Content uploaded by Philip Mohr
Author content
All content in this area was uploaded by Philip Mohr on Apr 21, 2022
Content may be subject to copyright.
Social Networks 32 (2010) 161–167
Contents lists available at ScienceDirect
Social Networks
journal homepage: www.elsevier.com/locate/socnet
Obesity-related behaviors in adolescent friendship networks
Kayla de la Hayea,b,∗, Garry Robinsc, Philip Mohrd, Carlene Wilsone,f
aCSIRO Preventative Health Flagship, Australia
bUniversity of Adelaide, Australia
cUniversity of Melbourne, Australia
dCSIRO Food and Nutritional Sciences, Australia
eFlinders University, Australia
fCancer Council South Australia, Australia
article info
Keywords:
Obesity
Health behavior
Adolescents
Social networks
Exponential random graph models
abstract
This study examines obesity-related behaviors within adolescent friendship networks, because adoles-
cent peers have been identified as being important determinants of many health behaviors. We applied
ERGM selection models for single network observations to determine if close adolescent friends engage
in similar behaviors and to explore associations between behavior and popularity. Same-sex friends were
found to be similar on measures of organized physical activity in two out of three school-based friend-
ship networks. Female friends were found to engage in similar screen-based behaviors, and male friends
tended to be similar in their consumption of high-calorie foods. Popularity (receiving ties) was also asso-
ciated with some behaviors, although these effects were gender specific and differed across networks.
Crown Copyright © 2009 Published by Elsevier B.V. All rights reserved.
The proportion of children who are overweight or obese is esti-
mated to be between 20% and 25% in Australia (Olds et al., 2004).
The prevalence of childhood obesity in many other affluent coun-
tries is equally high and, as in Australia, has risen dramatically over
the past couple of decades (WHO, 2003). The economic and soci-
etal costs of this ‘epidemic’ are predicted to be immense because
obese children have an increased risk for a number of medical con-
ditions as well as negative long-term psychosocial consequences
(Zametkin et al., 2004). Overweight adolescents are also at risk of
being overweight in early adulthood (Crossman et al., 2006). As
their behaviors are more malleable than adults, this age may be an
effective time for intervention (Jeffery et al., 2000), so it is important
to understand factors associated with adolescent overweight.
Behaviors associated with rising childhood obesity include food
consumption patterns that have increased energy intake, and
declining levels of physical activity diminishing overall energy
output (Zametkin et al., 2004). Specifically, fast food has been asso-
ciated with increased energy and fat intake (French et al., 2001),
and dietary patterns characterized by over-consumption of energy-
dense, low-fiber, and high-fat foods have been associated with
increased fatness in children (Johnson et al., 2008). Over the past
two decades, an overall decline in physical activity amongst chil-
dren and adolescents has also been reported, which has largely
∗Corresponding author at: CSIRO Food and Nutritional Sciences, PO Box 10041,
Adelaide, SA 5000, Australia. Tel.: +61 8 8303 8911; fax: +61 8 8303 8899.
E-mail address: kayla.delahaye@csiro.au (K. de la Haye).
been attributed to decreased active play and locomotion (Olds et al.,
2004). Screen time, which includes time spent watching television,
computing, and playing video games, has been found to be a strong
competitor for children’s leisure time (Olds et al., 2004). Research
suggests there is a strong relationship between screen time, phys-
ical activity, and propensity for obesity; children who watch more
television are less likely to do vigorous physical activity and are
more likely to have higher body mass indexes (BMIs) (Andersen et
al., 1998).
Behavioral interventions need to be informed by an under-
standing of the important factors shaping obesity-related behaviors
amongst children and adolescents, and there is a growing body of
research highlighting the important role of the social environment.
Family, peer, and school environments have been identified as con-
texts in which adolescents’ health behaviors are established and
maintained (Williams et al., 2002). As adolescents spend increas-
ing time with friends, the potential for the norms and behaviors
of peers to be influential is increased (Peterson, 1989). Peers have
been found to influence adolescents’ consumption of snack foods
(Feunekes et al., 1998) and foods high in saturated fat (Monge-Rojas
et al., 2002). Acculturation to peer norms has also been associated
with lower levels of physical activity and higher frequency of fast
food consumption amongst Hispanic and Asian-American adoles-
cents (Unger et al., 2004). Social support from friends has been
found to be positively related to physical activity (Duncan et al.,
2005), and adolescent girls have been found to be more physically
active when they reported that their close friends engaged in high
levels of physical activity (Voorhees et al., 2005). Yet, contrasting
0378-8733/$ – see front matter. Crown Copyright © 2009 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.socnet.2009.09.001
162 K. de la Haye et al. / Social Networks 32 (2010) 161–167
results have also been reported: Pearson et al. (2006) indicated that
adolescents tended to form friendships with school students who
differed from them in sporting behaviors. Accordingly, it cannot
be taken for granted that adolescent friendships are based around
universal behavioral similarity: although similarity on some behav-
iors (or attributes) may be a central feature of the friendship, dyads
may differ on other, less salient, attributes. Adolescent obesity is
also a predictor of marginalization and social stigma amongst peers
(Strauss and Pollack, 2003), so that health behaviors associated
with overweight may also differ based on social hierarchies or sta-
tus features of friendships.
In sum, this literature suggests that adolescent friendships are
an important social context in which obesity-related behaviors take
place. Understanding the role of friendship influences on these
behaviors, however, needs to take into account the complex struc-
tures of adolescent friendship ties, beyond the simple examination
of dyads and the identification of individual membership within
small groups. In this article, by using statistical models for net-
works, we control for basic friendship dependencies at the same
time as we examine associations between friendship ties and self-
reported behavior. Examination of self-reported behavior, and not
perceptions of the behaviors of friends (i.e. perceived reports),
is particularly important in adolescent research because this age
cohort has been found to project their own behavior onto their
peers (Ryan, 2001).1
1. Social networks and health behaviors
Measuring the complex patterns of adolescents’ friendship ties
can serve to highlight how network structure and behaviors are
interdependent. Theories of social influence identify both direct
(e.g. imitation) and indirect (e.g. internalization of group norms)
mechanisms of interpersonal influence, which arise through social
interaction. Social ties can also be influential as conduits of
resources, information and social support, and positions or roles
within these social-structural contexts can be an additional source
of behavioral control (Friedkin, 1998). Social networks have been
found to play an important role in adult obesity, with individ-
ual weight status strongly influenced by the weight status of
close friends and other non-biological ties (Christakis and Fowler,
2007). Recent research has also found that adolescents’ weight
is associated with their friends’ weight cross-sectionally (Valente
et al., 2009) and longitudinally: one study has claimed these
effects were explained by shared environmental factors (Cohen-
Cole and Fletcher, 2008b), while others have reported that network
effects on adolescent overweight were indicative of social influence
(Fowler and Christakis, 2008; Trogdon et al., 2008). Unfortunately,
these studies were not able to specify the mechanisms of interper-
sonal influence on weight status, or whether similarity in weight
status was explained by similarity on obesity-related behaviors. As
well, only the Valente et al. (2009) study adequately considered the
effect of possible shared friendships (i.e. transitivity)2; a common
feature of friendship networks that should be controlled to avoid
overestimating network effects.
Associations between network structure and individual
attributes, such as health behaviors, can also be driven by pro-
1Although perceived behavior of network partners has been found to be an impor-
tant source of influence on some health behaviors (Rice et al., 2003; Valente et al.,
1997), this paper is interested in exploring associations between friendship ties
and ‘actual’ behavior. Drawing conclusions about the potential social–psychological
mechanisms that may underpin associations between friendship ties and obesity-
related behaviors is beyond the scope of the current study, but would be a fruitful
topic for future research.
2Accounting for shared ties amongst ‘alters’ was not possible in the Christakis
and Fowler (2007) study given limitations when the data were originally collected.
cesses of social selection. Studies of adolescent smoking have
found individuals tend to form friendships with peers who engage
in similar smoking behaviors (Kobus, 2003). For single network
observations it is not possible to disentangle processes of selection
and influence (Steglich et al., submitted for publication), but we
are able to test for associations between network structure and
behavior which, if present, would justify subsequent longitudinal
research.
Recent developments in social network analysis, employing
Exponential Random Graph (or p*) Models (ERGMs), provide a
sophisticated method for modeling the structure of complex social
networks (Robins et al., 2007). New ERGM specifications allow us
to make dependence assumptions about the presence of network
ties that go beyond the level of the dyad, to account for formations
of larger group structures (Snijders et al., 2006). These statistical
models assume that networks are self-organizing: “relational ties
come into being in ways that may be shaped by the presence or
absence of other ties (and possibly node-level attributes)” (Robins
et al., 2007, p. 177). The formation of ties, and thus the overall
network structure, is therefore assumed to be based on structural
or ‘endogenous’ processes such as tie reciprocation or transitivity
(i.e. shared friendships), as well as ‘exogenous’ processes involving
node-level attributes, including social influence and social selection
(Robins et al., 2001). In this paper, we apply ERGM selection mod-
els for single network observations to understand whether there
are associations between node-level attributes and tie-level vari-
ables without drawing firm inferences about whether selection or
influence (or both) processes are operating.
ERGM parameters represent a range of different tie configu-
rations, each of which relates to specific structural processes or
interactions between network ties and individual-level attributes
(Robins et al., 2007). These parameters can be estimated simultane-
ously to determine which effects significantly explain the network
structure: i.e. which particular configurations of ties occur more or
less than would be expected at chance levels, given the number of
nodes and density of the network, and given other effects in the
model. This enables us to gain an understanding of the structural
building blocks of adolescent friendship networks, and to explore
interdependence between individual attributes and friendship ties
(such as friend similarity) within these complex social structures.
The present study used ERGMs to investigate associations
between adolescents’ friendship networks and obesity-related
behaviors. The major aim of the study was to determine if
close adolescent friends are similar on a number of obesity-
related behaviors. Our review of the research literature on obesity
suggested that three general areas of behavior might be theoret-
ically and empirically important: high-calorie food consumption
(Johnson et al., 2008), physical activity (Zametkin et al., 2004), and
sedentary screen-based behaviors (Andersen et al., 1998). The sec-
ondary aim was to explore if popularity (receiving friendship ties)
was associated with these behaviors, which may serve to highlight
additional social processes relevant to the behaviors in question.
2. Method
2.1. Respondents
Male and female students from two independent middle schools
in a major Australian city were invited to take part in the study.
Participants from School 1 were in year 8, and predominantly 13
years old (76%). Participants from School 2 were in year 8 (82%
were 13 years old) and year 9 (87% were 14 years old). Each school
year level was defined as a separate peer network to explore sim-
ilarities and differences within and across age groups and school
contexts. Response rates within each of these three peer networks
K. de la Haye et al. / Social Networks 32 (2010) 161–167 163
were excellent, ranging from 81% to 93%. Students and their par-
ents/guardians were informed of the study via an information letter
mailed to students’ homes, which included information on the
study and provided an opportunity to opt-out of participation. Stu-
dents were also given the opportunity to opt-out of the study on the
day of data collection. To maximize the response rate and obtain
optimally complete network data, the information letter also stated
that respondents would be entered into a draw for one of two gift
vouchers.
2.2. Procedure and materials
Respondents completed a paper-based questionnaire developed
for this study in their classroom. The questionnaire assessed self-
reported frequency of engaging in an obesity-related behavior;
high-calorie food consumption, physical activity, and screen time,
“during a normal week of the school year”. The questionnaire also
included items measuring students’ age, gender, year level, and
friendship ties within the school. Participants’ height and weight
were measured in order to calculate BMI, which is included in the
subsequent analyses as a control variable.3
2.2.1. High-calorie food consumption
Four items assessed weekly consumption of fast food, savory
snack foods, sweet snack foods and high-calorie drinks. To aid
understanding and recall, each item provided respondents with
examples of these foods. Respondents rated their consumption fre-
quency on a 5-point scale that was slightly modified to suit each
food type, where 1 represented the lowest frequency and 5 the
highest frequency (e.g. 1 =less than once a week, 5 = three or more
times a day). Overall scores for high-calorie food consumption were
calculated by taking a mean of the four food items, with Cronbach’s
alphas ranging from 0.60 to 0.69 across the three networks.
2.2.2. Physical activity
Four items assessed the weekly quantity and frequency of
respondents’ participation in both organized and non-organized
physical activities, which were defined as in the Adolescent Physical
Activity Recall Questionnaire (Booth et al., 2002). Organized activ-
ities were described as “activities played in teams or supervised by
coaches, including training and practice”, whereas non-organized
activities included “activities that are not supervised by adults and
don’t involve training”, with 30 examples of both activity types
provided to aid comprehension and recall. Respondents rated the
frequency they engaged in both types of activities on a 5-point scale,
where 1 = less than once a week and 5 = every day. The weekly quan-
tity of activity, measured in number of hours, was also rated on a
5-point scale, where 1 = none and 5 =seven or more hours a week.
Overall scores for organized physical activity and non-organized
physical activity were calculated by taking a mean of the frequency
and quantity measures for each activity type. Cronbach’s alpha coef-
ficients ranged from 0.87 to 0.93 for organized physical activity, and
from 0.77 to 0.79 for non-organized physical activity.
2.2.3. Screen time
Six items measured the number of hours spent watching
TV/movies, playing video/computer games, and using the Internet
on “a normal school day”, and on “a normal day of the week-
end”. Responses were given on a 5-point scale, where 1 = none
and 5 = more than four hours a day. An overall score for TV/movie
3Because weight status has been associated with friendship selection and social
marginalization amongst adolescent peers, controlling for BMI allows us to explore
associations between friendship ties and obesity-related behaviors independent of
these effects.
watching was calculated by taking a mean of the weekday and
weekend scores (Cronbach’s alpha coefficients ranged from 0.53
to 0.75).4An overall score for ‘other’ screen activities was cal-
culated by taking a mean of the weekday and weekend scores
of video/computer gaming and Internet use (Cronbach’s alphas
ranged from 0.66 to 0.81).
2.2.4. Measurement of directed friendship ties
Respondents listed the first and last names of all of their “close
friends” (defined as friends they “hang around with” the most), not
including siblings that attended the same school and were in their
year level. Respondents were also provided with a response option
indicating that their close friends were not at this school. The num-
ber of friends to nominate was not specified, although 15 lines were
provided.
The three school friendship networks were found to be strongly
segregated by gender: the proportion of friendship ties that were
inter-gender amongst the 2 year 8 groups was 18% (School 1)
and 17% (School 2), and was 10% amongst the year 9 group. As
the social processes underpinning friendship selection, and poten-
tially social influence, likely differ for inter-gender vs. intra-gender
friendship ties in early adolescence, we decided to consider only
intra-gender friendship ties in this paper and to explore male
and female friendship networks separately.5The demographic and
network characteristics for each of these six friendship networks
(three male, three female) are summarized in Table 1, and the
descriptive statistics of the behavioral measures are summarized
in Table 2.
2.3. Analysis methods
Statistical analysis was conducted with PNet (Wang et al., 2006),
a program for the simulation and estimation of ERGMs. PNet imple-
ments Markov Chain Monte Carlo Maximum Likelihood Estimation
to estimate model parameters and standard errors, based on a
fixed number of nodes. Parameter estimates of zero indicate that
the effect being modeled occurs at a rate consistent with chance,
whereas a positive parameter suggests the effect is more preva-
lent and a negative parameter that the effect is less prevalent than
chance, given the other effects in the model. Effects are tested using
at-ratio (the parameter estimate divided by the standard error) and
are assessed as significant when the absolute value of this ratio
exceeds two (Snijders et al., 2006).6Once convergent estimates
were achieved goodness of fit was assessed via simulation of the
estimates in PNet.
Each model included parameters for both structural (endoge-
nous) and node-level effects; these parameters and their
corresponding graph configurations are outlined in Table 3. Fol-
lowing the suggestion of Snijders et al. (2006), graph density was
fixed in most models to facilitate convergence of the estimation
algorithm.
To test the hypothesis that friends would be similar on measures
of obesity-related behaviors, the models included parameters for an
4Because the ‘TV/movie watching’ summary measure was only based on two
items, and the Cronbach’s alpha coefficients for Network 1 and Network 2 were
acceptable (both 0.75), the overall score was used despite the low alpha coefficient
in Network 3 (0.53).
5Initial analyses of friendship networks that included both males and females
found some similar effects to the analyses reported below, but also obscured many
effects that were gender based. Accordingly, a separate analysis of male and female
friendship networks seems reasonable.
6We label an effect as significant following the usual practice in the ERGM lit-
erature of a parameter estimate that is more than twice its standard error. Caution
needs to be adopted about exact probabilities, however, as underlying distributions
are not known.
164 K. de la Haye et al. / Social Networks 32 (2010) 161–167
Table 1
Demographic and network characteristics of the adolescent friendship networks.
Characteristic Male friendship networks Female friendship networks
N1 N2 N3 N1 N2 N3
N90 57 55 74 51 58
School 122122
Year level 889889
Out-degree mean (S.D.) 5.2 (3.0) 5.7 (3.2) 6.2 (3.4) 5.1 (2.8) 5.0 (2.2) 5.4 (2.7)
Min,Max 0,14 0,15 0,14 0,12 0,10 1,11
In-degree mean (S.D.) 5.2 (2.9) 5.7 (3.4) 6.2 (3.3) 5.1 (2.5) 5.0 (2.4) 5.4 (2.2)
Min,Max 0,11 0,15 0,18 0,13 0,11 2,10
Table 2
Means (standard deviations) for the measures of obesity-related behaviors.
Behavior Male friendship networks Female friendship networks
N1 N2 N3 N1 N2 N3
Fast fooda1.9 (0.8) 2.2 (0.9) 2.0 (0.8) 1.8 (0.8) 1.7 (0.7) 2.1 (0.8)
Savory snacksb2.3 (0.9) 2.4 (0.9) 2.5 (1.1) 1.9 (0.8) 2.3 (1.0) 2.2 (1.0)
Sweet snacksb2.4 (0.9) 2.4 (1.0) 2.7 (0.9) 2.3 (0.9) 2.5 (0.9) 2.4 (1.0)
High-calorie drinkb2.5 (1.2) 2.8 (1.3) 2.8 (1.1) 2.1 (1.1) 2.3 (1.1) 2.2 (1.1)
Organized PA
Frequencyc3.2 (1.0) 3.4 (1.0) 2.9 (1.0) 2.8 (1.2) 3.1 (1.1) 2.8 (1.2)
Quantityd3.6 (1.1) 3.5 (1.1) 3.2 (1.4) 3.2 (1.3) 3.4 (1.2) 3.2 (1.4)
Non-org. PA
Frequencyc3.5 (1.1) 3.3 (1.2) 3.5 (1.2) 3.6 (1.2) 3.2 (1.1) 3.1 (1.0)
Quantityd3.5 (1.1) 3.3 (1.0) 3.3 (0.8) 3.4 (1.0) 3.2 (0.9) 2.9 (0.9)
TV/movies (avg. weekday/end)e3.3 (0.8) 3.5 (0.8) 3.4 (0.6) 3.0 (0.8) 3.3 (0.9) 3.5 (0.7)
Gaming (avg. weekday/end)e2.7 (1.0) 2.7 (1.0) 2.9 (1.1) 1.7 (0.8) 2.0 (0.9) 1.9 (1.2)
Internet (avg. weekday/end)e2.3 (0.9) 2.6 (1.2) 2.6 (1.2) 2.3 (0.8) 2.5 (0.9) 3.0 (0.9)
a1 =almost never, 2 = less than once a week, 3= one to two times a week, 4=three to six times a week, 5 = every day.
b1 =less than once a week, 2 = once or twice a week, 3= three to six times a week, 4=one to two times a day, 5 = three or more times a day.
c1 =less than once a week, 2 = once a week, 3= two to three times a week, 4=four to six times a week, 5 = every day.
d1 =none, 2 = one hour or less a week, 3= two to three hours a week, 4=four to six hours a week, 5 = seven or more hours a week.
e1 =none, 2 = less than one hour a day, 3= one to two hours a day, 4=three to four hours a day, 5 = more than four hours a day.
absolute difference effect (parameter 8) for each of the continuous
behavioral measures. This effect models the absolute difference of
a variable between nodes who share a directed tie: a significant
negative parameter estimate indicates a propensity for nodes that
are tied to be similar (i.e. have less of a difference than expected
by chance) on the attribute in question, given other effects in the
model. To explore associations between popularity and individ-
ual behavior, parameters for receiver effects (parameter 7) were
also included for each continuous behavioral attribute. Receiver
effects model associations between values on an individual vari-
able and in-degree, with positive estimates indicating high values
on this attribute are associated with receiving more ties. Sender
effects (parameter 6) for each behavioral variable were included as
a control and model associations between values on an individual
Table 3
Parameters included in the ERGMs for the directed friendship networks.
Parameter Tie configuration Description
1 Reciprocity Models the tendency for ties to be reciprocated.
2 Popularity (k-instar) Models the in-degree distribution and tendency for popularity.
3 Expansiveness (k-outstar) Models the out-degree distribution and reflects social activity or
expansiveness.
4 Transitive closure (directed k-triangles: AKT-T) Models the tendency for 2-paths to close, meaning for ‘a friend of a friend to
become a friend’ (i.e. shared friendship).
5 Multiple connectivity (directed k-2 paths: A2P-T) Models nodes that are connected by many 2-paths; a pre-cursor to transitivity.
6 Sender Models the tendency for ties to be sent from nodes with a particular attribute
(black node) to any node (white node).
7 Receiver Models the tendency for ties to be sent from any node (white node) to nodes
with a particular attribute (black node).
8 Absolute difference Models the tendency for ties to be sent to nodes with similar or different
scores on a continuous attribute.
Note: See Snijders et al. (2006) and Robins et al. (2009) for additional information on model parameters.
K. de la Haye et al. / Social Networks 32 (2010) 161–167 165
variable and out-degree. All behavioral measures were standard-
ized to normalize data.
Node-level effects for each of the five behavioral measures were
first modeled independently. As there were no significant abso-
lute difference effects for TV/movie watching or non-organized
physical activity these measures were not included further in the
analyses. Final models were developed using a backward selection
process whereby node-level effects for all other behaviors were
modeled simultaneously to test for competing effects, and non-
significant variables removed one step at a time. Parameters for
node-level effects associated with BMI (absolute difference, sender,
and receiver effects) were also included in the final models, to con-
trol for the possibility that associations between friendship ties and
behavior may be explained by associations between friendship ties
and body mass.7In all models, node-level effects were estimated
in conjunction with parameters for structural effects (parameters
1 through 5) that were also expected to explain network struc-
ture and thus need to be controlled while examining the behavioral
effects.
3. Results
3.1. Structural effects
The final model parameter estimates and standard errors for the
structural effects in each of the six friendship networks (three male,
three female) are presented in Table 4, with significant parameters
denoted by an asterisk (*).
Endogenous effects were found to explain the structure of the
observed networks, and thus need to be accounted for when test-
ing the hypotheses. Across all friendship networks, there were
significant reciprocity effects, meaning friendship ties tended to
be reciprocated between dyads. All six networks also had signifi-
cant positive transitive closure parameters, coupled with negative
multiple connectivity parameters. This combination signifies a ten-
dency for ‘a friend of a friend to become a friend’, common in
adolescent friendship networks (e.g. Espelage et al., 2007), and
also indicates that the networks were characterized by segmented,
clustered friendship groups. Most networks also had significant
negative popularity and expansiveness effects, meaning nodes
with high in-degrees and out-degrees were not likely, unless they
formed as a result of other transitive clique-like structures. This
can be interpreted to mean that nodes who are popular or socially
expansive are so within the context of cliques or clusters of friends.
3.2. Difference effects
The hypothesis that adolescent friends would be similar on
obesity-related behaviors was tested by model parameters for
absolute difference effects, with significant negative estimates indi-
cating the hypothesis was supported. The parameter estimates and
standard errors for the preliminary models (with effects for each
behavior modeled independently) and final models are presented
in Table 5, with significant parameters denoted by an asterisk (*),
and significant or marginally significant effects in bold.
The final models illustrate some clear trends in friend similarity
on obesity-related behaviors. In Network 1 and Network 3, both
male and female friends were found to engage in similar amounts
of organized physical activity, as indicated by the significant nega-
tive difference effects in the final models. However in Network 2,
there was no evidence that male or female friends were alike on
7Models that did not control for BMI-related effects were also run. The significant
absolute difference effects did not differ from those reported below suggesting that
behavioral similarity amongst friends is not explained by similarity on BMI.
organized physical activity. The final models also suggest there are
some important gender differences in tendencies towards behav-
ioral homophily. Female friends in all three networks were found
to be alike on ‘other’ screen activities, which include time spent
playing video or computer games, and Internet use. There was also
a trend (although marginally significant) for male friends in Net-
work 2 to be similar on other screen activities, but this effect was
not significant amongst male friends in Network 1 and Network 3.
Finally, there was some evidence that male friends were alike in
their consumption of high-calorie foods. In the preliminary models
there were significant negative difference effects for high-calorie
food consumption amongst males in Network 1 and Network 2,
however once friend similarity on other behaviors was accounted
for in the final models, this effect only remained marginally signif-
icant in Network 2. This suggests that although male friendships
in these two networks might not have been based on similarity
in high-calorie food consumption, as similarity amongst friends
seems better explained by other correlated behaviors, it remains
that male friends were nonetheless alike in the amount of high-
calorie foods they reported consuming. Finally, across all three
networks, there was no evidence that friends (male or female) were
similar in the amount of non-organized physical activities they did,
or on the amount of TV/movies they watched.
3.3. Receiver effects
Receiver effects modeled associations between individual
obesity-related behaviors and popularity (as measured by in-
degree). For males, participation in organized physical activity was
positively associated with receiving ties in Network 1 (significant
estimate of 0.170 (S.E. =0.064)) and Network 3 (marginally signifi-
cant estimate of 0.154 (S.E. = 0.083)), indicating that boys who did
the most organized physical activity tended to be the most popular.
Worth noting is that in these two male friendship networks, both
receiver effects and difference effects were significant for organized
physical activity. Amongst males in Network 3, there was also a sig-
nificant positive receiver effect for high-calorie food consumption
(parameter estimate =0.175, S.E. = 0.075), where popular boys also
tended to be the highest consumers of unhealthy snack and junk
foods.
There were no receiver effects found in the female friendship
networks, apart from a marginally significant positive association
between in-degree and screen activities in Network 1 (parameter
estimate =0.183, S.E.= 0.099). This can be interpreted as a trend for
popular girls to be the highest users of video/computer games and
Internet. Again, this receiver effect was paired with a significant
difference effect showing a tendency for similarity amongst female
friends on screen activities in this network.
4. Discussion
Adolescent school friends were found to be similar on some
obesity-related behaviors, particularly leisure activities. Organized
physical activity was an important factor in adolescent friendships
in two of the three networks, with male and female friends tend-
ing to be alike in the extent that they participated in activities such
as sports and training. Female friends in all three networks were
found to be similar on sedentary screen-based activities, includ-
ing video/computer gaming and Internet use. Perhaps surprisingly,
male friends in two of the three networks were not alike on screen
activities, and whether this reflects differences in the social nature
of the screen activities is not known but is an area for future
research. With regards to high-calorie food consumption, only male
friends in two of the networks were alike in the amount of snack
foods and fast food they consumed. However, this tendency for sim-
166 K. de la Haye et al. / Social Networks 32 (2010) 161–167
Table 4
Model parameter estimates (standard errors) for structural network effects.
Parameter Male friendship networks Female friendship networks
N1 N2 N3 N1 N2 N3
Reciprocity 1.77 (0.21)* 2.05 (0.23)* 2.56 (0.27)* 2.60 (0.28)* 2.06 (0.28)* 2.85 (0.32)*
Popularity −0.76 (0.18)* −0.06 (0.22) −1.11 (0.34)* −1.11 (0.26)* −0.47 (0.27) −1.40 (0.52)*
Expansiveness −0.68 (0.18)* −0.99 (0.29)* −0.59 (0.28)* −0.59 (0.22)* −0.56 (0.29) −0.45 (0.34)
Transitive closure 1.63 (0.11)* 1.01 (0.10)* 1.33 (0.12)* 1.75 (0.15)* 0.94 (0.10)* 0.98 (0.12)*
Multiple connectivity −0.14 (0.02)* −0.16 (0.02)* −0.10 (0.02)* −0.10 (0.03)* −0.22 (0.03)* −0.25 (0.04)*
Table 5
Model parameter estimates (standard errors) for absolute difference effects.
Parameter Male friendship networks Female friendship networks
N1 N2 N3 N1 N2 N3
Preliminary models
High-calorie food −0.093 (0.040)* −0.119 (0.045)* 0.119 (0.061) −0.038 (0.050) −0.053 (0.065) 0.006 (0.064)
Organized PA −0.089 (0.039)* −0.033 (0.045) −0.105 (0.039)* −0.053 (0.039) 0.008 (0.056) −0.098 (0.035)*
Non-org. PA −0.015 (0.035) −0.030 (0.042) −0.055 (0.043) −0.055 (0.038) 0.019 (0.061) −0.064 (0.050)
TV/movies 0.015 (0.039) −0.041 (0.045) −0.033 (0.040) −0.046 (0.038) −0.048 (0.050) 0.047 (0.054)
Other screen −0.007 (0.043) −0.092 (0.046)* −0.048 (0.039) −0.093 (0.053) −0.128 (0.059)* −0.088 (0.038)*
Final models
High-calorie food −0.082 (0.044) 0.061 (0.045)
Organized PA −0.098 (0.032)* −0.079 (0.037)* −0.064 (0.030)* −0.073 (0.035)*
Other screen −0.069 (0.035) −0.099 (0.038)* −0.114 (0.044)* −0.104 (0.039)*
ilarity on food consumption seemed to be a consequence of friend
similarity on other correlated health behaviors such as physical
activity and screen activities.
Although effects of friend similarity on obesity-related behav-
iors showed some differences across school contexts and age
groups, the interpretation of any clear trends driven by these vari-
ables was limited by the small number of networks sampled. As
contextual factors such as school policies or facilities that impact
on food consumption and physical activity are modifiable, it may be
important to consider in future how they mediate social influence
in a school setting.
Longitudinal research is needed to explore the dynamic pro-
cesses that underlie behavioral similarity amongst peers. These
effects may result from processes of social influence, whereby ado-
lescents adopt behaviors that are similar to those of their friends.
It is also likely that adolescents form friendship ties as a result
of common extra-curricular activities, whether they are active or
sedentary, that may result in similarity through friendship selec-
tion. In any event, our cross-sectional findings point to the potential
value of network-based health initiatives that target organized
physical activity, screen activities, and high-calorie food consump-
tion at school, because the school-peer environment seems to
function to maintain groups of friends whose behaviors are obesity-
protecting or obesity-promoting. The clustering of similar friends
into small social groups suggests network analysis could be used
to design interventions that target particular ‘unhealthy’ friendship
groups. Alternatively, interventions harnessing influential peers to
promote healthy behaviors and norms have been found to be suc-
cessful in adolescent school environments (Campbell et al., 2008)
and could be trialed to address these particular obesity-related
behaviors.
Although our results indicate some behaviors are associated
with friendship, it is also of interest to see which behaviors were
not the bases of friendship formation in these networks. Adolescent
friends were not found to be alike on measures of non-organized
physical activity, which includes active play and locomotion, or on
time spent watching TV and movies. As friendship selection may
be less likely to be based on preferences for active transportation
or TV viewing habits than sports or screen activities, there may be
little similarity amongst close friends. Additionally, if these activi-
ties tend to occur outside of the peer environment, social contexts
such as the family may be more influential than peers, and thus a
more relevant milieu for interventions (Koehly and Loscalzo, 2009).
Whether activities engaged in by groups of adolescents can be
described as “social” is also matter for debate; some activities may
be largely driven by individual or environmental factors.
Previous studies have consistently found friends to be alike on
a number of health behaviors, however the results of this study
are more varied. Some of these prior results were based on per-
ceived reports of friend’s behavior, and as noted above, perceived
reports in this age group may overestimate behavioral similarity
amongst friends (Ryan, 2001). This study has also accounted for
transitive closure, whereas previous results based on independent
dyadic friendships may be epiphenomenal of the structural depen-
dencies in the network. In fact, “network effects” have been found
for health outcomes such as acne and height using regression mod-
els commonly employed in peer influence research (Cohen-Cole
and Fletcher, 2008a). The authors suggest these implausible net-
work effects are a result of inadequately controlled environmental
confounders, however they may also be explained to some extent
by failing to control for friendship selection and network structure.
Comparisons between results obtained using regression models
and ERGMs, when studying longitudinal network effects, is an area
that warrants further investigation.8
Additionally, we found evidence that popularity within ado-
lescents’ school friendship networks was associated with some
obesity-related behaviors, although these effects were localized
within networks suggesting that the underlying social processes
may be unique to the local culture developed within each year level
8Thanks to the suggestions of one reviewer, we also ran dyadic independent
network models that did not control for network structure. Although some node-
level effects were similar to those reported in this paper, several effects that were
significant in the dyadic independent models dropped off in models that included
parameters for structural effects. Additionally, a few node-level effects that were
not significant in the dyadic independent models became significant once network
structure was accounted for. We believe this lends support to claims that controlling
for endogenous network processes is important when testing hypotheses of simi-
larity or influence amongst network partners. It is risky to rely on regressions when
the data contains the type of independencies implicit in a network representation.
K. de la Haye et al. / Social Networks 32 (2010) 161–167 167
cohort. Importantly, the tendency for behaviors to be associated
with popularity (receiving ties) occurred only alongside effects for
similarity amongst friends on that behavior. For example, amongst
male friends in Network 1 and Network 3, there was a tendency for
friends to be alike on organized physical activity and for physical
activity to be associated with social status, with popular students
tending to do the most sport or organized physical activities. This
highlights the role popular students may have in making particu-
lar health behaviors salient to friendship formation amongst their
peers, or alternatively may be indicative of social pressures on
popular students to adopt behaviors that are socially valued and rel-
evant in their local peer network. These popular students may also
be an important source of social influence, particularly within their
respective friendship clusters (as indicated by the structural effects
showing popularity and expansiveness tended to be localized in
smaller friendship groups).
To conclude, research has suggested that weight status in ado-
lescents and adults is influenced by their friendship ties and that
over time obesity can spread through social networks (Christakis
and Fowler, 2007; Fowler and Christakis, 2008). This paper has
applied ERGMs to explore obesity-related behaviors in male and
female adolescent friendship networks, and has found that friends
are similar on some leisure activities and food consumption behav-
iors as early as 13–14 years of age; a potential mechanism for the
social ‘contagion’ of overweight and obesity.
References
Andersen, R.E., Crespo, C.J., Bartlett, S.J., Cheskin, L.J., Pratt, M., 1998. Relation-
ship of physical activity and television watching with body weight and level
of fatness among children. Journal of the American Medical Association 279,
938–942.
Booth, M.J., Okely, A.D., Chey, T., Bauman, A., 2002. The reliability and validity of
the Adolescent Physical Activity Recall Questionnaire. Medicine and Science in
Sports and Exercise 34, 1986–1995.
Campbell, R., Starkey, F., Holliday, J., Audrey, S., Bloor, M., Parry-Langdon, N., Hughes,
R., Moore, L., 2008. An informal school-based peer-led intervention for smok-
ing prevention in adolescence (ASSIST): a cluster randomised trial. Lancet 371,
1595–1602.
Christakis, N.A., Fowler, J.H., 2007. The spread of obesity in a large social network
over 32 years. The New England Journal of Medicine 357, 370–379.
Cohen-Cole, E., Fletcher, J.M., 2008a. Detecting implausible social network effects in
acne, height, and headaches: longitudinal analysis. British Medical Journal 337,
a2533.
Cohen-Cole, E., Fletcher, J.M., 2008b. Is obesity contagious? Social networks vs. envi-
ronmental factors in the obesity epidemic. Journal of Health Economics 27,
1382–1387.
Crossman, A., Sullivan, D.A., Benin, M., 2006. The family environment and Ameri-
can adolescents’ risk of obesity as young adults. Social Science & Medicine 63,
2255–2267.
Duncan, S.C., Duncan, T.E., Strycker, L.A., 2005. Sources and types of social support
in youth physical activity. Health Psychology 24, 3–10.
Espelage, D.L., Green Jr., H.D., Wasserman, S., 2007. Statistical analysis of friendship
patterns and bullying behaviors among youth. New Directions for Child and
Adolescent Development 118, 61–75.
Feunekes, G.I., de Graaf, C., Meyboom, S., van Staveren, W.A., 1998. Food choices
and fat intake of adolescents and adults: associations of intakes within social
networks. Preventative Medicine 27, 645–656.
Fowler, J.H., Christakis, N.A., 2008. Estimating peer effects on health in social net-
works: a response to Cohen-Cole and Fletcher; and Trogdon, Nonnemaker, and
Pais. Journal of Health Economics 27, 1400–1405.
French, S.A., Story, M., Neumark-Sztainer, D., Fulkerson, J.A., Hannan, P., 2001. Fast
food restaurant use among adolescents: associations with nutrient intake, food
choices and behavioral and psychosocial variables. International Journal of Obe-
sity 25, 1823–1833.
Friedkin, N.E., 1998. A Structural Theory of Social Influence. Cambridge University
Press, New York.
Jeffery, R.W., Drewnowski, A., Epstein, L.H., Stunkard, A.J., Wilson, G.T., Wing, R.R.,
et al., 2000. Long-term maintenance of weight loss: current status. Health Psy-
chology 19 (Suppl. 1), 5–16.
Johnson, L., Mander, A.P., Jones, L.R., Emmett, P.M., Jebb, S.A., 2008. Energy-dense,
low-fiber, high-fat dietary pattern is associated with increased fatness in child-
hood. American Journal of Clinical Nutrition 87, 846–854.
Kobus, K., 2003. Peers and adolescent smoking. Addiction 98 (Suppl. 1), 37–55.
Koehly, L.M., Loscalzo, A., 2009. Adolescent obesity and social networks. Preventing
Chronic Disease 6(3): http://www.cdc.gov/pcd/issues/2009/jul/08 0265.htm.
Monge-Rojas, R., Nunez, H.P., Garita, C., Chen-Mok, M., 2002. Psychosocial aspects
of Costa Rican adolescents’ eating and physical activity patterns. Journal of Ado-
lescent Health 31, 212–219.
Olds, T., Dollman, J., Ridley, K., Boshoff, K., Hartshorne, S., Kennaugh, S., 2004.
Children and sport. Adelaide, South Australia: University of South Australia, Aus-
tralian Sports Commission. Published in digital form at http://www.ausport.gov.
au/research/youthandsport2004full.asp.
Pearson, M., Steglich, C., Snijders, T., 2006. Homophily and assimilation among sport-
active adolescent substance users. Connections 27, 47–63.
Peterson, C., 1989. Looking Forwards Through the Life Span, 2nd ed. Prentice Hall,
Singapore.
Rice, R.E., Donohew, L., Clayton, R., 2003. Peer network, sensation seeking, and drug
use among junior and senior high school students. Connections 25, 32–58.
Robins, G.L., Pattison, P.E., Elliott, P., 2001. Network models for social influence
processes. Psychometrika 66, 161–190.
Robins, G., Pattison, P., Kalish, Y., Lusher, D., 2007. An introduction to expo-
nential random graph (p*) models for social networks. Social Networks 29,
173–191.
Robins, G., Pattison, P., Wang, P., 2009. Closure, connectivity and degree distribu-
tions: exponential random graph (p*) models for directed social networks. Social
Networks 31, 105–117.
Ryan, A.M., 2001. The peer group as a context for the development of young adoles-
cent motivation and achievement. Child Development 72, 1135–1150.
Snijders, T.A., Pattison, P.E., Robins, G.L., Handcock, M.S., 2006. New specifications
for exponential random graph models. Sociological Methodology 36, 99–153.
Steglich, C.E.G., Snijders, T.A.B., Pearson, M., submitted for publication. Dynamic
networks and behavior: separating selection from influence.
Strauss, R.S., Pollack, H.A., 2003. Social marginalization of overweight children.
Archives of Pediatrics & Adolescent Medicine 157, 746–752.
Trogdon, J.G., Nonnemaker, J., Pais, J., 2008. Peer effects in adolescent overweight.
Journal of Health Economics 27, 1388–1399.
Unger, J.B., Reynolds, K., Shakib, S., Spruijt-Metz, D., Sun, P., Johnson, C.A., 2004.
Acculturation, physical activity, and fast-food consumption among Asian-
American and Hispanic adolescents. Journal of Community Health 29, 467–481.
Valente, T.W., Watkins, S.C., Jato, M.N., Van Der Straten, A., Tsitsol, P.M., 1997. Social
network associations with contraceptive use among Cameroonian women in
voluntary associations. Social Science & Medicine 45, 677–687.
Valente, T.W., Fujimoto, K., Chou, C., Spruijt-Metz, D., 2009. Adolescent affiliations
and adiposity: a social network analysis of friendships and obesity. Journal of
Adolescent Health 45, 202–204.
Voorhees, C.C., Murray, D., Welk, G., Birnbaum, A., Ribisil, K.M., Johnson, C.C., et al.,
2005. The role of peer social network factors and physical activity in adolescent
girls. American Journal of Health Behavior 29, 183–190.
Wang, P., Robins, G.L., Pattison, P.E., 2006. PNet: a program for the sim-
ulation and estimation of exponential random graph models. School of
Behavioural Science, University of Melbourne. Available from: http://www.sna.
unimelb.edu.au/pnet/pnet.html.
WHO, 2003. Obesity and overweight. Published in digital form at: http://
www.who.int/hpr/NPH/docs/gs obesity.pdf.
Williams, P.G., Holmbeck, G.N., Greenley, R.N., 2002. Adolescent health psychology.
Journal of Consulting and Clinical Psychology 70, 828–842.
Zametkin, A.J., Zoon, C.K., Klein, H.W., Munson, S., 2004. Psychiatric aspects of child
and adolescent obesity: a review of the past 10 years. Journal of the American
Academy of Child and Adolescent Psychiatry 43, 134–150.
