnicity (white vs. nonwhite), and grade level.
38 | Pickering
Suicidal thoughts and behaviors. The Youth Risk Behavior Survey (Centers for Disease
Control and Prevention, 2010) was used to measure suicidal thoughts and behaviors. Each
student was asked whether in the preceding 12 months he/she had: seriously considered suicide;
planned suicide; made one or more suicide attempts; or made an attempt that resulted in injury
requiring medical treatment. Students were assigned to one of three categories indicating high to
low suicide risk: suicide attempt (SA) with or without injury, seriously considered suicide (SI)
without attempt, and no STB (NS).
Intervention exposure. Exposure to the Sources of Strength intervention was measured at
the end of the first school year and categorized into four different dichotomous exposure
modalities corresponding to various levels of engagement:
(1) Exposure to a presentation or assembly consisted of answering “yes” to either: Have
you seen a presentation or assembly about… (a) strengths that help teens get through hard
times?, or (b) helping suicidal teens by getting adults involved?
(2) Exposure to posters or videos was assessed by answering “yes” to: Have you seen
posters or videos at school about strengths?
(3) Direct peer communication was measured by answering “yes” to either: Has a friend
or other student… (a) told you about Sources of Strength?, or (b) talked to you about using
strengths?
(4) Participation in an activity consisted of answering “yes” to either: (a) Have you
participated in a Sources of Strength activity such as adding your trusted adult to a poster?, or (b)
Has a friend or other student asked you to name adults you can go to for help?
39 | Pickering
Assessment Metrics
Assessment metrics, including values for sociometric metrics and demographic
characteristics, were normalized within school to produce z-scores. These scores indicated the
average characteristics of peer leaders with respect to the general student body at these schools,
with a value of 1 indicating a standard deviation difference in that metric compared to the
average for all students.
Selection concordance. To address the concordance of the adult-selected peer leaders
with the proposed selection methods, we measured the percent of peer leaders for each method
who were also adult-selected peer leaders. We additionally computed concordance among all
other alternate peer leader selection methods. For example, if the school-level concordance
between APL and OPL methods was 20%, then 20% of the peer leaders selected by opinion
leadership at that school had also been chosen as adult-selected peer leaders.
Sociometric characteristics. The average in-degree, out-degree, coreness, closeness,
betweenness, and opinion leader nominations were normalized within school and computed
across schools.
Clustering. To determine the amount of peer leader clustering, for each selection method
we calculated the average number of peer leaders within one step of (i.e., directly connected to)
any given peer leader.
Representativeness. To assess demographic representation, for each selection method we
calculated the gender, race, age, and grade level of peer leaders and compared these values to the
school mean.
Reach. To determine the proximity of these peer leaders to at-risk students, we calculated
the distance of each peer leader to the closest student (other than the given peer leader) in each of
40 | Pickering
the three risk categories. A lower value therefore represented being closer in friendship steps to
these peers. Risk categories included: 1) indicating suicide ideation or suicide attempt, 2) being
in the periphery of the network (Borgatti & Everett, 2000), and 3) naming no trusted adults at
school. In the case that a peer leader was disconnected from all students within a risk category,
the maximum distance in the network was assigned. One school (10073) had no suicide attempts
and was excluded from the statistics on distance to closest student with attempt. Figure 7 shows
the distribution of at-risk students within the network of one sample school. Figure 8 illustrates
how the proportion of at-risk students varied across all schools. The smallest risk group was
suicidality (15.4% of students), followed by peripheral students (16.2%), with many students not
naming a trusted adult (32.0%).
Figure 7. Distribution of at-risk students in one sample network. Highlighted nodes include
those with suicide ideation or attempt (top left), those who did not name any trusted adults (top
right), and those who were peripheral in the network (bottom left).
41 | Pickering
Figure 8. Distribution of the proportion of at-risk students across schools, by risk type.
Analysis
Data import, cleaning, and analysis were performed in R (R Core Team, 2014). The
creation of network objects and network metrics were performed using the iGraph package
42 | Pickering
(Csardi & Nepusz, 2006). To determine the relationship between concordance of peer leader
selection methods with exposure, schoolwide exposure to the four Sources of Strength modalities
was regressed against the percent concordance with each peer leader selection method, with and
without adjusting for log-transformed school size. Regression analysis on these 20 school-level
observations was performed in R using the glm package.
Results
Sample
Across the 20 schools, average enrollment in the study was 82.2% (range 65.9-98.3%) for
a final baseline sample size of 5,746 students. The number of students participating ranged from
54 to 841 per school, with 9 to 45 adult-nominated peer leaders per school. Figure 6 presents the
proportion of the school population trained as peer leaders by school size, with the number of
peer leaders indicated. Schools with fewer than 200 students generally trained between 15-30%
of students, while schools with more than 400 students did not train more than 10%. A total of
4,026 participants completed information on Sources of Strength exposures at the end of the first
year.
Selection Concordance
There was general discordance of the adult-selected peer leaders with peer leaders chosen
through different identification methods (Table 3). The amount of concordance was as low as
13.3% for key players peer leaders and as high as 21.6% for opinion leaders. Concordance was
the highest between closeness and betweenness (54.2%) and lowest between coreness and
betweenness (11.1%). Degree centrality had consistently high concordance, as it was related to
opinion leaders (35.5%), coreness (31.8%), closeness (32.9%), betweenness (30.9%), and even
to key players (30.3%).
43 | Pickering
Table 3. Concordance among peer leader selection methods.
Adult Opinion Degree Core Close Between
Key
Player
Adult-
Selected
459
(100%)
Opinion
Leader
99
(21.6%)
459
(100%)
Degree
85
(18.5%)
163
(35.5%)
459
(100%)
Coreness
70
(15.3%)
94
(20.5%)
146
(31.8%)
459
(100%)
Closeness
71
(15.5%)
93
(20.3%)
151
(32.9%)
84
(18.3%)
459
(100%)
Betweenness
66
(14.4%)
88
(19.2%)
142
(30.9%)
51
(11.1%)
249
(54.2%)
459
(100%)
Key Players
61
(13.3%)
86
(18.7%)
139
(30.3%)
57
(12.4%)
70
(15.3%)
110
(24%)
459
(100%)
CPL1
104
(22.7%)
258
(56.2%)
227
(49.4%)
88
(19.2%)
128
(27.9%)
142
(30.9%)
127
(27.7%)
CPL3
102
(22.2%)
205
(44.7%)
246
(53.6%)
94
(20.4%)
158
(34.4%)
185
(40.3%)
136
(29.6%)
CPL5
98
(21.3%)
217
(47.2%)
203
(44.2%)
78
(17.0%)
103
(22.4%)
127
(27.7%)
153
(33.3%)
Peer Leader Characteristics
As expected, sociometric selection methods produced individuals with the highest values
of that characteristic (Table 4). Adults did nominate peer leaders with higher network
characteristics than the average student, but these values were relatively low in comparison to the
other selection methods. Consistent with their role as respected members of the community,
opinion leader peer leaders had high values of in-degree (M = +1.10, SE = 0.05) but were
generally not bridging nodes as they had relatively low betweenness (M = +0.12, SE = 0.05).
Peer leaders selected through the Key Players algorithm had higher values of social network
characteristics compared to the general population, but these values were modest in relation to
the other selection methods.
44 | Pickering
The most clustering occurred when using coreness and closeness to select peer leaders, as
these sets of individuals typically had over 3 friends within one friendship step (3.66 and 3.53,
respectively). The Key Players algorithm produced peer leaders with the fewest direct
connections to other peer leaders (0.42). While being instructed to select students from diverse
groups within school, adult-selected peer leaders on average had ties to 1.34 other peer leaders.
Only the adult-selected peer leader sets had any individuals who were disconnected from other
students in the network (0.9%); all other methods produced peer leaders who were connected to
the network. Figure 9 illustrates the general trends toward clustering and network position in one
school. Consistent with Table 3, peer leaders selected through coreness and closeness appear to
be highly clustered, while peer leaders selected through the Key Players algorithm appear to be
uniformly spread throughout the network. While opinion leaders were generally more spread out
in the network, they still tended to cluster in local pockets.
There were large demographic differences among the different selection methods. A
greater proportion of adult-selected peer leaders were female compared to the rest of the
population (M = +0.22, SE = 0.05); coreness and closeness also produced sets of peer leaders
that were generally more female than average (M = +0.10 and +0.11, respectively, SE = 0.05).
There were no other gender differences among selection methods.
While adult-selected peer leaders matched the racial composition of the student
populations, most of the other sociometric selection methods produced peer leaders that were
overwhelmingly more ethnically white than the general population, with the most ethnically
white students under the opinion leader selection method (M = +0.18, SE = 0.04). The
betweenenss and Key Players selection methods produced samples that were racially
representative. Adult-selected peer leaders and all sociometric methods produced peer leaders
45 | Pickering
that were younger than the general population. Conversely, opinion leaders tended to be
approximately half a year and grade level older than other students. The proportion of peer
leaders with suicide ideation and suicide attempt matched that of the general population under
almost all selection methods. Peer leaders chosen under the degree condition had a lower
proportion of students with suicide ideation than the general population (M = -0.12, SE = 0.04)
and those chosen under the closeness condition had a lower proportion of students with suicide
attempt than the general population (M = -0.08, SE = -0.04).
Every selection method produced peer leaders who were closer to at-risk students than
the general population. Peer leaders selected under the degree and betweenness conditions were
the closest to individuals with suicide ideation (M = -0.32 and -0.31), while adult-selected and
opinion leaders were the least close (M = -0.19 and -0.22). Likewise, peer leaders selected under
the betweenness condition were the closest to those with suicide attempt (M = -0.31), while
adult-selected, opinion leaders, and coreness peer leaders were the least close (M = -0.13, -0.19,
and -0.15). When translated, this indicates that if peer leaders were chosen under the
betweenness condition instead of the adult-selected condition, on average approximately one
more student with suicide ideation could be directly reached for every 5 peer leaders, and one
more student with suicide attempt could be directly reached for every 2 peer leaders.
The Key Players algorithm did the best job at being able to reach peripheral students and
students who were isolated from adults (M = -0.28 and -0.24, respectively), though peer leaders
selected by degree were also close to students who were isolated from adults (M = -0.24). Adult-
selected peer leaders were not as close to peripheral students (M = -0.06) or students isolated
from adults (M = -0.10).
46 | Pickering
Table 4. Metrics of the 459 peer leaders chosen by various methods. For all sections except
connectedness, displayed is the average in relation to the school-level mean value and SE. bold
values significantly different from 0.
Selection Method
Metric
M
(SD) Adult
Opinion
Leader Degree Coreness Closeness
Between-
ness
Key
Player
Social Network Metrics
In-Degree
4.08 +0.39 +1.10
(0.05)
+1.98
(0.03)
+0.93
(0.04)
+1.02
(0.04)
+0.98
(0.05)
+0.81
(0.05) (2.94) (0.05)
Out-Degree
4.79
(2.69)
+0.24
(0.04)
+0.34
(0.04)
+0.50
(0.03)
+0.58
(0.02)
+0.66
(0.02)
+0.67
(0.02)
+0.48
(0.03)
Coreness
4.97
(1.96)
+0.37
(0.04)
+0.65
(0.03)
+0.84
(0.02)
+1.11
(0.02)
+0.71
(0.02)
+0.55
(0.02)
+0.39
(0.03)
Closeness
0.10
(0.04)
+0.32
(0.05)
+0.44
(0.05)
+0.55
(0.05)
+0.53
(0.04)
+0.80
(0.05)
+0.71
(0.05)
+0.36
(0.04)
Betweenness
1573
(2467)
-0.04
(0.04)
+0.12
(0.05)
+0.45
(0.06)
-0.02
(0.04)
+0.87
(0.08)
+1.28
(0.09)
+0.31
(0.06)
Connectedness
PLs within 1
Step
1.34
(0.06)
2.17
(0.08)
2.43
(0.08)
3.66
(0.11)
3.53
(0.09)
2.27
(0.07)
0.42
(0.04)
%
Disconnected
2.9%
(16.9%)
0.9%
(0.4%)
- - - - - -
Demographics
Gender
(Female)
49.4%
(50.0%)
+0.22
(0.05)
+0.08
(0.05)
-0.04
(0.05)
+0.10
(0.05)
+0.11
(0.05)
+0.03
(0.05)
0.00
(0.05)
Race (White)
72.1%
(44.9%)
+0.05
(0.05)
+0.18
(0.04)
+0.12
(0.04)
+0.11
(0.04)
+0.14
(0.04)
+0.03
(0.05)
+0.03
(0.05)
Age
15.7
(1.3)
-0.16
(0.04)
+0.32
(0.05)
-0.15
(0.04)
-0.14
(0.05)
-0.32
(0.04)
-0.19
(0.04)
-0.09
(0.05)
Grade Level
10.4
(1.1)
-0.04
(0.04)
+0.45
(0.05)
-0.06
(0.05)
-0.06
(0.05)
-0.25
(0.04)
-0.18
(0.04)
-0.05
(0.05)
Suicide
Ideation
8.8%
(28.4%)
+0.08
(0.05)
-0.04
(0.04)
-0.12
(0.04)
-0.07
(0.04)
-0.04
(0.04)
+0.05
(0.05)
-0.07
(0.04)
Suicide
Attempt
7.6%
(26.5%)
-0.02
(0.04)
-0.04
(0.04)
+0.05
(0.05)
-0.05
(0.04)
-0.08
(0.04)
+0.05
(0.05)
+0.07
(0.05)
Average distance…
…to closest
SI
2.07
(1.81)
-0.19
(0.03)
-0.22
(0.02)
-0.32
(0.01)
-0.27
(0.02)
-0.29
(0.01)
-0.31
(0.02)
-0.28
(0.02)
…to closest
SA
2.29
(1.82)
-0.13
(0.03)
-0.19
(0.02)
-0.27
(0.02)
-0.15
(0.02)
-0.23
(0.02)
-0.31
(0.02)
-0.27
(0.02)
…to closest
peripheral
2.41
(1.71)
-0.06
(0.03)
-0.12
(0.02)
-0.18
(0.02)
-0.05
(0.02)
-0.11
(0.02)
-0.21
(0.02)
-0.28
(0.03)
…to closest
adult isolate
1.54
(1.75)
-0.10
(0.03)
-0.16
(0.01)
-0.24
(0.01)
-0.17
(0.01)
-0.20
(0.01)
-0.23
(0.01)
-0.24
(0.01)
47 | Pickering
Concordance and Exposure
School-level percent concordance of the empirical adult-selected peer leaders with
alternate selection methods (i.e., “selection concordance”) was related to schoolwide intervention
exposure for some modalities (Table 5). Selection concordance was not a significant predictor of
schoolwide exposure to a presentation, nor did it significantly predict schoolwide activity
participation in analyses adjusted for school size. Schoolwide rates of direct peer communication
were significantly larger when schools had peer leader sets that more closely aligned with
opinion leaders, closeness, and all custom peer leader selection methods. In analyses adjusted for
school size, this effect was significant for concordance with opinion leaders and marginal for the
custom selection methods. Exposure to a poster/video was significant for concordance with all
methods except coreness.
Table 5. Relationship between selection method concordance and exposure to Sources of
Strength across four modalities in 20 schools. *p<.05, +p<.10
Selection Method Modality
Unadjusted
Presentation Poster/Video Direct Peer Activity
Opinion Leader 0.32 (0.34) 1.05 (0.30)* 0.89 (0.22)** -0.01 (0.39)
Degree 0.50 (0.31) 0.82 (0.31)* 0.41 (0.27) 0.10 (0.37)
Coreness 0.23 (0.37) 0.73 (0.38) 0.41 (0.31) 0.23 (0.42)
Closeness 0.56 (0.3) 1.04 (0.27)* 0.58 (0.25)* 0.44 (0.35)
Betweenness 0.29 (0.27) 0.83 (0.24)* 0.40 (0.22) 0.63 (0.27)*
Key Players 0.46 (0.35) 0.87 (0.36)* 0.59 (0.29) 0.30 (0.41)
CPL1 0.30 (0.33) 0.89 (0.27)* 0.61 (0.22)* 0.06 (0.34)
CPL3 0.38 (0.33) 1.02 (0.30)* 0.66 (0.26)* -0.03 (0.39)
CPL5 0.55 (0.31) 1.07 (0.28)* 0.68 (0.24)* 0.23 (0.37)
Adjusted
Presentation Poster/Video Direct Peer Activity
Opinion Leader 0.18 (0.38) 0.74 (0.29)* 0.82 (0.24)* -0.49 (0.35)
Degree 0.41 (0.37) 0.41 (0.34) 0.21 (0.32) -0.51 (0.36)
Coreness 0.04 (0.41) 0.28 (0.37) 0.20 (0.34) -0.26 (0.40)
Closeness 0.52 (0.37) 0.72 (0.31)* 0.46 (0.31) -0.07 (0.38)
Betweenness 0.11 (0.44) 0.48 (0.38) 0.23 (0.36) 0.18 (0.43)
Key Players 0.34 (0.39) 0.50 (0.34) 0.43 (0.32) -0.12 (0.39)
48 | Pickering
CPL1 0.19 (0.33) 0.62 (0.26)* 0.51 (0.24)+ -0.32 (0.31)
CPL3 0.28 (0.35) 0.76 (0.27)* 0.55 (0.27)+ -0.36 (0.34)
CPL5 0.49 (0.39) 0.74 (0.32)* 0.59 (0.30)+ -0.42 (0.38)
Discussion
In peer-led interventions, the use of sociometric information to inform selection of peer
leader seeds in the network is considered superior to methods that do not use this information.
Less is known, though, about the network and demographic characteristics of peer leader sets
chosen under various sociometric methods. This study provides information on how the
characteristics of peer leader sets vary under several selection methods, performed across several
schools of varying size, structure, and demographic composition.
Our analyses confirm that the intent of adult-selected peer leader sets—namely, to
contain a diverse sample of students—is being achieved. Peer leaders selected by adults tended
to cluster less than other selection methods and had higher values of network centrality
characteristics than the general population. This suggests that adults may be tapping into
implicitly observed information about the school network even without using formal analytic
methods. Nonetheless, there is still potential to optimize peer leader selection as adults tended to
choose individuals who were less central, more female, and not particularly close to at-risk
students compared to the other selection methods.
No single method consistently maximized diagnostic metrics, and certain methods
produced peer leaders with characteristics incongruent with selection goals (e.g., sociometric
selection methods generally selected more white, younger, female peer leaders). Among all
methods, peer leaders chosen based on degree also had generally high values on all other social
network characteristics and were reasonably close to at-risk students in the network.
49 | Pickering
Interestingly, though the Key Players algorithm was designed to produce a set of
individuals maximally connected to others in the network, it additionally performed quite well at
producing a representative sample of individuals. Students chosen by the Key Players algorithm
aligned with the student population on all demographic characteristics. If there is demographic
clustering in the network and the algorithm selects individuals from various network positions,
this may explain why demographic groups are represented equally within these selection sets.
Peer leader sets chosen by betweenness also had individuals who were demographically similar
in terms of gender and race. This may be because individuals who are high in betweenness tend
to be bridges of groups, and in schools groups tend to be defined by gender and race.
Suicide ideation and attempt did not differ for any peer leader sets except those chosen by
degree and closeness. The degree metric provided peer leader sets with less suicide ideation than
the general population. While there could be several possible reasons for this, one interpretation
could relate to the constraints placed on popular individuals within networks. That is, popular
students may have the ability to spread information throughout the networks and set some trends,
but they generally tend to be reflective of the sentiment of the network overall (Valente, 2010). It
has been shown that students in Sources of Strength who have suicide ideation or attempt tend to
be less popular than those without suicidality—on average, suicidal students are 86% as popular
as non-suicidal students (Wyman et al., 2018). While this may put popular students in a place to
be role models of behavior, it also may cause them to be less empathetic to the needs of suicidal
students within the network. Also consistent with evidence from this study, peer leaders chosen
through closeness had lower attempts, likely because individuals with suicide attempt tend to be
located in sparse, less cohesive friendship groups.
50 | Pickering
Adults chose peer leaders who were close to at-risk students, but in almost every case the
network-informed methods happened to choose individuals who were closer. The betweenness
and degree methods chose individuals who were the closest to those with suicidal ideation and
attempts and to those who were isolated from adults, and the Key Players algorithm chose peer
leaders who were the closest to peripheral students. This may be an important consideration for
interventions that wish to deliver information to target students that fit these particular
distributions; for example, reaching students on the periphery of the network. Other studies have
shown that proximity to peer leaders plays a large part into whether an individual is exposed to
information; individuals are less likely to be exposed the further they are from peer leaders, and
this effect tapers off at 3 steps way (Pickering et al., 2018). In this regard—the ability to reach
students at particular at-risk positions within the network—network-informed selection is
superior to methods that do not use this information.
Of all selection methods, coreness was the most unique in its ability to reach individuals
with suicide ideation but not suicide attempt. This may be reflective of the nature of how suicide
ideation and attempt spread through the network. That is, individuals with high coreness may be
close to those with suicide ideation because this phenomenon may be more likely to spread
through social circles while suicide attempt does not. When examined together, the Key Players
algorithm did the best at reaching at-risk students across the four risk categories.
Concordance between adult-selected peer leaders and all selection methods was not
related to the proportion of students who were exposed to a presentation. This modality was
hypothesized to be the least affected by peer interactions; it therefore is not surprising that there
is a diminished role of peer leaders at facilitating exposure to this modality, and thus the
selection method does not matter as much. On the other hand, schools with high concordance
51 | Pickering
between adult-selected peer leaders and opinion leaders had higher rates of peer-to-peer
communication. If this effect were simply due to the ability to reach several students, then degree
concordance would also be predictive of exposure. Since it is not, this implies that opinion
leaders are effective at diffusing the message to peers for reasons that extend beyond simply their
ability to reach several people; opinion leaders may be more persuasive in their communication,
their opinions may be more sought after by other students, or they may have more resources
(e.g., social capital) that enable them to reach other students.
Several concordance measures were related to exposure to a poster/video. This is
unexpected considering our hypothesis that peer-to-peer communication should depend more on
peer leader selection than poster/video exposure. One explanation for this could be that peer
leaders affect poster/video exposure not simply due to network positioning, but because certain
characteristics of peer leaders may help them to be better at delivering the modality of posters
and videos to the school. If this is the case, then the data indicate that opinion leaders, students
chosen by closeness, and students chosen by the custom peer leader selection methods may be
better at influencing this modality. It could be the case that opinion leaders know how to deliver
persuasive poster/video messages, or students with high closeness know how to use media to
reach people around them. The custom peer leader selection methods drew upon different
demographic groups, so perhaps these peer leaders are better able to design messages that can
reach diverse populations within the school community.
These data come from a larger randomized control trial, but the data from this study is
observational. That is, peer leader sets in the study were not assigned based on any condition and
were simply observed. A stronger study method would be to assign peer leader sets based on the
various selection methods and observe intervention exposures within each of those schools. Also,
52 | Pickering
our analysis of selection methods assumes that the only differences in individuals are the
characteristics measured here: network position and demographics. There may be several other
considerations for peer leaders that may influence diffusion that are not measured here:
willingness to participate, attitudes toward the intervention, school attendance, student
personality type, etc. Though these characteristics may affect diffusion, they are not measured
with the current survey and may not be feasible to obtain through survey methods at the
beginning of a school year.
This study shows that network information can be used to help influence the selection of
peer leaders at school. By considering several different selection methods, researchers can see
the peer leader sets produced by these methods and evaluate these students’ demographic
compositions, positions in the network overall, and proximity to at-risk students. There is room
for improvement when considering methods for peer leader selection. Adult-selected methods
produce peer leaders in suboptimal network positions, while methods based solely on network
characteristics may produce peer leaders who are not demographically representative of the
school they attend. Opinion leaders may be better equipped to deliver the message of the
intervention, but also may not be demographically representative. Future work should determine
how these peer leader sets translate to diffusion of interventions through the school network.
53 | Pickering
Figure 9. Peer leaders selected using various methods in a sample school. From left to right
methods include: adult (red), opinion leader (yellow), degree (dark blue), closeness (aqua),
betweenness (baby blue), coreness (green-blue), and key players (green).
54 | Pickering
Study 2
Training and Social Support Influence Diffusion in A Peer-
Led School-Based Intervention
Background
A common strategy of school-based peer-led interventions is to train students in a school
or community as intervention advocates. These peers’ roles range from formal peer educators in
classroom settings to informal champions of the intervention who attempt to change attitudes and
behavior through informal interactions with others in their community. The use of peer leaders or
peer educators has been shown to be superior to the use of adults or intervention staff when
delivering intervention programs to adolescent groups.
Little is known about the factors that can contribute to how effective these peer leaders
are at delivering the intervention. Indeed, there is a dearth of information about how training
influences the effectiveness of change agents once they have been identified (Valente, 2017).
Other studies have shown that individuals who remain in workplace training programs generally
have more positive attitudes toward learning, greater perceived benefits of the training, and a
strong individual locus of control (Noe, 1986; Noe & Wilk, 1993). Studies on nursing educators
found that these educators were less likely to experience burnout in their role in a university
system when they had perceived social support from their chairperson and peers (Fong, 1993,
2016). The support of the nurse’s chairperson was the strongest predictor of a sense of
accomplishment in these programs—which may be analogous to teacher, adult, and/or
intervention staff providing support in school programs. Additionally, lack of peer support was
associated with burnout (Fong, 2016).
55 | Pickering
This lends evidence to the positive, beneficial effect of social support on adherence to
training programs. Indeed, peer support has been linked to behavioral maintenance, meaning that
individuals may be more likely to continue work as a peer leader with the support of their peers.
For example, “buddy systems”, where a participant goes through a training program or
intervention with a friend, have been used to improve several types of health-related behavioral
change (Hurdle, 2001).
While social support can lead to increased participation in school-based educational
programs, there is less evidence about how support can affect a peer leader’s ability to diffuse an
intervention message through a school. In one study examining an HIV risk avoidance
intervention, network proximity to multiple peer health advocates, or being in a network sector
where these health advocates clustered, led to a greater likelihood of exposure and adoption of
the intervention (Li et al., 2012). It is unknown, though, whether this effect is actually due to
peer support, or rather to other factors such as message reinforcement from multiple sources.
There are several ways that interventions like Sources of Strength may bolster social
support to enhance the diffusion potential of peer leaders. Three potential sources of support
have been identified: formal training (intervention support), adults support, and peer support.
There is additional evidence that a schoolwide culture of support may affect diffusion beyond
individual support. In a previous study, the schoolwide percent of students trained as peer leaders
contributed to an increased likelihood of an individual being exposed to the intervention, above
and beyond an individual’s proximity to a peer leader (Pickering et al., 2018). Therefore, multi-
level models were run to determine the effect of support at the individual level and at the school
level on network diffusion (Figure 10).
56 | Pickering
Figure 10. Conceptual model for Study 2.
The specific aims for this study are as follows:
2.1.-2.3. The effects of formal support/training (2.1), adult support (2.2), and peer support
(2.3) are assessed in a multivariable multi-level model. We examine the significance
of the parameter estimates across all four exposure modalities.
2.4. The additional effect of school-level support on intervention diffusion is examined in
the same model as the level-1 predictors, and includes total meetings at the school,
average adult support, average adults named, and total number peer leaders at the
school.
57 | Pickering
Methods
Recruitment
Of the 40 schools in the Sources of Strength evaluation trial, 20 were randomized to
receive the intervention immediately and are included in this study. Participating schools were
located in rural, small town, and micropolitan areas of New York and North Dakota and were
recruited from regions with past five-year youth suicide rates above the state averages.
Participants were followed for four waves of data collection spanning two years. In this study,
baseline network data are used along with follow-up data on peer leader experiences and
exposure to the Sources of Strength intervention. Of the schools in this study, 16 were from New
York and 4 were from North Dakota, ranging in size from 63-1,207 students (M=366). Two
schools served American Indian reservations.
Students were recruited in two phases: 1) at the beginning of the school year for the
school-wide Sources of Strength evaluation and 2) immediately following the first phase for the
selection of peer leaders. All students at school were invited to enroll in the evaluation study by
completing web-based assessments during the fall and spring over two years. Information letters
were sent to parents that allowed the option to decline their child’s participation. The University
of Rochester IRB approved the study protocol, and research personnel collected opt-out forms
and verbal assent with eligible students. Students were given information about how to access
help and support for themselves and other students if necessary.
Student Survey
Information on each student’s demographics and network characteristics was collected at
the beginning of the school year through a baseline survey. At the end of the school year, a
similar survey asked about exposure to the Sources of Strength intervention.
58 | Pickering
Demographics. The baseline survey administered to all students collected information on
student sex, ethnicity (white vs. nonwhite), and age.
Social network. Students were asked to name up to seven students in school who are their
closest friends. Additionally, students were asked to name up to seven adults at school who they
can trust and talk to about personal things.
Intervention exposure. Exposure to the Sources of Strength intervention was measured at
the end of the first school year and categorized into four different dichotomous exposure
modalities corresponding to various levels of peer engagement:
(1) Exposure to a presentation or assembly consisted of answering “yes” to either: Have
you seen a presentation or assembly about… (a) strengths that help teens get through hard
times?, or (b) helping suicidal teens by getting adults involved?
(2) Exposure to posters or videos was assessed by answering “yes” to: Have you seen
posters or videos at school about strengths?
(3) Direct peer communication was measured by answering “yes” to either: Has a friend
or other student… (a) told you about Sources of Strength?, or (b) talked to you about using
strengths?
(4) Participation in an activity consisted of answering “yes” to either: (a) Have you
participated in a Sources of Strength activity such as adding your trusted adult to a poster?, or (b)
Has a friend or other student asked you to name adults you can go to for help?
Peer Leader Selection
Peer leader selection was preceded by the training of adult advisors at school and baseline
assessments. Recruitment was performed following standardized procedures by distributing
59 | Pickering
nomination forms to staff members at school which asked for nominations of up to 6 students
whose “voices are heard” by other students. Nominations were reviewed with an attempt to
select individuals who reflected diverse groups within their school, with a target of 5-10% of the
school population selected as a peer leader. The size of peer leader teams, therefore, was
dependent on school size and staff selection. A total of 959 students were invited (19 to 86 per
school), with 798 (83.2%) enrolling with parent permission and youth assent. Of these, 459 (9-45
per school) were retained as active peer leaders.
Peer Leader Support
At the end of the first school year, peer leaders were asked to fill out a survey asking
about their experiences as a peer leader. In addition, records were kept on the number of training
meetings the peer leader attended, the total number of training meetings at each school, and the
ratio of meetings attended.
Peer leader training. Peer leader training was measured as the percent of meetings the
peer leader attended at the school. At the school level, peer leader training was measured as the
total number of training meetings offered.
Adult support. Adult support was measured through two methods: First, through the peer
leader survey as the mean of the following questions, assessed on a 4-point Likert scale: “As a
peer leader in Sources of Strength…” 1) “I’ve felt supported by the adult advisors for SoS”; 2)
“I’ve felt supported by the staff and teachers at the school.” Adult support was also measured by
the number of trusted adults the student named in the network section of the general survey.
School-level adult support was measured as the mean of the individual-level values at each
school.
60 | Pickering
Peer support. Peer support was indicated by whether a peer leader named another peer
leader as a friend. School-level peer support was measured as the percentage of the school
population trained as peer leaders.
Outcomes
The primary outcome of interest was the proportion of individuals in a peer leader’s
social network exposed to the intervention, across each modality. Egocentric networks were
produced by including (1) all individuals within one friendship tie of the peer leader (1-step
network) and (2) all individuals within two friendship ties of the peer leader (2-step network).
Then, the proportion of individuals in this network exposed to each intervention modality was
computed. When an egocentric network included a peer leader or an individual with missing
data, this individual was removed from the calculation of exposure rates. This method is
illustrated in Figure 11. In this hypothetical situation, the peer leader (red) is connected within 1
step to individuals who are exposed to the intervention (green), individuals who are not exposed
(white), and another peer leader (blue). In this example, the exposure for the peer leader’s 1-step
network is 67%.
61 | Pickering
Figure 11. Hypothetical 1-step friendship network around a peer leader (red) with exposed
friends (green), unexposed friends (white), and students who have missing data or are also peer
leaders (blue). Exposure in this 1-step egocentric network is 67%.
Analysis
Data import, cleaning, and analysis were performed in R (R Core Team, 2014). The
creation of network objects and network metrics were performed using the iGraph package
(Csardi & Nepusz, 2006). Multi-level models with level-1 (student) and level-2 (school)
variables were performed in the lme package (Bates et al., 2017). These models assessed the
effect of each type of support on intervention exposure, with a random intercept for school.
Because school size was hypothesized to moderate the effect of school-level peer support on
intervention exposure, an interaction term between these two variables was included in the
model. Models were built in a stepwise fashion, starting with a base model that did not include
any support variables. Next, each domain of support (i.e., training, adult, peer) was included
individually in the model, followed by a final model with all variables included.
62 | Pickering
Results
Sample
Across the 20 schools, average participation in the study was 82.2% (range 65.9-98.3%)
for a final baseline sample size of 5,746 students. The number of students participating ranged
from 54 to 841 per school, with 9 to 45 adult-nominated peer leaders per school. Figure 6
presents the proportion of the school population trained as peer leaders by school size, with the
number of peer leaders indicated. Schools with fewer than 200 students generally trained
between 15-30% of students, while schools with more than 400 students did not train more than
10%. A total of 438 peer leader participants completed information on Sources of Strength
exposures at the end of the first year. Of these, 4 peer leaders were isolates and no information
was able to be computed on their egocentric networks. The final analytic sample size was 434
peer leaders.
Peer leaders were generally able to reach a large audience. The mean size of peer leaders’
1-step networks was 8.75 (SE = 3.09) and 2-step networks was 38.90 (SE = 16.07). There were
few peer leaders who were not connected to any individual, and in some cases peer leaders were
able to reach over 90 students in their 2-step egocentric networks (Figure 12). Peer leaders were
overwhelmingly white ethnicity (80%), though this was consistent with the average across
schools (78%). Peer leaders were 60% female (general population 50%), and on average 15.5
years of age (general population 15.8). Peer leaders on average attended 59% of meetings, with
an average of 8.3 (SD = 3.7) meetings per school during the first school year. On average, peer
leaders named 3.3 (SD = 2.4) trusted adults and had an average value of 3.4 (SD = 0.56) on the
peer leader survey asking about social support (range: 1-4), in the range of “agree” to “strongly
agree” that they felt supported. Peer leaders tended to not cluster together (M = 1.3 peer leader
63 | Pickering
friends named, SD = 1.2), but in some cases peer leaders were connected to up to 6 other peer
leaders. On average, 12% of schools were trained as peer leaders (SD = 0.07).
Figure 12. Distribution of the sizes of peer leaders' 1- and 2-step egocentric networks.
Demographics and Exposure
Across schools, exposure to the Sources of Strength intervention was highest for
poster/video exposure (62%), followed by direct peer communication (61%), exposure to a
presentation (56%), and participation in an activity (54%). In all models, males had lower
exposure to direct peer communication in their 1-step networks (base model B = -6.1, p < .05),
and had lower exposure to all modalities except presentation in their 2-step networks (base
model B’s range from -3.7 to -3.3, p’s < .01). Ethnicity was not related to intervention exposure
64 | Pickering
in peer leaders’ 1- or 2-step networks in any model. There is some evidence that older students
are more disengaged from the Sources of Strength program. These peer leaders’ 1- and 2-step
egocentric networks had less exposure to the intervention in some models—typically, to
presentation and activity participation, and to a lesser extent peer-to-peer communication, but
never to poster/video exposure.
65 | Pickering
Table 6. Intervention diffusion to peer leaders’ local social networks as a function of
demographic characteristics and social support.
0. Base
1-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) 2.2 (-2.5, 6.9) -2.8 (-6.8, 1.1) -6.1 (-9.9, -2.2)* -3.4 (-7.6, 0.9)
Ethnicity
(ref=nonwhite) -0.4 (-6.8, 6) 1.7 (-3.8, 7.2) 3.8 (-1.6, 9.1) -1.5 (-7.3, 4.4)
Age -2 (-4.1, 0.1) 0.9 (-0.9, 2.6) -1.1 (-2.8, 0.6) -1.2 (-3.1, 0.6)
Training (% Meet)
Adult Support
Peer Support
Level-2
School Size -3.6 (-11.1, 3.9) -6.9 (-15.3, 1.5) -1 (-8.1, 6.2) -8.3 (-15.1, -1.5)*
Training (# Meet)
Adult Support
Peer Support
Peer Support x Size
0. Base
2-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) -1.2 (-3.3, 0.9) -3.3 (-5.1, -1.4)** -3.6 (-5.4, -1.7)** -3.7 (-5.7, -1.6)**
Ethnicity
(ref=nonwhite) -1 (-3.9, 1.9) -0.4 (-3.1, 2.2) 0.6 (-2, 3.2) -0.3 (-3.2, 2.6)
Age -1.3 (-2.3, -0.4)* -0.1 (-1, 0.7) -0.6 (-1.4, 0.2) -0.8 (-1.7, 0.1)
Training (% Meet)
Adult Support
Peer Support
Level-2
School Size -3.2 (-11, 4.6) -8.9 (-16.5, -1.3)* -2.5 (-9.6, 4.5) -8.9 (-16.5, -1.4)*
Training (# Meet)
Adult Support
Peer Support
Peer Support x Size
66 | Pickering
I. Training
1-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) 2.7 (-2, 7.3) -3.3 (-7.1, 0.4) -6.3 (-10.1, -2.5)* -2.4 (-6.6, 1.8)
Ethnicity
(ref=nonwhite) 4.7 (-2.1, 11.5) 3.2 (-2.4, 8.7) 4.3 (-1.3, 10) 4.5 (-1.7, 10.7)
Age -2.2 (-4.2, -0.1)* 1.2 (-0.4, 2.8) -0.8 (-2.5, 0.9) -1.1 (-2.9, 0.8)
Training (% Meet) 5.9 (-2.6, 14.3) 4.7 (-2.1, 11.6) 5.3 (-1.7, 12.3) 11.7 (4.1, 19.4)*
Adult Support
Peer Support
Level-2
School Size -2 (-10.9, 6.9) -11 (-20.5, -1.4)* -2.6 (-11.7, 6.5) -7.7 (-15.7, 0.3)
Training (# Meet) 0.7 (-0.9, 2.3) -0.4 (-2.2, 1.3) -0.1 (-1.7, 1.6) -1.5 (-2.9, 0)
Adult Support
Peer Support
Peer Support x Size
I. Training
2-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) -1 (-2.9, 0.9) -3.4 (-5.1, -1.7)** -3.6 (-5.3, -1.9)** -3.8 (-5.7, -2)**
Ethnicity
(ref=nonwhite) 2.6 (-0.2, 5.4) 1.1 (-1.5, 3.6) 2.8 (0.2, 5.3)* 4.4 (1.6, 7.2)*
Age -1.5 (-2.3, -0.6)** 0.2 (-0.5, 0.9) -0.3 (-1, 0.5) -0.8 (-1.6, 0)*
Training (% Meet) 2.6 (-0.9, 6.1) 3.8 (0.6, 6.9)* 4.7 (1.5, 7.9)* 4.5 (1.1, 7.9)*
Adult Support
Peer Support
Level-2
School Size -2.5 (-12.1, 7.1) -10 (-19.1, -0.9)* -3.2 (-12.2, 5.9) -8.5 (-17.2, 0.2)
Training (# Meet) 0.9 (-0.9, 2.6) 0 (-1.6, 1.6) 0.1 (-1.5, 1.7) -1.7 (-3.3, -0.2)*
Adult Support
Peer Support
Peer Support x Size
67 | Pickering
II. Adult Support
1-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) 1.5 (-3.5, 6.4) -3.9 (-8.2, 0.3) -7.1 (-11.1, -3)** -5.1 (-9.5, -0.7)*
Ethnicity
(ref=nonwhite) -3.3 (-10.4, 3.8) 0.4 (-5.7, 6.6) 2.7 (-3, 8.4) -0.2 (-6.5, 6.1)
Age -2.7 (-4.9, -0.6)* 0.8 (-1.1, 2.6) -2.1 (-3.9, -0.3)* -2.1 (-4.1, -0.2)*
Training (% Meet)
Adult Support -1.7 (-4.3, 0.8) -0.5 (-2.7, 1.7) -0.3 (-2.4, 1.8) -1.2 (-3.5, 1.1)
Peer Support
Level-2
School Size -5.7 (-13.8, 2.3) -8.1 (-17, 0.9) -5.4 (-11.1, 0.3) -5.3 (-12.4, 1.7)
Training (# Meet)
Adult Support 17.2 (0.6, 33.7) 6.4 (-11.7, 24.5) 23.2 (11.3, 35.1)* -12.4 (-26.8, 2.1)
Peer Support
Peer Support x Size
II. Adult Support
2-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) -2 (-4.3, 0.3) -3.7 (-5.7, -1.7)** -4.1 (-5.9, -2.2)** -4.5 (-6.8, -2.3)**
Ethnicity
(ref=nonwhite) -1.5 (-4.9, 1.9) -1.8 (-4.7, 1) -0.2 (-2.8, 2.5) -0.8 (-4, 2.4)
Age -1.7 (-2.8, -0.7)* -0.2 (-1.1, 0.6) -0.6 (-1.4, 0.2) -1 (-2, 0)*
Training (% Meet)
Adult Support -0.8 (-2, 0.4) -0.3 (-1.4, 0.7) -0.8 (-1.7, 0.1) -0.5 (-1.7, 0.6)
Peer Support
Level-2
School Size -5.8 (-14.2, 2.5) -10.8 (-18.8, -2.8)* -6.2 (-12.2, -0.2) -7.3 (-14.8, 0.2)
Training (# Meet)
Adult Support 14.1 (-2.3, 30.5) 9 (-6.7, 24.7) 20.9 (9, 32.8)* -12.6 (-27.4, 2.3)
Peer Support
Peer Support x Size
68 | Pickering
III. Peer Support
1-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) 2.4 (-2.3, 7.1) -2.7 (-6.6, 1.2) -6 (-9.9, -2.1)* -3 (-7.2, 1.1)
Ethnicity
(ref=nonwhite) -1.8 (-8.3, 4.7) -0.4 (-5.8, 5.1) 2.6 (-2.8, 8) -3.6 (-9.3, 2.2)
Age -1.8 (-3.8, 0.3) 1.2 (-0.6, 2.9) -0.9 (-2.6, 0.8) -0.8 (-2.7, 1)
Training (% Meet)
Adult Support
Peer Support 2.8 (0.7, 4.8)* 2.9 (1.2, 4.6)* 1.6 (-0.1, 3.4) 4.6 (2.7, 6.4)**
Level-2
School Size -1.2 (-11.3, 8.9) 1.1 (-8.4, 10.7) 3.4 (-5.4, 12.3) -6.3 (-15.7, 3.1)
Training (# Meet)
Adult Support
Peer Support 1.1 (-6.3, 8.6) 9.6 (1, 18.3)* 6.8 (-1.3, 14.8) -0.3 (-7.2, 6.6)
Peer Support x Size NS 7 (2.1, 12)* 5.8 (1.2, 10.4)* NS
III. Peer Support
2-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) -1.2 (-3.3, 0.9) -3.2 (-5.1, -1.3)** -3.5 (-5.4, -1.7)** -3.6 (-5.7, -1.6)**
Ethnicity
(ref=nonwhite) -1 (-3.9, 1.9) -1.1 (-3.8, 1.5) 0 (-2.6, 2.6) -0.6 (-3.5, 2.4)
Age -1.3 (-2.3, -0.4)* 0 (-0.8, 0.8) -0.5 (-1.3, 0.3) -0.7 (-1.6, 0.2)
Training (% Meet)
Adult Support
Peer Support 0 (-0.9, 1) 1.2 (0.4, 2)* 1 (0.2, 1.8)* 0.5 (-0.4, 1.4)
Level-2
School Size -1 (-11.5, 9.4) -0.7 (-10, 8.5) 4.4 (-4.1, 12.9) -4.8 (-14.7, 5.2)
Training (# Meet)
Adult Support
Peer Support 1.4 (-6.2, 9.1) 9.1 (0.7, 17.4) 9.1 (1.4, 16.8)* 2.2 (-5.1, 9.6)
Peer Support x Size NS 5.8 (0.9, 10.6)* 5.8 (1.4, 10.2)* NS
69 | Pickering
IV. All Predictors
1-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) 2.6 (-2.3, 7.4) -3.9 (-7.8, 0.1) -6.2 (-10.2, -2.2)* -2.6 (-6.9, 1.6)
Ethnicity
(ref=nonwhite) 0.7 (-6.9, 8.3) -0.8 (-6.9, 5.4) 1.5 (-4.7, 7.7) 4.5 (-2.1, 11)
Age -2.9 (-5, -0.8)* 1 (-0.8, 2.7) -2 (-3.8, -0.3)* -1.9 (-3.7, -0.1)*
Training (% Meet) -0.8 (-10.2, 8.7) 3.7 (-4, 11.3) 1.4 (-6.3, 9) 8.6 (0.5, 16.7)*
Adult Support -1.5 (-4, 1) -1.1 (-3.1, 0.9) -0.5 (-2.6, 1.5) -1.2 (-3.3, 1)
Peer Support 2.9 (0.7, 5.1)* 2.9 (1.1, 4.6)* 2.1 (0.3, 4)* 4 (2.1, 5.9)**
Level-2
School Size 1.7 (-10.2, 13.6) -0.7 (-11.3, 9.9) 1.9 (-5.5, 9.3) -3.8 (-13.9, 6.3)
Training (# Meet) 0 (-2.1, 2) -0.5 (-2.3, 1.2) -0.7 (-1.9, 0.5) -0.4 (-2.2, 1.3)
Adult Support 18.4 (-2, 38.7) 1.9 (-16.4, 20.1) 23.1 (9.9, 36.3)* -12.3 (-29.6, 5.1)
Peer Support 3.2 (-5, 11.3) 11.3 (2.2, 20.4)* 8.3 (2, 14.6)* -0.1 (-7.1, 6.8)
Peer Support x Size NS 8 (1.8, 14.2)* 5.4 (1.1, 9.7)* NS
IV. All Predictors
2-Step Networks Presentation Poster/Video Direct Peer Activity
Level-1
Gender (ref=female) -1.9 (-3.9, 0.2) -3.6 (-5.5, -1.8)** -3.7 (-5.6, -1.9)** -4.4 (-6.4, -2.5)**
Ethnicity
(ref=nonwhite) 1.9 (-1.3, 5.2) -0.8 (-3.7, 2.2) 1.5 (-1.4, 4.4) 3.3 (0.2, 6.4)*
Age -2.1 (-3, -1.1)** -0.1 (-0.9, 0.7) -0.4 (-1.2, 0.4) -1.3 (-2.1, -0.4)*
Training (% Meet) 0.5 (-3.6, 4.6) 2.8 (-0.9, 6.4) 2.1 (-1.5, 5.7) 2.9 (-1, 6.7)
Adult Support -0.1 (-1.2, 0.9) -0.1 (-1, 0.9) -0.5 (-1.4, 0.4) 0 (-1, 1)
Peer Support -0.2 (-1.1, 0.8) 0.9 (0, 1.7)* 1 (0.1, 1.8)* 0.1 (-0.8, 1)
Level-2
School Size -1.9 (-14.2, 10.4) -5.2 (-16.3, 5.9) -1.2 (-9.4, 6.9) -4.2 (-14.7, 6.2)
Training (# Meet) 0.4 (-1.7, 2.6) -0.6 (-2.5, 1.3) -1.3 (-2.7, 0.1) -1.2 (-3.1, 0.6)
Adult Support 10.2 (-10.3, 30.7) 10.6 (-7.9, 29) 26.2 (12.6, 39.8)* -9.4 (-26.9, 8)
Peer Support 1.8 (-6.7, 10.2) 3.9 (-3.8, 11.5) 3.9 (-1.7, 9.5) 2.5 (-4.7, 9.7)
Peer Support x Size NS NS NS NS
70 | Pickering
Support and Exposure
The relationship between support and exposure within peer leaders’ egocentric networks
is presented in Table 6. The model was built in a stepwise fashion, with individual support
domains (i.e., training, adult, and peer support) being added individually (models I – III), and
then collectively (model IV). Though adult support was measured both by survey ratings and the
number of trusted adults named, survey ratings had higher t-values and was more predictive of
egocentric network exposure than number of trusted adults named, which was not significant in
any model; therefore, survey ratings of adult support were used instead.
Training. The percent of meetings an individual attended, and the schoolwide number of
meetings were generally not predictive of egocentric network exposure for any modality. The
exception to this was the effect of meeting attendance on 1-step network exposure to activity
participation. Individuals who attended a greater proportion of meetings had higher exposure in
their 1-step egocentric networks (B = 8.6, 95% CI = [0.5, 16.7]).
Adult support. Adult support was not predictive of egocentric network exposure to a
presentation, a poster/video, or activity participation at the individual and school level. Peer
leader ratings of adult support did not predict peer communication. However, in schools with
higher average ratings of adult support, peer leaders had 1- and 2-step egocentric networks with
greater exposure to peer communication. This effect was quite stark; a 1 SD (0.36-unit on the
Likert scale survey response) increase in schoolwide mean ratings of adult support was
associated with a 23% increase in 1-step and a 26% increase in 2-step egocentric network
exposure, or approximately 1 in 4 friends.
Peer support. Training with a peer increased 1-step egocentric network exposure to all
intervention modalities and was most significant at increasing activity participation (B = 4.0 for
training with an additional friend, 95% CI = [2.1, 5.9]). Beyond the effect of an individual
71 | Pickering
training with a friend, the proportion of the school trained as peer leaders was additionally
predictive of 1-step egocentric network exposure to a poster/video and direct peer
communication, (B = 11.3 and 8.3, respectively, p’s < .05). This effect was even greater in larger
schools; the interaction between school size and percent trained as peer leaders was significant
for these modalities as well (interaction B = 8.0 and 5.4 respectively, p’s < .05). To be sure
percent of meetings attended did not depend on schools having too many meetings, correlation
analyses were performed at the school level and showed no relationship between mean percent
meetings attended and total meetings offered (r = .07, p = .39).
The effect of peer support was more limited within the 2-step egocentric networks.
Training with an additional peer only increased the 2-step egocentric network exposure to a
poster/video by 0.9% and to direct peer communication by 1.0% (p’s < .05). There was no effect
of percent of school trained as peer leaders on 2-step egocentric network exposure.
Post-hoc analyses were conducted to determine the dose-response effect of peer leaders
training with an additional friend (Table 7). The variable indicating training with a friend was
split into dummy variables indicating having 1, 2, or 3 or more peer leader friends. Having a
friend who was a peer leader increased the exposure in peer leaders’ 1-step egocentric networks
across all modalities except direct peer communication, when the benefit began at 2 friends (all
B > 5.5, p’s < .05), but there was no additional increase in exposure beyond 1 friend for the
presentation, poster/video, and peer communication modalities (Figure 11). Having additional
peer leader friends appeared to increase activity exposure in peer leaders’ 2-step egocentric
networks; the increase in exposure was 5.5% (95% CI = [0.1, 10.9]) for one friend, 9.1% (95%
CI = [3.1, 15.1]) for two friends, and 15.4% (95% CI = [8.7, 22.2]) for three or more friends.
72 | Pickering
Figure 13. Relationship between having additional peer leader friends and exposure to four
Sources of Strength modalities. Values represent the increase in 1- or 2-step egocentric network
exposure for training with an additional friend.
73 | Pickering
The relationship between number of peer leader friends and 2-step egocentric network
exposure was not as strong for presentation or activity modalities, where the effect was not
significant. The exposure to poster/video in 2-step egocentric networks increased when training
with 1 peer leader friend, but there was no marginal increase in exposure after 1 friend.
Similarly, the exposure to direct peer communication in 2-step egocentric networks increased
when training with 2 peer leader friends, but there was no marginal increase in exposure after 2
friends.
Table 7. Relationship between number of peer leader friends and 1- and 2-step egocentric
network exposure to intervention modalities. Presented are regression coefficients and 95%
confidence intervals. *p<.05, **p<.001.
# PL
Friends Presentation Poster/Video
Peer
Communication Activity
1-Step Network
1 9.5 (3.5, 15.4)* 7.4 (2.4, 12.4)* 2.8 (-2.2, 7.8) 5.5 (0.1, 10.9)*
2 10.5 (3.9, 17.2)* 11 (5.4, 16.6)** 6.3 (0.8, 11.9)* 9.1 (3.1, 15.1)*
3+ 8.8 (1.4, 16.3)* 9.4 (3.1, 15.7)* 5.2 (-1, 11.5) 15.4 (8.7, 22.2)**
2-Step Network
1 1.4 (-1.3, 4.1) 3.7 (1.3, 6.1)* 2.0 (-0.4, 4.3) -0.3 (-3.0, 2.3)
2 1.0 (-2.0, 4.0) 4.0 (1.4, 6.7)* 2.9 (0.2, 5.5)* 0.4 (-2.6, 3.4)
3+ 0.1 (-3.2, 3.5) 4.0 (1.0, 7.0)* 3.5 (0.5, 6.4)* 2.0 (-1.3, 5.4)
Discussion
School-based interventions are increasingly using peer leaders to deliver the intervention
to the school population resulting in more powerful, effective, and economical programs. While
some research has examined factors that contribute to remaining in training programs, less is
known about how types of support can affect the ability of peer leaders to disseminate an
intervention and influence other students in school-based programs. This study provides an
74 | Pickering
examination of the ways in which intervention training, adult support, and peer support are able
to enhance the ability of peer leaders to deliver an intervention.
This study provides evidence that certain types of support can improve the effectiveness
of peer leaders. We found that each of the three types of support are able to influence diffusion
of the Sources of Strength intervention from peer leaders, though the mechanisms through which
this occurs, and the specific contextual applications, vary depending on support type, exposure
modality, and reach of peer leaders (i.e., 1-step vs. 2-step networks).
The activity modality appeared to be the most responsive to support in multiple ways.
First, this was the only modality in which exposure benefitted by peer leaders’ having more
meetings attended. This implies that a higher level of training is necessary to engage peers in
activity participation beyond simple dissemination of the Sources of Strength message. Training
may provide students with useful skills that are able to engage them with other students and
persuade them to participate in an activity.
Second, peer support was highly related to proportion of the 1-step egocentric network
who participated in an activity. This implies that buddy-style training may have an added benefit
with this modality. It may require more resources to convince an individual to participate in an
activity than to simply listen to a peer talk about the intervention. Students must actively
participate when performing activities, which may be a greater barrier to behavior and thus may
require more than one trained individual to persuade them to do so. Alternately, students may
need to hear the intervention’s message from more than one source to be convinced to
participate.
The strong effects of support on activity participation in the 1-step network, and lack of
an effect in the 2-step network, demonstrates the limited influence peer leaders have to engage
75 | Pickering
other students. In fact, it may take more than one trained peer leader to students to participate in
Sources of Strength activities. Once a friend of a peer leader participates, that individual may not
have the resources, training, or motivation to get his or her friends to subsequently participate.
That is, peer leaders can convince students to participate in activities, but they have a limited
effect on getting the friends of their friends to participate.
Peer leader training at the individual and school level was generally not predictive of
exposure to the intervention. This may indicate that the number of meetings being held currently
at each school is adequate to train peer leaders. And, even if certain peer leaders attend a low
percent of meetings at school, individuals in their 1- and 2-step networks will nonetheless be
exposed to presentations, posters and videos, and direct peer communication. It may be that
communicating the intervention’s message does not require as much training resources as does
getting students engaged to participate in intervention activities.
Additionally, adult support was generally not predictive of peer leaders’ egocentric
network exposures. This may indicate that individuals who have trained in the program do not
require as many external reinforcements (e.g., adult reinforcement) to carry out their duties as
peer leaders, nor do they depend as strongly on the quality of adult support to provide resources
they need to be peer leaders.
That the school-level effect of adult support was predictive of direct peer communication
rates, but not individual-level adult support, suggests that the mechanism by which adult support
increases diffusion in these networks does not act directly on peer leaders. Instead, having a
culture of adult support at the school level may make these networks more conducive to
receiving information about the intervention. Schools with more supportive adults may also have
76 | Pickering
students who are more open to talking about difficult subjects like suicide prevention, or may be
schools in which individuals’ opinions are more readily shared and respected by others.
Adults may also have some level of influence when it comes to communicating
information about the intervention schoolwide. That is, individuals in schools with higher adult
support may gain information about the intervention through these adults, making the
intervention more salient to these students. This study does not examine the joint student-to-
student network and student-to-adult network, as the information about trusted adults was
collected only from students. However, adults may act as indirect ties, and two disconnected
students may transfer information about the intervention if they share this information with one
shared adult. Interventions such as Sources of Strength should focus on bringing adults on board
in a supportive role, as their support also appears to influence the spread of intervention
messaging.
The significant effect of peer support on exposure across all modalities and at multiple
levels (school and individual) suggests that the “redundancy problem” posed by Borgatti is not
so much of a concern in these networks (Borgatti, 2006). If connected peer leaders were in fact
contributing redundant information to these networks, we could expect to see no effect of peer
support on intervention exposure. However, there appears to be a boost to diffusion efficacy
when students train with peer leader friends. The exact mechanism though which this works is
not entirely clear, though. It may be that training with a friend increases person-level factors
(e.g., motivation to participate, effectiveness at communicating information, etc.), or that peer
leaders need assistance to reach all individuals in their egocentric networks, or that friends of
peer leaders may need reinforcement of the intervention message from other sources to
assimilate the conveyed information and/or retain this information.
77 | Pickering
Training with more than one friend was beneficial to diffusion of all intervention
modalities. However, the added benefit was limited to one friend for all modalities except
activity, which saw increasing gains in exposure as more peer leaders trained together. This
suggests that guidelines such as the buddy system may be beneficial when selecting peer leaders
(Hurdle, 2001). At a minimum, this finding should emphasize that methods to spread peer
leaders throughout the network (e.g., Key Players) are not in fact necessary as long as there is not
a high degree of peer leader clustering in the network.
In addition, training more of the school population as peer leaders, especially in large
schools, benefitted diffusion of poster/video exposure and peer communication in peer leaders’
egocentric networks above and beyond their own direct connections to other peer leaders. This
finding was also seen in previous studies on Sources of Strength, where it was shown that an
individual was more likely to be exposed to the intervention if a school had a greater proportion
of students trained as peer leaders, above and beyond the individual’s own proximity to a peer
leader (Pickering et al., 2018). Having more peer leaders in a school provides greater
opportunities for informal communications that happen outside of friendship networks, such as
incidental contact through “weak ties” or acquaintances (Granovetter, 1983). This is supported
by the lack of an effect of support on activity; the ability of information to pass through weak ties
is powerful, but persuasion and behavior change do not tend to act through these types of
connections. Even so, this study examines the friendship network as the primary mechanism
through which diffusion occurs, but it should be noted that a small level of influence can happen
between individuals who are not connected through friendship ties.
This study benefitted by being able to study diffusion from multiple peer leaders across
several schools of varying sizes and compositions. Additionally, there was a high response rate
78 | Pickering
of peer leaders at the end of the first year. Peer leaders’ egocentric networks were measured to
study the extent of diffusion in their social sphere, though it should be noted that diffusion may
have occurred outside these friendship ties. Additionally, the networks were assumed to be
unchanging across the first school year. It may have been the case that peer leaders actively
sought to form ties to other students within the network and increased the size of their egocentric
networks. If this were the case, assuming peer leaders did not lose individuals within their
egocentric networks, the estimates provided in this study would err on the conservative side.
We identified several ways that the Sources of Strength intervention can support peer
leaders once they have been identified and begun training, and there are important implications
for future iterations of this program. If the goal is to change behavior and get students to
participate in intervention activities, peer leaders need to attend a high percentage of training
meetings and undergo training with friends. Spreading awareness of the intervention messaging
may be an easier task and is also improved by having peer leaders train with a friend and having
several peer leaders at a school. Getting adults at school to support the intervention and its
activities can go a long way toward getting students talking about the intervention. Taken
together, the resources spent on support will depend on the program goals and types of messages
or behaviors targeted in the population of interest.
79 | Pickering
Study 3
Network Structure and Peer Leader Seeds Affect Diffusion
of a Suicide Preventive Intervention in High Schools: A
Simulation Study
Background
When school interventions use peer leader “seeds” to disseminate the intervention
messaging, knowledge of how these seeds will impact diffusion can lead to more effective
interventions that can be tailored to individual schools. The effect of choosing different
intervention seeds as initial adopters on subsequent diffusion has been previously studied. For
example, Valente and Davis (1999) found that intervention diffusion is fastest, and most
saturated, when opinion leaders are selected as initial seeds compared to random individuals or
those on the margins of the network.
When a complete network census is available, it may be beneficial to utilize this
information and employ multiple peer leader selection approaches. A one-size-fits-all method
may not be appropriate. This is because network structure may impact the spread of the
intervention in different areas (e.g., in clustered pockets of the network) and, as has been
hypothesized, different sociometric characteristics may be beneficial for diffusion at different
stages during the process. For instance, individuals with high degree values may help initiate a
quick spread of the intervention, while characteristics such as betweenness and closeness may be
more beneficial at helping diffusion reach saturation.
There is also evidence that network structure moderates the effectiveness of different
initial adopter seeds. For example, highly central individuals are more influential in a highly
80 | Pickering
centralized network (e.g., one with distinctly popular individuals) compared to a decentralizaed
network (e.g., individuals who generally have similar degree status) (Valente, 2010). Social
network structure was found to influence the diffusion of a substance use prevention intervention
(Rulison, Gest, & Osgood, 2015). In this study, diffusion was higher in networks that were more
structurally cohesive (i.e., students were connected through many independent paths), less
clustered, less centralized (i.e, fewer popular students who may act as gatekeepers to the
intervention). Diffusion was also higher when a greater proportion of students were intervention
participants and when they were more spread-out through the network. The social status of
intervention participants did not affect diffusion, though this study only measured intervention
exposure; opinion leaders may have had more of an effect at eliciting attitudinal and behavioral
change. Other studies have shown that networks that have more cliques and are more modular
have slower diffusion (Choi, Kim, & Lee, 2010), that density increases diffusion (Brown et al.,
2012), and that more centralized networks have slower diffusion (Valente, 2010).
The Sources of Strength intervention provides a unique opportunity to study the spread of
intervention messaging in that, in addition to promoting diffusion broadly, it is especially
important to diffuse intervention messaging to at-risk students at school. These students include
those at the periphery of the network (away from popular friendship groups), those with suicide
ideation and/or attempt (the behavior attempting to be mitigated by the intervention), and those
who are isolated from adults (reducing the ability to seek help in times of crisis). Additionally,
the Sources of Strength evaluation study has examined the intervention in 20 schools with a
range of sizes and network properties. Figure 14 shows two schools from the intervention of
similar size but with different network structures—the network on the bottom is more dense than
81 | Pickering
the top network yet also more clustered. It is this variation in network structure that allows
testing (via simulation) of how various selection methods may impact diffusion.
Figure 14. Two friendship networks from the Sources of Strength intervention. Nodes are
colored by adult-selected peer leader status, and by how many steps an individual is from a peer
leader.
82 | Pickering
The specific aims of this study are:
3.1.The effect of peer leader selection on diffusion is assessed by performing several
diffusion simulations and generating the average total exposure and average exposure
to at-risk groups in a “short” and “long” time period. The “long” time period will be
equivalent to the end of one school year in the Sources of Strength intervention. Since
the simulation will be run several times, bootstrapped confidence intervals of the
exposure are produced.
3.2.The effect of network structure on simulated network diffusion is assessed by using
each network structure variable as a predictor of diffusion under various selection
methods.
3.3.The reach to at-risk nodes is assessed by examining the exposure to these three
subgroups (suicide risk, periphery, isolated from adults) in addition to the general
student population.
83 | Pickering
Methods
The schools examined in this study have implemented Sources of Strength, a suicide
preventive intervention. Of the 40 schools examined in the evaluation trial of this intervention,
20 were randomized to implement the intervention immediately and are used in this study. These
schools are located in rural, small town, and micropolitan areas of New York (n = 16) and North
Dakota (n = 4) and were recruited from regions with past five-year youth suicide rates above the
state averages. Participants were followed for four waves of data collection spanning two years,
but as this study addresses diffusion over the course of one school year, only data from baseline
and the end of the first school year are used. Of the 40 schools in the parent study, the 20
randomized to receive the Sources of Strength immediately are included in this study. Schools
ranged in size from 63-1,207 students (M=366).
Students were recruited in two phases: 1) at the beginning of the school year for the
school-wide Sources of Strength evaluation and 2) immediately following the first phase for the
selection of peer leaders. All students at school were invited to enroll in the evaluation study by
completing web-based assessments during the fall and spring over two years. Information letters
were sent to parents that allowed the option to decline their child’s participation. The University
of Rochester IRB approved the study protocol, and research personnel collected opt-out forms
and verbal assent with eligible students. Students were given information about how to access
help and support for themselves and other students if necessary.
Survey Variables
Demographics. The baseline survey administered to all students collected information on
student sex, ethnicity (white vs. nonwhite), and grade level.
84 | Pickering
Suicidal thoughts and behaviors. The Youth Risk Behavior Survey (Centers for Disease
Control and Prevention, 2010) was used to measure suicidal thoughts and behaviors. Each
student was asked whether in the preceding 12 months he/she had: seriously considered suicide;
planned suicide; made one or more suicide attempts; or made an attempt that resulted in injury
requiring medical treatment. Students were assigned to one of three categories indicating high to
low suicide risk: suicide attempt (SA) with or without injury, seriously considered suicide (SI)
without attempt, and no STB (NS).
Intervention exposure. Exposure to the Sources of Strength intervention was measured at
the end of the first school year and categorized into four different dichotomous exposure
modalities corresponding to various levels of engagement. This simulation study examined
diffusion of information. Namely, direct peer communication was measured by answering “yes”
to either: Has a friend or other student… (a) told you about Sources of Strength?, or (b) talked to
you about using strengths?
Network Variables
Information from the friendship network was used to construct network-level variables
describing the structure of each network. These variables included: (a) Out-degree: the number
of friendship nominations made. Students who named no friends and received no friendship
nominations were considered peer isolates. (b) Coreness: for each student, the k-core is the
maximal subgraph in which each vertex has degree k; a larger value indicates denser and more
cohesive friendship groups. (c) Closeness to peer leader: the length of the shortest path to a peer
leader, categorized into 1, 2, 3 or more steps, and not connected. (d) Adult out-degree: total
number of trusted adults named. Students who named no trusted adults were considered adult
isolates.
85 | Pickering
Peer Leader Selection
Peer leader selection was preceded by the training of adult advisors at school and baseline
assessments. Standardized procedures were implemented for recruitment across the schools.
Nomination forms were distributed to staff members at school which asked for nominations of
up to 6 students whose “voices are heard” by other students. Nominations were reviewed with an
attempt to select individuals who reflected diverse groups within their school, with a target of 5-
10% of the school population selected as a peer leader. The size of peer leader teams, therefore,
was dependent on school size and staff selection. A total of 959 students were invited (19 to 86
per school), with 798 (83.2%) enrolling with parent permission and youth assent. Of these, 459
(9-45 per school) were retained as active peer leaders.
Alternate Peer Leader Selection Criteria
The number and proportion of student body trained as peer leaders varied by school
(Figure 6). The demographic and sociometric characteristics of alternate peer leader sets are
described elsewhere (Study 1). For any given school with a set of n peer leaders, an alternative
set of n peer leaders was chosen by the following selection methods. Any time a ranked method
produced a tie, students were selected randomly to break the tie.
Opinion leaders. Students were asked to name up to three students in school who they
considered to be student leaders who others listen to, producing a total number of nominations
per student (opinion leader in-degree). The top n opinion leaders at each school were selected as
opinion peer leaders (OPL).
Friendship network. Students were asked to name up to seven students in school who are
their closest friends. These nominations were used to produce several individual-level network
86 | Pickering
variables. This information included: (a) In-degree: the number of friendship nominations
received; (b) Coreness: for each student, the k-core is the maximal subgraph in which each vertex
has degree k, with larger values indicating a position in a more highly interconnected group of
friends; (c) Closeness: the reciprocal sum of distances to each other student in the network,
indicating a central proximity to all other students; and (d) Betweenness: the number of times an
individual is in the shortest path connecting any two nodes. The iGraph package in R (Csardi &
Nepusz, 2006) was used to compute all individual-level friendship network variables. The top n
students for each metric were selected as sociometric peer leaders (SPL).
Key players. The key players algorithm (KPP-POS) is a technique that selects a set of
maximally connected individuals as seed nodes in a network in order to optimize diffusion
(Borgatti, 2006). The key players algorithm was performed using the InfluenceR package in R
(Jacobs, Khanna, & Madduri, 2016) to identify n individuals in each school as key players
(KPL).
Hybrid methods. Three hybrid methods of peer leader selection were created. In all cases,
representative samples of the population were taken by stratifying the school population by
ethnicity, gender, and grade level and choosing a proportional number of peer leaders within that
stratum, rounded down. Then, the key players algorithm was used to bring the total number of
peer leaders up to n per school. These custom peer leaders (CPL) were selected by the following
algorithms:
1) Influence-weighted: the students with the highest opinion leader in-degree and
friendship in-degree were chosen within each stratum.
2) Centrally-weighted: the students with the highest closeness and betweenness were
chosen within each stratum.
87 | Pickering
3) Structurally-weighted: peer leaders were initially chosen by highest opinion leader
and friendship in-degree but were constrained to a maximum of 2 individuals per
stratum. This resulted in a greater proportion of peer leaders being chosen through the
key players algorithm.
Risk Subpopulations
Four target populations were addressed in this study. These populations include:
(1) All students.
(2) Suicidal students: students with suicide ideation or attempt in the past 12 months.
(3) Peripheral students: students who were classified as being in the periphery of the
network and may be harder to reach (Suzuki, 2011).
(4) Isolated from adults: students who did not name any trusted adults in the network
section of the survey and would have a more difficult time seeking help for
themselves or others if necessary.
Figure 6 shows the distribution of at-risk students within the network of one sample school.
Figure 7 illustrates how the proportion of at-risk students varied across all schools. Across all
schools, the smallest risk group was suicidality (15.4% of students), followed by peripheral
students (16.2%), with many students not naming a trusted adult (32.0%).
Simulation Approach
Baseline network data was used to simulate diffusion of peer communication through the
20 school networks using the set of adult-selected peer leaders (APL), all alternate peer leader
selection sets, and a random set of starting seeds. Briefly, for nodes A and B where A nominated
B (A→B), at each time point there is a probability pBA that node B influenced node A.
88 | Pickering
Consistent with literature and empirical findings, characteristics of B were modeled to
modify the probability of influence. These characteristics fell into two domains.
1) Opinion leadership. Students who were identified as opinion leaders in their school
had higher exposure to peer communication in their 1-step networks (72.0% vs.
62.5%, p<.001) and 2-step networks (70.4% vs. 62.7%, p<.001). Individuals directly
connected to opinion leaders were 15% more likely to have had exposure to the
intervention than individuals who were not connected to peer leaders. This effect was
reinforced in the literature, as studies have found an effect of opinion leaders’
network position and personal characteristics on their ability to diffuse information
(Valente & Davis, 1999; Van Eck et al., 2011). In this sample, schools that trained a
higher proportion of opinion leaders as peer leaders were found to have empirically
higher diffusion (see Study 1). Thus, when B was identified as an opinion leader, pBA
was modeled to increase proportional to the normalized opinion leader in-degree of
B, by 15% on average
2) Homophily. Guided by theory on the diffusion of innovations, individuals who are
similar to others (i.e., homophilous) are considered more likely to spread information
to their like peers (Rogers, 2010). This finding has been replicated several times, and
homophily has been shown to explain from 15% to up to 50% of behavioral diffusion
(Aral, Muchnik, & Sundararajan, 2009; De Choudhury, Sundaram, John, Seligmann,
& Kelliher, 2010), though it is difficult to determine the true effect of homophily as it
is highly correlated with influence in diffusion processes (Shalizi & Thomas, 2011).
As such, if A and B are homophilous (i.e., share a characteristic) on gender or
ethnicity, pBA was modeled to increase by 15%.
89 | Pickering
At each time point, all influence probabilities associated with A become converted to
dichotomous influence events. Drawing from Valente’s work on individual thresholds in
networks (Valente, 1996), a component was added to the model to evaluate whether the
proportion of individuals influencing the given node is above a particular threshold. For
modeling the spread of peer communication, the threshold was set at 0 (i.e., any friend
communicating intervention messaging to an individual would expose the individual to that
information). This threshold is expected to be larger for behavioral change, such as participating
in a Sources of Strength activity. The steps involved in the simulation model are presented in
Table 8. Time points in the simulation correspond to months in the intervention (i.e., 9 months
from the beginning to the end of the first school year).
The effect of increasing social cohesion was also modeled. Intervention schools became
more dense over time, increasing density by 0.0011 from baseline to year 1 follow-up. As such,
the model was set to have a 0.1% chance at time point to form each non-existing tie. At the end
of each simulation the rate of exposure to the intervention was calculated, producing a
distribution of diffusion rates. The simulation approach was implemented in R.
Table 8. Steps involved in each iteration of the diffusion simulation.
Step Instruction
1 Baseline probabilities are assigned to each edge in the network. Student B can
influence Student A if there is a friendship nomination from A to B.
2
Baseline probabilities are modified according to the following criteria:
2a If B is an opinion leader, probability increases proportional to the normalized
opinion leader in-degree, by 15% on average.
2b
If A and B share gender, probability increases by 15%.
2c
If A and B share ethnicity, probability increases by 15%.
3
Probabilities are assigned to yes/no influence event status.
4
If the proportion of influencing events is greater than the threshold, A adopts.
5
Compute the proportion of the network now exposed to the intervention.
6
Rewire the network so that each missing edge has a 0.1% chance of formation.
90 | Pickering
The simulation was run in each of the 20 schools under each of the peer leader selection
methods. For each school and simulation method, the proportion of the target population exposed
at time 4 and time 9 (representing months from the start of the intervention) were computed.
Target populations included the whole school, suicidal students, peripheral students, and students
who were isolated from adults.
Figure 15. One iteration of diffusion simulation in one school. Dark red nodes are APL seeds.
Nodes are colored by time to exposure, from time 1 (dark orange) to time 9 (dark blue).
91 | Pickering
Regression Analysis
To determine how the effect of each selection method on diffusion at time 4 and time 9
depended on network structure, linear regression was performed. Predictors included selection
methods moderated by network structure variables, and the outcome was school-level average
rate of diffusion. These network structure variables included:
a) Degree centralization: the tendency for certain students to have many friendship
nominations and other students to have few friendship nominations. Networks with high
degree centralization have clearly popular students.
b) Closeness centralization: the tendency for certain students to be close to several other
students in the network and other students to be less close. Networks with high closeness
centralization have students who are clearly central in the network.
c) Betweenness centralization: the tendency for certain students to be bridges between
subgroups within the network. Networks with high betwenness centralization have
students who are clearly identified as brokers/bridges between subgroups.
d) Density: the number of ties present out of the number of all possible ties that could exist
(7N in a network of size N).
Simulations of Percent Peer Leaders
Simulation methods were additionally performed to determine the effect of selecting
more (or fewer) of a school population as peer leaders on intervention exposure. Simulation
conditions included selecting 5%, 10%, 15%, 20%, 25%, 30%, and 35% of the school as peer
leaders. The peer leader selection method that performed the best generally would be considered
for this analysis. Schools were classified into three size groups: small (<150 students), medium
(150-500 students), and large (>500 students) and diffusion after 4 months and 9 months were
92 | Pickering
averaged over schools within each size category. Again, the populations examined were all
students, students with suicide ideation or attempt, peripheral students, and students isolated
from adults at school.
Results
Overall Diffusion
The simulation model produced diffusion rates after 9 months for the APL method that
were consistent with the observed rates in these schools. There was good correlation between the
observed exposure and simulated exposure rates, at 0.64 (p = .002). This correlation is depicted
in Figure 16.
Figure 16. Correlation between observed exposure at T9 and average simulated exposure at T9
in 20 schools. Reference line of y=x is displayed in red.
The simulated diffusion curves (to the entire school population) are presented in Figure
17. Network-informed methods of peer leader selection appeared to perform better than random
93 | Pickering
methods or adult-selected methods, except in certain schools. The figure also demonstrates
variability in diffusion curves by method. In some schools (e.g., 10073) it was more difficult to
distinguish among the methods with respect to diffusion curves.
Figure 17. Simulated intervention diffusion over 9 time points in 20 schools, depending on peer
leader selection method. The outcome is percent of school exposed to the intervention.
Figure 18 shows the simulated exposure to the intervention averaged across all 20
schools after 4 months (T4) and 9 months (T9). In general, the random seeds did not perform
well at diffusing the intervention by the end of the first school year but did significantly better at
94 | Pickering
reaching peripheral students early on. Of all sociometric selection methods (i.e., degree,
coreness, closeness, betweenness), selecting peer leaders by degree consistently produced the
greatest diffusion in all subpopulations at all time points. Sociometric selection methods
generally performed poorer at influencing diffusion at 4 months compared to 9 months. When
considering diffusion at the end of the first school year, the custom peer leader selection methods
performed quite well, with the structurally-weighted algorithm (CPL5) producing the networks
with the greatest intervention diffusion. The CPL5 method also performed quite well at
providing quick diffusion, with high exposure rates at time 4. The adult-selection methods
performed quite well at diffusion to peripheral students, and in fact had the highest diffusion rate
to peripheral students after 4 months.
95 | Pickering
Figure 18. Simulated exposure to the intervention at time 9 and time 4 for several selection
methods. Displayed are the mean and 95% confidence interval of the school-level exposures
across 20 schools in four different sub-population.
96 | Pickering
Network Moderators of Diffusion
Because of the success of the degree peer leader selection method (vs. other sociometric
methods) and structurally-weighted algorithm (vs. other custom selection algorithms), these two
methods were used to represent the sociometric and custom methods, respectively, in regression
analyses. Exposure at time 9 was used as the outcome in these models, with predictors including
selection method (reference = APL) and several network measures. Network measures were
normalized to have a mean of 0 and standard deviation of 1. Interactions between selection
method and network variables were excluded if not significant at p=.05.
97 | Pickering
Table 9. Regression of selection method and network variables on intervention exposure after 9
months, in four subpopulations. The “network exposure variable” is indicated by the respective
columns, and exposure rate is the outcome in all models.
Exposure at T9, All Students
None Size Density Trans
Degree
Centralization
Closeness
Centralization
Betweenness
Centralization
Intercept 0.67 0.67 0.67 0.66 0.67 0.67 0.67
Network Var. - -0.02 0.13** 0.06* 0.03 0.04 0.05
Selection Method (ref = APL)
CPL5 0.09* 0.09* 0.09** 0.09* 0.09* 0.09* 0.09*
KPL 0.07 0.07 0.07** 0.07* 0.07 0.07 0.07
OPL 0.07 0.06 0.07** 0.07* 0.06 0.06 0.07
RAND -0.03 -0.03 -0.03 -0.03 -0.03 -0.03 -0.03
SPL-D 0.06 0.06 0.06* 0.06 0.06 0.06 0.06
Selection Method x Network
CPL5 x Network - NS NS NS NS NS NS
KPL x Network - NS NS NS NS NS NS
OPL x Network - NS NS NS NS NS NS
RAND x Network - NS NS NS NS NS NS
SPL-D x Network - NS NS NS NS NS NS
Exposure at T9, Suicidal Students
None Size Density Trans
Degree
Centralization
Closeness
Centralization
Betweenness
Centralization
Intercept 0.61 0.62 0.62 0.61 0.62 0.62 0.62
Network Var. - -0.01 0.12** 0.06* 0.03 0.02 0.05
Selection Method (ref = APL)
CPL5 0.06 0.06 0.06 0.07 0.06 0.06 0.06
KPL 0.06 0.06 0.06 0.06 0.05 0.05 0.06
OPL 0.02 0.02 0.02 0.03 0.01 0.01 0.02
RAND -0.05 -0.05 -0.05 -0.05 -0.05 -0.05 -0.05
SPL-D 0.01 0 0.01 0.01 -0.01 0 0
Selection Method x Network
CPL5 x Network - NS NS NS NS NS NS
KPL x Network - NS NS NS NS NS NS
OPL x Network - NS -0.12* -0.11* -0.10* NS -0.11*
RAND x Network - NS NS NS NS NS NS
SPL-D x Network - NS NS -0.11* NS NS -0.11*
98 | Pickering
Exposure at T9, Peripheral Students
None Size Density Trans
Degree
Centralization
Closeness
Centralization
Betweenness
Centralization
Intercept 0.39 0.39 0.39 0.39 0.39 0.39 0.39
Network Var.
0.02 0.08 0.03 0.01 0.02 0.01
Selection Method (ref = APL)
CPL5 0.02 0.02 0.02 0.02 0.01 0.01 0.02
KPL 0 -0.01 0 0 -0.01 -0.01 0
OPL 0 -0.01 -0.07 0 -0.01 -0.01 0
RAND -0.03 -0.03 -0.04 -0.03 -0.04 -0.04 -0.03
SPL-D -0.01 -0.02 -0.09 -0.01 -0.02 -0.02 -0.01
Selection Method x Network
CPL5 x Network - 0.08* NS -0.11* -0.10* NS -0.11*
KPL x Network - 0.08* NS NS -0.09* -0.11* NS
OPL x Network - 0.10* NS -0.10* -0.12** NS -0.12*
RAND x Network - NS NS NS NS NS NS
SPL-D x Network - 0.10** NS -0.11* -0.12** -0.11* -0.11*
Exposure at T9, Isolated from Adults
None Size Density Trans
Degree
Centralization
Closeness
Centralization
Betweenness
Centralization
Intercept 0.51 0.51 0.51 0.52 0.52 0.52 0.51
Network Var.
0.01 0.08* 0.02 0.01 0.04 0.01
Selection Method (ref = APL)
CPL5 0.08* 0.08* 0.08* 0.08* 0.08* 0.08* 0.08*
KPL 0.07 0.07 0.07* 0.07 0.07 0.07 0.07
OPL 0.06 0.06 0.06 0.06 0.06 0.06 0.06
RAND -0.02 -0.02 -0.02 -0.02 -0.02 -0.02 -0.02
SPL-D 0.06 0.05 0.06 0.06 0.05 0.05 0.06
Selection Method x Network
CPL5 x Network - NS NS NS NS NS NS
KPL x Network - NS NS NS NS NS NS
OPL x Network - NS NS NS NS NS NS
RAND x Network - NS NS NS NS NS NS
SPL-D x Network - NS NS NS NS NS NS
There was no interaction between selection method and network characteristics at
predicting diffusion to the general population. The CPL5 method did the best at diffusing the
intervention to this group. Beyond selection method, networks that were more dense and more
transitive had higher diffusion—schools that were 1 standard deviation more dense were
99 | Pickering
predicted to have 13% higher exposure at time 9 (p<.001), and schools that were 1 standard
deviation more transitive were predicted to have 6% higher exposure (p<.01).
Diffusion to students isolated from adults followed the same general trend as to the
general student population—diffusion was higher to students isolated from adults in schools that
were more dense (B = 0.08, p<.05). The CPL5 method also performed the best at diffusing the
intervention to this subgroup, reaching approximately 8% more adult-isolated students than the
adult selection method.
Significant interactions were observed in models examining suicidal students at school.
In this subgroup, density and transitivity increased intervention diffusion to suicidal students.
Network structure appeared to moderate the optimal selection method. That is, the opinion leader
selection method was more powerful when schools were less dense (interaction B = -0.12),
transitive (interaction B = -0.11), and centralized (interaction B = -0.10). Additionally, the
degree selection method performed better when schools had less transitivity (interaction B = -
0.11) and less betweenness centralization (interaction B = -0.11).
Diffusion to peripheral students appeared to be the most strongly related to network
structure. All methods performed better than the adult-selection method when schools were
larger (interaction B ranged from 0.08 to 0.10, p’s <.05). In addition, all methods performed
worse than the adult-selection method when schools were more highly centralized on degree
(interaction B ranged from -0.12 to -0.09, p’s <.05). The Key Players algorithm fared worse
when schools were more centralized on closeness, and the CPL5 and opinion leader method
performed worse when schools were more centralized on betweenness. The degree selection
approach performed worse when schools were centralized on any metric.
Percent Population Trained as Peer Leaders
100 | Pickering
As the CPL5 method generally produced the most diffusion in simulations, it was used as
the selection method when examining the percent of population trained as peer leaders.
Simulated diffusion curves dependent on percent of population trained as peer leaders are shown
in Figure 19.
The effect of adding more peer leaders was greater when considering diffusion after 4
months, but it was more difficult to achieve diffusion at this time point in large schools and to
peripheral nodes. In fact, diffusion to peripheral nodes in large schools was less than 10%, even
when 50% of the school population was trained as peer leaders. In medium and large schools,
exposure was generally never greater than 50% in at-risk subpopulations. Smaller schools,
though, continued to benefit when adding additional peer leaders.
101 | Pickering
Figure 19. Simulated diffusion to the entire network (ALL), peripheral students (PERI), suicidal
students (SUIC), and students isolated from trusted adults (TAS), in small (1), medium (2), and
large (3) schools. Simulated diffusion is shown at T4 and T9.
102 | Pickering
When considering diffusion after 9 months, schools of all size were able to achieve
greater than 75% exposure. Small schools achieved this by training 15% of the school as peer
leaders, and medium and large schools by training 20%. Among all school sizes, 75% of suicidal
students were reached by training 30% of the school population as peer leaders. It was more
difficult to reach students who were isolated from adults, but still over 50% of this subpopulation
was reached by training only 10% of the school population as peer leaders.
103 | Pickering
Figure 19. School where sociometric selection had the least impact (top, school 10073) and
greatest impact (bottom, school 10070) on diffusion to all students after 9 months.
104 | Pickering
Discussion
It has been hypothesized that the use of sociometric methods to inform selection of
“seeds” when trying to diffuse an intervention or innovation is superior to methods that do not
use such information (Valente et al., 2004). This study attempted to simulate diffusion through
20 school networks participating in the Sources of Strength intervention to determine if certain
selection methods would diffuse knowledge of the intervention better than the current adult-
selected method. The structurally-weighted algorithm (CPL5), which selects some localized
opinion leaders of diverse racial backgrounds and genders and some “key players” spaced in
strategic network positions, performed the best across a range of school sizes and network
characteristics.
The effect of network structure on diffusion confirm the findings of other studies. In this
simulation, we found that networks that were more dense and had higher transitivity had greater
exposure to the intervention, confirming findings of empirical studies on diffusion (Brown et al.,
2012; Rulison, Gest, et al., 2015). However, a handful of studies found centralization to be a
barrier to diffusion. This was not the case in this study—the centralization of networks did not
affect diffusion to the general population. Centralization did affect diffusion to some at-risk
students (peripheral and suicidal), though, when peer leader seeds were chosen by sociometric
methods such as degree or opinion leadership. When peer leaders were chosen in this fashion,
centralization did act as a barrier to the spread of the intervention to the periphery of the network
and to suicidal students who tend to be located in sparser regions of the network (Wyman et al.,
2018). And, while it may be the case that centrality matters more for diffusion in centralized
networks (Valente, 2010), choosing a set of individuals based on centrality is detrimental to
diffusion to the periphery and sparse pockets of the network in centralized networks.
105 | Pickering
It did appear that there was variation in diffusion depending on network structure—in
fact, some schools had no distinguishable difference among selection methods while some had
quite a stark difference. For example, in school 10073 which was small but high in modularity,
all network-informed methods produced lower diffusion than randomly selected seeds (except
CPL5, which produced 2.2% higher exposure; Figure 16, School 10073). This appeared to occur
more in smaller schools than larger schools. The school with the highest difference in diffusion
between random and network-informed methods was School 10070, in which the CPL5 method
produced 20% greater exposure than random methods. A comparison of the structure of these
schools is available in Figure 19.
Diffusion to the peripheral subpopulation was most dependent on network structure. In
fact, it seemed that in order to reach peripheral students at an early time point (T4), it was best to
employ random or adult-selected seeds. The random and adult-selected methods performed best
when trying to diffuse information to peripheral students (i.e., in areas of the network that were
less reachable by conventional methods). Compared to sociometric selection, random selection is
more likely to choose students who are on the periphery who may quickly hear about the
intervention, but diffusion beyond these subgroups appears to be limited as time goes on. This is
also the case for students chosen by adults. An additional benefit of adult selection is that they
may be able to tap into characteristics of these students that are not assessed through these
surveys, such as personality type, that may make these students good candidates as peer leaders.
Adult-selected methods produced low diffusion to students who were isolated from adults, but
this is unsurprising; adults would be less likely to nominate students to whom they do not have
connections.
106 | Pickering
It appeared that density was an important characteristic for reaching suicidal students.
This may have important implications for the Sources of Strength intervention. Optimizing
diffusion to suicidal students may not simply be a matter of peer leader selection, but may also
be affected by one of the intervention’s aims: increasing social integration and facilitating
connections to other students and capable adults at school. By increasing ties among all students
and not just popular students, several forms of centralization would likely decrease and make
several sociometric selection methods more powerful.
School size played a large part in the effectiveness of certain selection methods. Opinion
leaders, popular students, key players, and individuals chosen by the CPL5 method were better
able to reach peripheral students at large schools, and those chosen by the CPL5 method were
better able to reach students isolated from adults at large schools, than peer leaders chosen by
adults. This finding emphasizes the importance of strategic peer leader selection in larger
schools. In smaller schools the selection of peer leaders in suboptimal locations may not matter
as much, because these schools are more cohesive and students are generally close to each other.
In larger schools, there may be a large penalty for selecting peer leaders in suboptimal network
positions. Peer-to-peer diffusion therefore becomes a more powerful mechanism of exposing
students to the intervention in large schools compared to small schools.
Opinion leaders have a powerful effect of disseminating information and influencing
peers (Valente & Davis, 1999; Van Eck et al., 2011). This effect is due to both their position
within the network and their role as respected members of the community. One accepted
hypothesis about peer leaders’ ability to diffuse information networks relates to a cascade of
influence. That is, opinion leaders influence other people in the network of moderately high
influence, who in turn influence others with lesser influence (Watts & Dodds, 2007). It appears
107 | Pickering
that this works quite well in most situations, but in some cases, such as when schools are more
transitive and when there is high betweenness centralization, opinion leaders and popular
students are not optimal seeds. This may be because highly transitive schools have a
reinforcement effect—transitivity increases the chance that information will stay in highly
transitive pockets of the network. Similarly, when there are clear bridges in the network (i.e.,
high betweenness centralization), this may indicate cliques of individuals who are bridged by
few gatekeepers; if these gatekeepers are not reached, information may not pass between the two
groups.
The CPL5 selection method was consistently good at maximizing diffusion to a range of
target populations under different conditions and seems to be a good all-purpose method to
increase exposure via diffusion at intervention schools. It does require knowledge of the entire
network, though, which can be prohibitive in some situations compared to, say, taking the sum of
opinion leader nominations. However, random and adult-selected peer leaders did not perform
too well in these simulations (though adult-selected peer leaders are close to suicidal students)
and sociometric method are less representative of the student body (Study 1). We also found that
adult-selected methods tended to select few opinion leaders—only 20% of adult-selected peer
leaders were opinion leaders. Selection alternatives that do not use as much network information
are the CPL1 approach, which does not rely as heavily on the Key Players algorithm. This
method stratifies by key demographic characteristics (i.e., sex, ethnicity, and grade level) and
uses the most popular students and opinion leaders within each stratum, an approach that is
easier to implement, in comparison. For this reason, it seems to be an acceptable alternative in
situations when the entire network cannot be mapped or computing resources not available to run
the Key Players algorithm.
108 | Pickering
Peer leader models have generally recommended identifying approximately 10-15% of
the population as leaders (Valente & Pumpuang, 2007). This study expands on that
recommendation by examining different diffusion goals (diffusion at an early vs. late time point)
and different sub-populations. The marginal increase to diffusion after 9 months seemed quite
low in simulations where more than 15% of the population was trained as peer leaders. A greater
percent of the school may need to be trained to reach at-risk subpopulations, though. There was a
greater marginal increase in exposure for adding additional peer leaders when looking at
diffusion after 4 months vs. 9 months. This effect was especially stark for reaching at-risk
subgroups in small schools. On the other hand, even when selecting a high proportion of trained
peer leaders there was very little diffusion to at-risk subgroups in large schools. Under these
situations, the peer leader diffusion model may not be the best for disseminating information
quickly to at-risk subgroups. The potential of these peer-led interventions may lie in their ability
to diffuse information over the entire school year.
This study benefits from several strengths. The study was unique in that it was able to
draw on data from 20 schools of differing sizes and network structures. The schools had a high
overall response rate, which permitted an accurate representation of the network structure. We
also examined a diffusion phenomenon with fewer assumptions related to it. That is, diffusion of
information is more likely to happen between individuals than changes in attitude or behavior.
When examining the latter two, models may have to make assumptions about the number of
friends influencing an individual and thresholds to adoption.
Because of the modeling and simulation approach, several assumptions were made in this
study. Namely, that diffusion happens within the parameters set by this model (see Table 8).
There is some evidence of diffusion within networks independent of proximity to a peer leader
109 | Pickering
(see Study 2). That is, there may be influence within these networks that is not confined to
friendship ties, such as diffusion through informal routes of communication such as
acquaintances.
Still, these models can provide valuable information for the Sources of Strength study
and are generalizable to other interventions attempting to disseminate information throughout a
network. By examining diffusion to the entire population as well as several at-risk subgroups,
other interventions can use this information to determine the best selection method to use based
on the target populations within the network and the speed at which this information must be
disseminated. At the least, the network-informed selection in this study surpassed uninformed
methods of selection and have the ability to more optimally diffuse information through social
networks in health-related interventions.
110 | Pickering
Conclusions
In his 2012 paper, Valente discusses the power of using social network data to accelerate
behavior change in interventions (Valente, 2012). These network interventions include: 1) the
individual change agent approach, in which seeds in the network are identified to champion the
diffusion of information, an innovation, or behavior; 2) segmentation, in which an identified
subgroup adopts the innovation at the same time; 3) induction, in which powerful sources such as
media create buzz within a community and the innovation travels through word-of-mouth; and 4)
alteration, in which the networks are changed by adding, deleting, or otherwise altering ties of
the social network. This dissertation contributes to the literature about the individual change
agent approach and addresses several unanswered questions on the use of change agents (“peer
leaders,” in these studies) and their ability to diffuse an intervention through a network.
The first study examined the characteristics of peer leaders chosen empirically by adults
at school and several alternate peer leader sets chosen through various mechanisms that make use
of network data. There are several published algorithms for identifying which nodes (i.e.
individuals) in networks are most central (Freeman, 1978). We selected several of these—
degree, coreness, closeness, and betweenness—to evaluate sociometric selection methods. In
addition, we selected alternate peer leader sets based on opinion leadership (i.e., number of
nominations of individuals in the network who were respected and others listen to) and through
the Key Players algorithm. To our knowledge, this study was the first to show how the
demographic composition of these alternate peer leader sets varied depending on selection
method.
This is an especially important concept to address in diffusion studies. In the literature we
identified three characteristics that sets of “seeds” should have to promote successful diffusion,
111 | Pickering
including opinion leadership, strategic network positioning, and representativeness. It is alarming
to see that individuals chosen by almost every network-informed method (except betweenness
and Key Players) consisted of more white students than the general population. The adult-
selected methods produced peer leader sets that were more female than the general population. If
peer leader sets are unrepresentative, it may compromise the ability of peer leaders to diffuse the
intervention to groups at school that are less represented in these seeds. Surprisingly, the Key
Players algorithm produced peer leader sets whose demographic composition aligned with that of
the general population—even though this demographic information was not a part of the
algorithm’s selection method.
To address demographic representativeness, we developed three different custom
selection methods (CPL1, CPL3, CPL5) that used demographic stratification to produce peer
leader sets. Demographic stratification has been used before, but there is less information on its
use as part of a broader selection algorithm (e.g., Starkey et al., 2009). After stratifying by
demographic characteristics, the individuals with the highest values of popularity and opinion
leadership were selected (CPL1), or the individuals with the highest closeness and betweenness
were selected (CPL3), or some popular opinion leaders were selected along with several students
chosen by the Key Players algorithm (CPL5). In addition to producing demographically
representative samples of the network, we found empirical evidence that schools that happened
to choose peer leaders that aligned with these selection methods produced greater rates of
exposure to poster/video and peer communication modalities. Aside from these custom methods,
choosing the top opinion leaders as peer leaders also led to higher rates of exposure to
poster/video and direct peer communication modalities.
112 | Pickering
One important distinction made in this dissertation is between individuals who are
opinion leaders and those who are popular in the network. While these groups may both tend to
hold central positions in the network, the biggest demographic distinction between the two was
with their age; students selected as opinion leaders were older than the general population
whereas those selected as popular students (i.e., high in-degree) were younger than the general
population. Several studies have stressed the importance of peer leaders on diffusion (Flodgren et
al., 2007; Valente & Davis, 1999; Valente & Pumpuang, 2007; Van Eck et al., 2011), though it is
not clear exactly what constitutes an opinion leader; this definition can vary across studies. In
this study we show that students at school who are respected and other people listen to are more
powerful at diffusing the intervention than those who are simply popular.
Being an opinion leader does not necessarily mean that this status holds throughout the
entire school; rather, opinion leaders may act strongest on the local group of friends they care
about. Some studies have looked at specific groups of individuals within a network first and then
chosen opinion leaders within those groups (Buller et al., 1999). The custom selection methods
identified in this study, though, have a similar effect. Instead of selecting group-specific opinion
leaders, these methods can select demographically-representative opinion leaders. This aligns
more closely to the method of Buller et al. (1999) than selecting global opinion leaders.
The second study considered the different ways that support can be given to peer leaders
to make them more powerful agents of change within schools. We modeled the effect of three
types of support (training, adult, peer) at the individual and school level on four different
exposure modalities. These modalities varied in the amount of peer engagement required for
exposure from low (exposure to a presentation) to high (participation in an activity). As Valente
notes, “there is little to no evidence on what level of training is necessary for identified leaders to
113 | Pickering
be effective change agents” (Valente, 2017). Surprisingly, the effect of training at the individual
and school level did not appear to affect exposure to the intervention. This suggests that the
range of training in these schools is adequate and that a minimal amount of training is required to
get students exposed to the intervention. The one exposure modality that was affected by
training, though, was activity participation.
Peer support—training with friends who were also peer leaders—also increased diffusion
across all modalities. This finding addressed the “redundancy problem” of sociometric seed
selection methods, where the use of one particular metric will tend to produce seeds that cluster
in the network and have considerable overlap (i.e., redundancy) among the individuals that can
be reached (Borgatti, 2006). The peer leader selection methods from Study 1 produced peer
leader sets in which individuals were connected on average to two to almost four other peer
leaders, depending on the selection approach. In this study, though, we found that this wasn’t a
“problem” per se. In fact, there is increased exposure within peer leaders’ egocentric networks
when they train with one friend—though the added benefit after one friend is marginal.
The results with regard to peer support and training indicate an important distinction
between the diffusion of information compared to diffusion of behavior change. It appears one
way to get individuals to participate in an activity is to have peer leaders who are more highly
trained and who cluster together. Furthermore, the effect of training and peer support was seen in
the 1-step egocentric networks but not in the 2-step egocentric networks. This indicates an
immediacy to the effect peer leaders can have on activity participation. Information may spread
through these peer leader egocentric networks, but the spread of activity participation is far more
limited.
114 | Pickering
Taken together, the results of the second study expand the literature on how school-based
peer-led interventions can be more effective. Training is important when getting students
engaged, adult support is important for creating an environment that promotes discourse about
the intervention, and training with a friend does not actually lead to an undesirable redundancy
problem. Furthermore, there is an influence that peer leaders have that extends beyond their
friendship ties. That is, schools with a greater proportion of peer leaders trained had higher
exposure within peer leaders’ egocentric networks, even when adjusting for the effect of
individual peer leader support. Diffusion is not simply limited to friendship ties but, rather, can
occur through informal encounters (or, “weak ties”) with students who did not report being
friends at school.
The results of the first two studies were used to inform the peer leader diffusion model
created in the third study. Namely, we were able to confirm that opinion leaders are effective not
simply because of their position in the network, but rather because of their potential for
influence. We also showed that peer leaders who clustered did not experience a redundancy
penalty but, rather, had increased diffusion in their egocentric networks; custom selection
methods therefore were not designed to avoid having peer leaders who were friends with each
other.
The third study used a simulation approach across 20 different schools, 11 selection
methods, and 4 subpopulations with 1,000 simulations per condition to model diffusion
throughout the network. The simulation approach modeled diffusion using parameters that
approximated real life, as a function of opinion leadership, homophily, and an expected increase
in cohesiveness over time that is expected in school networks over the course of the academic
year. This study has implications for the way Sources of Strength, and other interventions, select
115 | Pickering
peer leaders. When sociometric information is available, it may be beneficial to use that
information to increase the effectiveness of the chosen peer leader sets. While selection method
is not as important in smaller schools, larger schools—with the potential to reach more
students—increasingly benefit from network-informed peer selection methods.
Social network thresholds help explain why people do or do not adopt (Valente, 1996).
Though thresholds were considered to have a negligible effect on communication (the diffusion
of information), the ability to include thresholds was incorporated into the simulation model for
future study. As a result, additional simulation studies will be able to use this information to
model phenomena that require greater influence than spreading a message.
The studies performed in this dissertation benefit from being conducted across several
high schools of differing sizes and network compositions. While the networks in these schools
are empirical, the simulation approach is only a model of real life and likely does not represent
the exact way diffusion occurs in these schools. A future direction could be to implement a
randomized controlled trial in which two proven methods—adult-selected (control) and CPL5
selection—are both implemented in trial schools to properly compare the effectiveness of each.
The scope of health behaviors that can be informed by these models is not limited to a
suicide preventive intervention. Other types of interventions that rely on the spread of
communication can benefit from the information in these studies. There may, though, be cultural
barriers that prevent diffusion of information when an intervention contains topics that are
sensitive or taboo in the community of interest. Other than this, these findings seem translatable
to other types of health behaviors.
Peer-led interventions have shown great promise in reducing the rates of suicidality in
schools. This dissertation demonstrates that, while these interventions are effective, there is still
116 | Pickering
room to improve peer leader selection methods in a way that maximizes peer leaders’ reach
within schools. The information gained from these studies have real-world implications that can
be useful for, but are by no means limited to, the Sources of Strength intervention.
117 | Pickering
Literature Cited
Aral, S., Muchnik, L., & Sundararajan, A. (2009). Distinguishing influence-based contagion from
homophily-driven diffusion in dynamic networks. Proceedings of the National Academy of Sciences,
106(51), 21544–21549.
Aseltine, R. H., James, A., Schilling, E. A., & Glanovsky, J. (2007). Evaluating the SOS suicide
prevention program: a replication and extension. BMC Public Health, 7(1), 161.
Backett-Milburn, K., & Wilson, S. (2000). Understanding peer education: insights from a process
evaluation. Health Education Research, 15(1), 85–96.
Bates, D., Maechler, M., Bolker, B., Walker, S., Christensen, R. H. B., Singmann, H., … Green, P.
(2017). Package ‘lme4.’ R Package Version, 1–1.
Bearman, P. S., & Moody, J. (2004). Suicide and friendships among American adolescents. American
Journal of Public Health, 94(1), 89–95.
Black, D. R., Tobler, N. S., & Sciacca, J. P. (1998). Peer helping/involvement: an efficacious way to meet
the challenge of reducing alcohol, tobacco, and other drug use among youth. Journal of School Health,
68(3), 87–93.
Bloor, M., Frankland, J., Langdon, N. P., Robinson, M., Allerston, S., Catherine, A., … others. (1999). A
controlled evaluation of an intensive, peer-led, schools-based, anti-smoking programme. Health
Education Journal, 58(1), 17–25.
Borgatti, S. P. (2006). Identifying sets of key players in a social network. Computational & Mathematical
Organization Theory, 12(1), 21–34.
Borgatti, S. P., & Everett, M. G. (2000). Models of core/periphery structures. Social Networks, 21(4),
375–395.
Brener, N. D., Simon, T. R., Anderson, M., Barrios, L. C., & Small, M. L. (2002). Effect of the incident at
Columbine on students’ violence-and suicide-related behaviors. American Journal of Preventive
Medicine, 22(3), 146–150.
Brent, D. A., & Brown, C. H. (2015). Effectiveness of school-based suicide prevention programmes. The
Lancet, 385(9977), 1489–1491.
Brown, C. H., Kellam, S. G., Kaupert, S., Muthén, B. O., Wang, W., Muthén, L. K., … others. (2012).
Partnerships for the design, conduct, and analysis of effectiveness, and implementation research:
Experiences of the Prevention Science and Methodology Group. Administration and Policy in Mental
Health and Mental Health Services Research, 39(4), 301–316.
Buller, D. B., Morrill, C., Taren, D., Aickin, M., Sennott-Miller, L., Buller, M. K., … Wentzel, T. M.
(1999). Randomized trial testing the effect of peer education at increasing fruit and vegetable intake.
Journal of the National Cancer Institute, 91(17), 1491–1500.
118 | Pickering
Burns, J., Dudley, M., Hazell, P., & Patton, G. (2005). Clinical management of deliberate self-harm in
young people: the need for evidence-based approaches to reduce repetition. Australian and New Zealand
Journal of Psychiatry, 39(3), 121–128.
Centers for Disease Control and Prevention. (2009). Connectedness as a strategic direction for the
prevention of suicidal behavior. Atlanta, GA: Author.
Centers for Disease Control and Prevention. (2010). Youth Risk Behavior Survey Questionnaire.
Retrieved from www.cdc.gov/yrbs
Centers for Disease Control and Prevention. (2012). Strategic direction for the prevention of suicidal
behavior: Promoting individual, family, and community connectedness to prevent suicidal behavior.
Atlanta, GA: Author. Available at: Http://Www. Cdc.
Gov/ViolencePrevention/Pdf/Suicide_Strategic_Direction_Full_Version-a. Pdf.
Centers for Disease Control and Prevention. (2017). QuickStats: Suicide Rates for Teens Aged 15-19
Years, by Sex - United States, 1975-2015. MMWR Morb Mortal Wkly Rep, 2017(66), 816.
http://dx.doi.org/10.15585/mmwr.mm6630a6
Chami, G. F., Ahnert, S. E., Kabatereine, N. B., & Tukahebwa, E. M. (2017). Social network
fragmentation and community health. Proceedings of the National Academy of Sciences, 201700166.
Choi, H., Kim, S.-H., & Lee, J. (2010). Role of network structure and network effects in diffusion of
innovations. Industrial Marketing Management, 39(1), 170–177.
Christakis, N. A., & Fowler, J. H. (2007). The spread of obesity in a large social network over 32 years.
New England Journal of Medicine, 357(4), 370–379.
Christakis, N. A., & Fowler, J. H. (2008). The collective dynamics of smoking in a large social network.
New England Journal of Medicine, 358(21), 2249–2258.
Ciffone, J. (2007). Suicide prevention: An analysis and replication of a curriculum-based high school
program. Social Work, 52(1), 41–49.
Coleman, J. S., Katz, E., & Menzel, H. (1966). Medical innovation: A diffusion study. Bobbs-Merrill Co.
Corrigan, P. W., Druss, B. G., & Perlick, D. A. (2014). The impact of mental illness stigma on seeking
and participating in mental health care. Psychological Science in the Public Interest, 15(2), 37–70.
Cowie, H., & Olafsson, R. (2000). The role of peer support in helping the victims of bullying in a school
with high levels of aggression. School Psychology International, 21(1), 79–95.
Csardi, G., & Nepusz, T. (2006). The igraph software package for complex network research.
InterJournal, Complex Systems, 1695.
Cuijpers, P. (2002). Effective ingredients of school-based drug prevention programs: A systematic
review. Addictive Behaviors, 27(6), 1009–1023.
119 | Pickering
Cusimano, M. D., & Sameem, M. (2011). The effectiveness of middle and high school-based suicide
prevention programmes for adolescents: a systematic review. Injury Prevention, 17(1), 43–49.
De Choudhury, M., Sundaram, H., John, A., Seligmann, D. D., & Kelliher, A. (2010). “ Birds of a
Feather”: Does User Homophily Impact Information Diffusion in Social Media? ArXiv Preprint
ArXiv:1006.1702.
Eaton, D. K., Kann, L., Kinchen, S., Shanklin, S., Ross, J., Hawkins, J., … others. (2008). Youth risk
behavior surveillance–United States, 2007. Morbidity and Mortality Weekly Report. Surveillance
Summaries (Washington, DC: 2002), 57(4), 1–131.
Eaton, D. K., Kann, L., Kinchen, S., Shanklin, S., Ross, J., Hawkins, J., … others. (2010). Youth risk
behavior surveillance—United States, 2009. MMWR Surveill Summ, 59(5), 1–142.
Flodgren, G., Parmelli, E., Doumit, G., Gattellari, M., O’Brien, M. A., Grimshaw, J., & Eccles, M. P.
(2007). Local opinion leaders: effects on professional practice and health care outcomes. Cochrane
Database of Systematic Reviews, (1).
Fong, C. M. (1993). A longitudinal study of the relationships between overload, social support, and
burnout among nursing educators. Journal of Nursing Education, 32(1), 24–29.
Fong, C. M. (2016). Role overload, social support, and burnout among nursing educators. Journal of
Nursing Education, 29(3), 102–108.
Forrest, S., Strange, V., & Oakley, A. (2002). A comparison of students’ evaluations of a peer-delivered
sex education programme and teacher-led provision. Sex Education: Sexuality, Society and Learning,
2(3), 195–214.
Frankham, J. (1998). Peer education: The unauthorised version. British Educational Research Journal,
24(2), 179–193.
Freedenthal, S. (2010). Adolescent Help-Seeking and the Yellow Ribbon Suicide Prevention Program: An
Evaluation. Suicide and Life-Threatening Behavior, 40(6), 628–639.
Freeman, L. C. (1978). Centrality in social networks conceptual clarification. Social Networks, 1(3), 215–
239.
Gardner, W., Klima, J., Chisolm, D., Feehan, H., Bridge, J., Campo, J., … Kelleher, K. (2010).
Screening, triage, and referral of patients who report suicidal thought during a primary care visit.
Pediatrics, 125(5), 945–952.
Goldston, D. B., Walrath, C. M., McKeon, R., Puddy, R. W., Lubell, K. M., Potter, L. B., & Rodi, M. S.
(2010). The Garrett Lee Smith memorial suicide prevention program. Suicide and Life-Threatening
Behavior, 40(3), 245–256.
Gould, M. S., Marrocco, F. A., Hoagwood, K., Kleinman, M., Amakawa, L., & Altschuler, E. (2009).
Service use by at-risk youths after school-based suicide screening. Journal of the American Academy of
Child & Adolescent Psychiatry, 48(12), 1193–1201.
120 | Pickering
Gould, M. S., Marrocco, F. A., Kleinman, M., Thomas, J. G., Mostkoff, K., Cote, J., & Davies, M.
(2005). Evaluating iatrogenic risk of youth suicide screening programs: a randomized controlled trial.
Jama, 293(13), 1635–1643.
Granovetter, M. (1983). The strength of weak ties: A network theory revisited. Sociological Theory, 201–
233.
Hall, M., Cross, D., Howat, P., Stevenson, M., & Shaw, T. (2004). Evaluation of a school-based peer
leader bicycle helmet intervention. Injury Control and Safety Promotion, 11(3), 165–174.
Hamdan, S., Story, M., French, S. A., Fulkerson, J. A., & Nelson, H. (2005). Perceptions of adolescents
involved in promoting lower-fat foods in schools: associations with level of involvement. Journal of the
American Dietetic Association, 105(2), 247–251.
Harden, A., Oakley, A., & Oliver, S. (2001). Peer-delivered health promotion for young people: a
systematic review of different study designs. Health Education Journal, 60(4), 339–353.
Holliday, J. (2006). Identifying and using influential young people for informal peer-led health
promotion. Cardiff University.
Hooven, C., Herting, J. R., & Snedker, K. A. (2010). Long-term outcomes for the promoting CARE
suicide prevention program. American Journal of Health Behavior, 34(6), 721–736.
Hunter, G., Ward, J., & Power, R. (1997). Research and development focusing on peer intervention for
drug users. Drugs: Education, Prevention and Policy, 4(3), 259–270.
Hurdle, D. E. (2001). Social support: A critical factor in women’s health and health promotion. Health &
Social Work, 26(2), 72–79.
Iyengar, R., Van den Bulte, C., Eichert, J., & West, B. (2011). How Social Network and Opinion Leaders
Affect the Adoption of New Products. GfK Marketing Intelligence Review, 3(1), 16–25.
Iyengar, R., Van den Bulte, C., & Valente, T. W. (2011). Opinion leadership and social contagion in new
product diffusion. Marketing Science, 30(2), 195–212.
Jacobs, S., Khanna, A., & Madduri, K. (2016). Bader D. influenceR: Software Tools to Quantify
Structural Importance of Nodes in a Network. 2015. URL: Https://Cran. r-Project.
Org/Web/Packages/InfluenceR/Index. Html [WebCite Cache ID 72BTZVmZT].
Kaess, M., Brunner, R., Parzer, P., Carli, V., Apter, A., Balazs, J. A., … others. (2014). Risk-behaviour
screening for identifying adolescents with mental health problems in Europe. European Child &
Adolescent Psychiatry, 23(7), 611–620.
Kelly, J. A., Amirkhanian, Y. A., Kabakchieva, E., Vassileva, S., Vassilev, B., McAuliffe, T. L., …
others. (2006). Prevention of HIV and sexually transmitted diseases in high risk social networks of young
Roma (Gypsy) men in Bulgaria: randomised controlled trial. Bmj, 333(7578), 1098.
121 | Pickering
Kempe, D., Kleinberg, J., & Tardos, É. (2003). Maximizing the spread of influence through a social
network. In Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery
and data mining (pp. 137–146). ACM.
Kim, D. A., Hwong, A. R., Stafford, D., Hughes, D. A., O’Malley, A. J., Fowler, J. H., & Christakis, N.
A. (2015). Social network targeting to maximise population behaviour change: a cluster randomised
controlled trial. The Lancet, 386(9989), 145–153.
King, K. A., Strunk, C. M., & Sorter, M. T. (2011). Preliminary effectiveness of surviving the Teens®
Suicide Prevention and Depression Awareness Program on adolescents’ suicidality and self-efficacy in
performing help-seeking behaviors. Journal of School Health, 81(9), 581–590.
Komro, K. A., Perry, C. L., Murray, D. M., Veblen-Mortenson, S., Williams, C. L., & Anstine, P. S.
(1996). Peer-Planned Social Activities for Preventing Alcohol Use Among Young Adolescents. Journal
of School Health, 66(9), 328–334.
Komro, K. A., Perry, C. L., Veblen-Mortenson, S., Williams, C. L., & Roel, J. P. (1999). Peer leadership
in school and community alcohol use prevention activities. Journal of Health Education, 30(4), 202–208.
Lenzer, J. (2012). Controversial mental health program closes down. BMJ: British Medical Journal
(Online), 345.
Li, J., Weeks, M. R., Borgatti, S. P., Clair, S., & Dickson-Gomez, J. (2012). A social network approach to
demonstrate the diffusion and change process of intervention from peer health advocates to the drug using
community. Substance Use & Misuse, 47(5), 474–490.
LoMurray, M. (2005). Sources of strength facilitators guide: Suicide prevention peer gatekeeper training.
Bismarck, ND: The North Dakota Suicide Prevention Project.
March, J. S., Silva, S., Petrycki, S., Curry, J., Wells, K., Fairbank, J., … others. (2007). The Treatment for
Adolescents With Depression Study (TADS): long-term effectiveness and safety outcomes. Archives of
General Psychiatry, 64(10), 1132–1144.
McGuire, L., & Flynn, L. (2003). The Columbia TeenScreen Program: Screening youth for mental illness
and suicide. TEN, 5(2), 56–62.
Mellanby, A. R., Rees, J. B., & Tripp, J. H. (2000). Peer-led and adult-led school health education: a
critical review of available comparative research. Health Education Research, 15(5), 533–545.
Menacho, L. A., Galea, J. T., & Young, S. D. (2015). Feasibility of recruiting peer educators to promote
HIV testing using Facebook among men who have sex with men in Peru. AIDS and Behavior, 19(2), 123–
129.
Milburn, K. (1995). A critical review of peer education with young people with special reference to
sexual health. Health Education Research, 10(4), 407–420.
Miller, W., & MacGilchrist, L. (1996). A model for peer-led work. Health Education, 96(2), 24–29.
122 | Pickering
Noe, R. A. (1986). Trainees’ attributes and attitudes: Neglected influences on training effectiveness.
Academy of Management Review, 11(4), 736–749.
Noe, R. A., & Wilk, S. L. (1993). Investigation of the factors that influence employees’ participation in
development activities. Journal of Applied Psychology, 78(2), 291.
Orpinas, P., Kelder, S., Frankowski, R., Murray, N., Zhang, Q., & McAlister, A. (2000). Outcome
evaluation of a multi-component violence-prevention program for middle schools: the Students for Peace
project. Health Education Research, 15(1), 45–58.
Pearlman, D. N., Camberg, L., Wallace, L. J., Symons, P., & Finison, L. (2002). Tapping youth as agents
for change: evaluation of a peer leadership HIV/AIDS intervention. Journal of Adolescent Health, 31(1),
31–39.
Pena, J. B., & Caine, E. D. (2006). Screening as an approach for adolescent suicide prevention. Suicide
and Life-Threatening Behavior, 36(6), 614–637.
Perry, C. L., Williams, C. L., Komro, K. A., Veblen-Mortenson, S., Stigler, M. H., Munson, K. A., …
Forster, J. L. (2002). Project Northland: Long-term outcomes of community action to reduce adolescent
alcohol use. Health Education Research, 17(1), 117–132.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of persuasion. In
Communication and persuasion (pp. 1–24). Springer.
Phelps, F., Mellanby, A., Crichton, N., & Tripp, J. (1994). Sex education: the effect of a peer programme
on pupils (aged 13-14 years) and their peer leaders. Health Education Journal, 53(2), 127–139.
Pickering, T. A., Wyman, P. A., Schmeelk-Cone, K., Hartley, C., Valente, T. W., Pisani, A. R., …
LoMurray, M. (2018). Diffusion of a Peer-Led Suicide Preventive Intervention Through School-Based
Student Peer and Adult Networks. Frontiers in Psychiatry, 9, 598.
https://doi.org/10.3389/fpsyt.2018.00598
Pisani, A. R., Schmeelk-Cone, K., Gunzler, D., Petrova, M., Goldston, D. B., Tu, X., & Wyman, P. A.
(2012). Associations between suicidal high school students’ help-seeking and their attitudes and
perceptions of social environment. Journal of Youth and Adolescence, 41(10), 1312–1324.
Power, R., Jones, S., Kearns, G., Ward, J., & Perera, J. (1995). Drug user networks, coping strategies, and
HIV prevention in the community. Journal of Drug Issues, 25(3), 565–581.
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/
Reis, C., & Cornell, D. (2008). An evaluation of suicide gatekeeper training for school counselors and
teachers. Professional School Counseling, 11(6), 386–394.
Robinson, J., Hetrick, S. E., & Martin, C. (2011). Preventing suicide in young people: systematic review.
Australian and New Zealand Journal of Psychiatry, 45(1), 3–26.
Rogers, E. M. (2003). Elements of diffusion. Diffusion of Innovations, 5(1.38).
123 | Pickering
Rogers, E. M. (2010). Diffusion of innovations. Simon and Schuster.
Rulison, K. L., Feinberg, M., Gest, S. D., & Osgood, D. W. (2015). Diffusion of intervention effects: The
impact of a family-based substance use prevention program on friends of participants. Journal of
Adolescent Health, 57(4), 433–440.
Rulison, K. L., Gest, S. D., & Osgood, D. W. (2015). Adolescent peer networks and the potential for the
diffusion of intervention effects. Prevention Science, 16(1), 133–144.
Shaffer, D., Scott, M., Wilcox, H., Maslow, C., Hicks, R., Lucas, C. P., … Greenwald, S. (2004). The
Columbia SuicideScreen: Validity and reliability of a screen for youth suicide and depression. Journal of
the American Academy of Child & Adolescent Psychiatry, 43(1), 71–79.
Shalizi, C. R., & Thomas, A. C. (2011). Homophily and contagion are generically confounded in
observational social network studies. Sociological Methods & Research, 40(2), 211–239.
Starkey, F., Audrey, S., Holliday, J., Moore, L., & Campbell, R. (2009). Identifying influential young
people to undertake effective peer-led health promotion: the example of A Stop Smoking In Schools Trial
(ASSIST). Health Education Research, 24(6), 977–988.
Stephenson, J., Strange, V., Forrest, S., Oakley, A., Copas, A., Allen, E., … others. (2004). Pupil-led sex
education in England (RIPPLE study): cluster-randomised intervention trial. The Lancet, 364(9431), 338–
346.
Stoebenau, K., & Valente, T. W. (2003). Using network analysis to understand community-based
programs: a case study from highland Madagascar. International Family Planning Perspectives, 167–173.
Story, M., Lytle, L. A., Birnbaum, A. S., & Perry, C. L. (2002). Peer-led, school-based nutrition
education for young adolescents: feasibility and process evaluation of the teens study. Journal of School
Health, 72(3), 121–127.
Suzuki, T. (2011). Analysis of core/periphery structures using R. Sociological Theory and Methods, 26,
421–428.
Taylor, L. M., Oldershaw, A., Richards, C., Davidson, K., Schmidt, U., & Simic, M. (2011).
Development and pilot evaluation of a manualized cognitive-behavioural treatment package for
adolescent self-harm. Behavioural and Cognitive Psychotherapy, 39(5), 619–625.
Tobler, N. S., & Stratton, H. H. (1997). Effectiveness of school-based drug prevention programs: A meta-
analysis of the research. Journal of Primary Prevention, 18(1), 71–128.
Tompkins, T. L., Witt, J., & Abraibesh, N. (2010). Does a gatekeeper suicide prevention program work in
a school setting? Evaluating training outcome and moderators of effectiveness. Suicide and Life-
Threatening Behavior, 40(5), 506–515.
Turner, G., & Shepherd, J. (1999). A method in search of a theory: peer education and health promotion.
Health Education Research, 14(2), 235–247.
124 | Pickering
U.S. Department of Health and Human Services. (2012). Office of the Surgeon General and National
Action Alliance for Suicide Prevention. In 2012 National Strategy for Suicide Prevention: Goals and
Objectives for Action. HHS Washington, DC.
Valente, T. W. (1996). Social network thresholds in the diffusion of innovations. Social Networks, 18(1),
69–89.
Valente, T. W. (2010). Social networks and health: Models, methods, and applications (Vol. 1). Oxford
University Press New York.
Valente, T. W. (2012). Network interventions. Science, 337(6090), 49–53.
Valente, T. W. (2017). Putting the network in network interventions. Proceedings of the National
Academy of Sciences, 114(36), 9500–9501.
Valente, T. W., & Davis, R. L. (1999). Accelerating the diffusion of innovations using opinion leaders.
The Annals of the American Academy of Political and Social Science, 566(1), 55–67.
Valente, T. W., Fujimoto, K., Chou, C.-P., & Spruijt-Metz, D. (2009). Adolescent affiliations and
adiposity: a social network analysis of friendships and obesity. Journal of Adolescent Health, 45(2), 202–
204.
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.
Valente, T. W., Hoffman, B. R., Ritt-Olson, A., Lichtman, K., & Johnson, C. A. (2003). Effects of a
social-network method for group assignment strategies on peer-led tobacco prevention programs in
schools. American Journal of Public Health, 93(11), 1837–1843.
Valente, T. W., & Pumpuang, P. (2007). Identifying opinion leaders to promote behavior change. Health
Education & Behavior, 34(6), 881–896.
Valla, J.-P., Bergeron, L., & Smolla, N. (2000). The Dominic-R: a pictorial interview for 6-to 11-year-old
children. Journal of the American Academy of Child & Adolescent Psychiatry, 39(1), 85–93.
Van Eck, P. S., Jager, W., & Leeflang, P. S. (2011). Opinion leaders’ role in innovation diffusion: A
simulation study. Journal of Product Innovation Management, 28(2), 187–203.
Walrath, C., Garraza, L. G., Reid, H., Goldston, D. B., & McKeon, R. (2015). Impact of the Garrett Lee
Smith youth suicide prevention program on suicide mortality. American Journal of Public Health, 105(5),
986–993.
Wasserman, D., Hoven, C. W., Wasserman, C., Wall, M., Eisenberg, R., Hadlaczky, G., … others.
(2015). School-based suicide prevention programmes: the SEYLE cluster-randomised, controlled trial.
The Lancet, 385(9977), 1536–1544.
Watts, D. J., & Dodds, P. S. (2007). Influentials, networks, and public opinion formation. Journal of
Consumer Research, 34(4), 441–458.
125 | Pickering
Whitlock, J., Wyman, P. A., & Moore, S. R. (2014). Connectedness and suicide prevention in
adolescents: Pathways and implications. Suicide and Life-Threatening Behavior, 44(3), 246–272.
Wiist, W. H., Jackson, R. H., & Jackson, K. W. (1996). Peer and community leader education to prevent
youth violence. American Journal of Preventive Medicine, 12(5), 56–64.
Wilton, T., Keeble, S., Doyal, L., & Walsh, A. (1995). The effectiveness of peer education in health
promotion: Theory and practice. University of the West of England, Faculty of Health and Community
Studies Bristol.
Wyman, P. A., Brown, C. H., Inman, J., Cross, W., Schmeelk-Cone, K., Guo, J., & Pena, J. B. (2008).
Randomized trial of a gatekeeper program for suicide prevention: 1-year impact on secondary school
staff. Journal of Consulting and Clinical Psychology, 76(1), 104.
Wyman, P. A., Brown, C. H., LoMurray, M., Schmeelk-Cone, K., Petrova, M., Yu, Q., … Wang, W.
(2010). An outcome evaluation of the Sources of Strength suicide prevention program delivered by
adolescent peer leaders in high schools. American Journal of Public Health, 100(9), 1653–1661.
Wyman, P. A., Pickering, T. A., Schmeelk-Cone, K., Valente, T. W., Pisani, A. R., & Brown, C. H.
(2018). Network Epidemiology of Suicial Students in Schools: Evidence from the Sources of Strength
Study. Manuscript in Preparation.
126 | Pickering
Appendix A. A list of commonly-used terms in the manuscript.
Term Definition
Ego Network For any given ego, the set of individuals who nominated the
ego or were nominated by the ego as a friend.
Homophily The state of two individuals possessing similar qualities.
Network Census The complete measurement of all friendship ties within a
school network.
Opinion Leader An individual who students at school believe is a leader that
other students listen to.
Peer Educator Generally synonymous with "peer leader," though typically
delivers the intervention in a more structured, less ecological
setting.
Peer Leader An individual who was selected to undergo training to
disseminate an intervention; a hypothetical seed to spread
intervention messaging in the network.
Sociometric Relating to the measurement of the friendship network, the
friendship relations of individuals at school.
Suicidality Having either suicide attempt or ideation within the past year.
Suicide Attempt (SA) Having attempted suicide 1 or more times within the past year.
Suicide Ideation (ID) Having seriously considered suicide within the past year with
no suicide attempts.
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Investigating factors that influence peer relationships and obesity during middle childhood
PDF
Exploring the role of peer influence, linguistic acculturation, and social networks in substance use
PDF
Contextualizing social network influences on substance use among high risk adolescents
PDF
A network analysis of online and offline social influence processes in relation to adolescent smoking and alcohol use
PDF
Social selection and influence of alcohol & marijuana implicit cognitions and behaviors: a longitudinal investigation of peer social network dynamics
PDF
System dynamics of in-home family eating behavior: insights from intensive longitudinal data using Ecological Momentary Assessment and wearable sensors
PDF
Social network influences on depressive symptoms among Chinese adolescents
PDF
Stage of readiness and learning styles -- a tailored HIV/AIDS prevention intervention for youth detained in the Los Angeles County Juvenile Justice system
PDF
Psychosocial and cultural factors in the primary prevention of melanoma targeted to multiethnic children
PDF
Adolescent social networks, smoking, and loneliness
PDF
Understanding the dynamic relationships between physical activity and affective states using real-time data capture techniques
PDF
Sleep health and variability in youth: a real-time data capture study to examine influences on daily dietary intake patterns and longitudinal weight trajectories
PDF
Effects of sugar and fiber consumption in minority adolescents and self-tracking as a potential dietary intervention tool
PDF
The influence of contextual factors on the processes of adoption and implementation of evidence-based substance use prevention and tobacco cessation programs in schools
PDF
Mixed methods investigation of user engagement with a smoking cessation app
PDF
A sociocultural and developmental approach to intimate partner violence among a sample of Hispanic emerging adults
PDF
Addressing federal pain research priorities: drug policy, pain mechanisms, and integrative treatment
Asset Metadata
Creator
Pickering, Trevor Anthony
(author)
Core Title
Diffusion of a peer‐led suicide preventive intervention in secondary schools: strategies to increase effectiveness of peer‐led interventions
School
Keck School of Medicine
Degree
Doctor of Philosophy
Degree Program
Preventive Medicine (Health Behavior Research)
Publication Date
03/01/2019
Defense Date
12/03/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
diffusion,diffusion of innovations,health interventions,intervention diffusion,OAI-PMH Harvest,peer educators,peer leaders,peer‐led interventions,social network analysis,social networks,suicide prevention
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Valente, Thomas (
committee chair
), de la Haye, Kayla (
committee member
), Huh, Jimi (
committee member
), Wyman, Peter (
committee member
)
Creator Email
tpickeri@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-128581
Unique identifier
UC11676914
Identifier
etd-PickeringT-7126.pdf (filename),usctheses-c89-128581 (legacy record id)
Legacy Identifier
etd-PickeringT-7126.pdf
Dmrecord
128581
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Pickering, Trevor Anthony
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
diffusion
diffusion of innovations
health interventions
intervention diffusion
peer educators
peer leaders
peer‐led interventions
social network analysis
social networks
suicide prevention