Scott Gest
Human Development & Family Studies
Penn State University
Prevention Research Seminar
November 17, 2010


A network perspective on peer norms
Intervention effects on peer norms in
middle/high school (PROSPER Peers)

Correlates of classroom-level peer norms

Everyday teaching practices and peer norms

The promise/hope of setting level program
impact
◦ Diffusion, mutual reinforcement, continuation

Prevention programs often seek to
reorganize social systems
◦ Dividing schools into smaller sub-units

Other programs attempt setting-level
change via individual-level change
◦ Seek to improve classroom climate by teaching
individual social skills

In Theory

In Research

In Prevention/Intervention Practice

Public health emphasis on social networks
◦ Prominent in major explanations of deviance
◦ Differential association, social learning, problem
behavior theory, peer cluster theory
◦ Friends’ deviance highly reliable predictor
◦ Emphases on refusal skills, friendship choice,
parent monitoring about friends

Descriptive norms
◦ what most people do (“what is”)
◦ Setting-level measure: average across individuals

Injunctive norms
◦ what people are expected to do (“what ought to be”)
◦ Setting-level measure: average across individuals

Norm salience
◦ the extent to which a behavior is associated with
positive or negative social sanctions
◦ Setting-level measure: correlation between level of
behavior and centrality in peer network
Youth
In-Degree Reach
A
5
8
B
5
10
C
3
8
D
3
4
Deviant behavior
associated with
less central
position.
Deviant behavior
associated with
more central
position.
Contrasts with norm as attribute mean over indivs.

Reduced selection of deviant friends
◦ Students encouraged to select prosocial peers who
will help them meet positive goals
◦ Parents encouraged to monitor students’ activities
and peer affiliations

Link to network structure
◦ Altered friendship selection tendencies will place
antisocial youth in less central/influential structural
positions

School based programs targeting substance
use, community partnerships with university
extension:
◦ Random assignment of communities

26 communities, 2 grade cohorts, 5 waves
◦ Iowa & Pennsylvania, Small towns
◦ 10,000+ Students per wave; 14,000+ total
◦ 368 school/cohort/wave networks

Questionnaires assess friendships
◦ Fall of 6th grade & Spring of 6th, 7th, 8th, & 9th
◦ Also assess variety of attitudes and behaviors

Lead Investigators
◦ Wayne Osgood, Mark Feinberg, Scott Gest, Jim Moody (Duke Univ.),
Karen Bierman

Other Investigators
◦ Derek Kreager, Sonja Siennick (Florida State Univ.), Kelly Rullison
(UNC Greensboro), Michael Cleveland, Suellen Hopfer

Graduate Assistants
◦ Wendy Brynildsen, Robin Gautier, Lauren Molloy, Dan Ragan,
Debra Temkin, April Woolnough

PROSPER Study Lead Investigators
◦ Dick Spoth, Mark Greenberg, Cleve Redmond, Mark Feinberg

Funders
◦ National Institute of Drug Abuse, Division of
Epidemiology, Services and Prevention
Research, Prevention Research Branch
◦ William T. Grant Foundation

Family-focused intervention – 6th Grade
◦
◦
◦
◦

Strengthening Families Program
Seven 2-hr sessions
Overarching theme: “Love & Limits”
Prior results suggest diffusion of effects
School-focused intervention – 7th grade
◦ Life Skills Training, Project Alert, or All Stars
◦ 11 to 18 sessions
◦ Relevant focus: Selecting prosocial peers, helping
peers make good choices, and resisting negative
influence from peers
Who are your best and closest friends in your grade?
First name
Last Name
(or if you don’t
know their last
name, . . . )
How often do you “hang out”
with this person outside of
school, (without adults
around)?
1) Never . . . 5) Almost Every Day
YOUR BEST FRIEND or FRIENDS
OTHER CLOSE FRIENDS


Questionnaire response rate: 87.2%
Usable friendship choices, Overall: 81.9%
◦ Of respondents: 93.9%

Name matching, Overall: 81.5%
◦ Inter-rater agreement, 98%
◦ Non-matches primarily out of school

Reciprocation rate
◦ Overall: 48%,
◦ 1st choice: 76%

Regression Coefficient
◦ Within-school association across
individuals
◦ Between a measure of deviant behavior
(IV) and a network measure of influence
potential (DV)
◦ (Mean effect later standardized for
effect size)







Degree: Number of Friendships
Closeness Centrality: Mean distance to reach others
Reach: Number of Direct & Indirect Friendships
Bonacich Centrality: Links to well connected others
Information Central: Harmonic mean dist to others
Betweenness Central: Import. in connecting others
Composite: Standardized sum of above
◦ All for both incoming and total friendships
◦ Transformed to normalize distributions and make
independent of network size

Substance Use:
◦ 30 days, 4 substances, IRT scoring

Substance Use Attitudes:
◦ Composite of 4 scales, 22 items

Delinquency:
◦ Past year, 12 items, IRT scoring

Composite antisocial
◦ Sum of other 3, standardized


No significant pretest diffs btwn trt & ctrl
Waves 2 – 5 as outcome
◦ Pooled test (impact doesn’t vary over time)

Multilevel model:
◦ 5 levels: Comm pairs for random assignment,
Community, School, Cohort, Wave
 School & Cohort crossed
 Level 1 variance heterogeneous by s. e. of b
◦ Controls for wave, state, Network size (Log,
quadratic), and pretest
◦ Estimated by Bayesian MCMC in MLwiN
◦ BDIC criterion for variance comps & controls
PROSPER Program Effects on Behavioral
Trtent Vs. Norms
Control
Measures Defining Relationship
Antisocial
Social Status
Behavior &
(Undirected)
Attitudes
Composite
Composite
Composite
Substance Use
Composite
Subst Use Attitudes
Composite
Delinquency
Total Friends
Composite
Closeness Centrality
Composite
Direct & Indirect Frnds
Composite
Bonacich Central.
Composite
Information Central.
Composite
Betweenness Central.
Composite
p < .05, 2 tail
p < .10, 2 tail
Diff.,
Std b
-0.052
-0.034
-0.048
-0.046
-0.045
-0.056
-0.041
-0.080
-0.063
-0.021
Comparison
Std.
Error
z
0.022 -2.38
0.018 -1.93
0.023 -2.07
0.022 -2.05
0.025 -1.84
0.115 -0.49
0.019 -2.15
0.039 -2.03
0.055 -1.15
0.067 -0.31
N = 253 - 256 networks
p
0.017
0.053
0.039
0.040
0.066
0.625
0.032
0.043
0.252
0.760
Std. b for Compos. Social Status
with Compos. Antisocial
0.05
0.00
-0.05
-0.10
-0.15
-0.20
Pretest 6th Fall
6th Grade 7th Grade 8th Grade 9th Grade
Spring
Spring
Spring
Spring
Control
Treatment

Clear evidence of beneficial intervention
impact on behavioral norms
◦ Reduced social status of antisocial relative to
prosocial youth
◦ Most consistent for composite measures
◦ Non-signif often equally strong (lower power)

Modest effect size
◦ Approximately .05 difference in correlation
Implications

Solid initial support for social network
approach to setting level intervention
◦ Setting level impact distinct from individual effects
Next Steps

Simulations needed to determine implied
reductions in deviance
◦ Networks are complex systems and influence
processes are reciprocal and dynamic

Assess evidence for mediation
◦ More deviance reductions where bigger network
effects?


Are positive peer norms associated with
better youth experiences & outcomes?
How might everyday teaching practices shape
peer norms?


Standard finding
in peer relations
literature:
aggression is
negatively
correlated with
acceptance (being
“liked most”)
But this masks
considerable
variation across
classrooms
norms
against aggression
norms
supporting aggression

TRAILS study (Siegwart Lindenberg, PI)
◦ Dijkstra, Gest, Lindenberg, Sentsa & Veenstra
(in prep)

N = 3,231 youth in 164 classrooms
Mage = 13.60

Peer nominations

◦ acceptance/rejection, behavior

Student reports
◦ fun, boredom, excitement at school

Teacher reports
◦ student achievement

Overall, there were weak norms against academic
achievement, but considerable variability.
◦ Mr (164) = -.11, SDr = .28
◦ 10th %ile r = -.42


90th %ile r = .27
Similar variability in norms for prosocial behavior
(weakly supporting) and bullying (moderately
against)
Norm for achievement, prosocial and bullying were
intercorrelated, so overall classroom norms in each
classroom were categorized as positive or negative

In classrooms categorized as having a positive profile of peer
norms:
◦ Students reported




more fun & excitement
less boredom
more emotional and practical support from peers
less peer rejection
◦ Teachers reported
 better academic adjustment
◦ Median d = .15


Peer norms may serve as useful marker of classroom social
atmosphere & student experience
What teaching practices may be related to these norms?


Goal: Identify teaching practices associated with
emerging peer network features (including peer
norms) and links to student outcomes
1st, 3rd & 5th grade classrooms
◦ Assessed three times within same school year


Each assessment includes measures of…
◦ Teaching practices (Observations, Teacher report)
◦ Peer-reported networks
◦ Youth behavior (peer & teacher report)
◦ Youth perceptions of classroom & school
Sample
◦ N = 39 classrooms from pilot study year (cross-sectional)
◦ Final sample will include >150 classrooms


Investigators: Scott Gest, Phil Rodkin (U of
Illinois), Tom Farmer
PA Lab
◦ Grad students: Deborah Temkin, Rebecca Madill,
Kathleen Zadzora, Rachel Abenavoli
◦ Project Coordinator: Gwen Kreamer
◦ Data manager: Larissa Witmer
◦ Undergraduate assistants: Kristen Granger, Kara
McKee

Funding
◦ William T. Grant Foundation & Spencer Foundation
◦ Institute of Education Sciences

1930s – Lewin
◦ teachers can and must influence “social atmosphere” of the classroom

1940s-50s – Gronlund (1959)
◦ Teachers should try to prevent “cliques & cleavages” in “social fabric”.
Based on clinical wisdom, mostly from descriptive case studies.

1970s – Hallinan
◦ Network analysis to show that reading groups foster friendships

1980s-90s – Cairns & Cairns
◦ “invisible hand” of the teacher in shaping peer networks

2000s – Farmer
◦ Elaboration on how teachers can deliberately influence social related
to status, affiliations and aggression

Now
◦ potential for network concepts & methods to strengthen and test
theories about teacher influence on peer networks
• Generally weak associations (little power with N=39)
• Emotional support associated with less peer support for aggression
• Instructional support associated with less peer support for prosocial
behavior and achievement

Causality unclear:
 Peer norms  teaching?
 Teaching  peer norms?
 Longitudinal design (Sep/Nov/May) will help some


There is striking diversity in teachers’ use of grouping
strategies, even within the same school – what do teachers
think their strategies are accomplishing? Sparse empirical
literature to guide practice.
Might peer norms partially mediate any associations between
teaching practices and student experiences of the classroom
(e.g., perceptions of support, achievement motivation)?

What do we really know about “desirable” and
“undesirable” features of peer networks?
◦ How are peer norms related to student experiences
and academic learning over time?
◦ Are more tight-knit networks always better?
◦ Are cliques always bad?
◦ Are hierarchies always bad?
◦ Might the desirability of these features depend on
which behaviors are valued in the network?

What are teachers (and schools) already
doing that may have systematic effects on
peer networks?
◦ Decades of clinical wisdom but little data
◦ Tremendous variation in very basic practices
◦ Variation in teacher awareness of peer dynamics,
goals for network, and planful action
◦ Building an evidence base could clarify processes
governing emerging network features and provide a
stronger foundation for professional training &
development

How can we apply this knowledge to advance
“prevention science” in schools?
◦ There is value in articulating hypotheses about how
school-based interventions may impact peer networks at
the setting level
 Many “universal” interventions target all students in the
school setting
 Implications for peer network are sometimes explicit, but
more often implicit
◦ Network concepts can sharpen program theories and
indices can permit strong tests
◦ Consideration of network dynamics quickly leads to
recognize of potential tradeoffs in intervention effects