The Impact of Peer Social Networks on Adolescent

The Impact of Peer Social Networks on Adolescent. Alcohol Use Initiation Marlon P. Mundt, PhD. From the Department of Family Medicine, School of Medicine and Public Health, University of Wisconsin, Madison, Wis No direct support was received from grant P01-HD31921 for this analysis. Address correspondence to Marlon P. Mundt, PhD, Department of Family Medicine, 1100 Delaplaine Ct, Madison, Wisconsin 53715 (e-mail:

marlon.mundt@fammed.wisc.edu).Received for publication February 23, 2011; accepted May 16, 2011.

ABSTRACT

A C

OBJECTIVE: Early adolescent alcohol use is a major public health problem. Drinking before the 14th birthday is associated with a fourfold increase in risk of alcohol dependence in adult- hood.

The objective of this study is to evaluate the association between adolescent social network characteristics and alcohol initiation prospectively over time. METHODS: The study analyzes data from the National Longi- tudinal Study of Adolescent Health, a nationally representative survey of 7th- through 11th-grade students enrolled between 1995 and 1996.

Generalized estimating equations are used to model the risk of alcohol use initiation at 1-year follow-up among nondrinkers at wave 1 of the study. RESULTS: Both an adolescent’s friends’ alcohol use and the adolescent’s social network characteristics displayed an inde- pendent main effect on alcohol initiation. In comparison with abstainers, alcohol initiators had more popular friends as measured by more peer nominations as friends (indegree)

CADEMIC PEDIATRICS opyright ª 2011 by Academic Pediatric Association 414

and having more friends up to 3 steps removed (3-step reach), and more friends who drank. An adolescent’s risk of alcohol use onset increased 13% (95% CI, 4%–22%) for every additional friend with high indegree, 3% (95% CI, 0.3%–6%) for every additional 10 friends within 3-step reach, and 34% (95% CI, 14%–58%) for each additional friend who drank alcohol, and after controlling for confounders. CONCLUSION:

The findings suggest that, in addition to well- established demographic risk factors, adolescents are at height- ened risk of alcohol use onset because of their position in the social network in relationship to their friends and the friends of their friends.

KEYWORDS: alcohol; adolescent; social network; peer

ACADEMIC PEDIATRICS 2011;11:414–421

WHAT’S NEW

Peer social networks impact adolescent alcohol use onset. Alcohol initiators have more friends and friends of friends who drink, are in closer proximity to more popular individuals, and interact with more friends, and more friends of friends, than abstainers.

ADOLESCENT ALCOHOL INITIATION is a major public health problem. One quarter of all adolescents begin drinking alcohol by the age of 13 years old.1 Drinking before the 14th birthday is associated with a fourfold increase in risk of becoming alcohol dependent in adulthood.2 Early alcoholinitiationleadstoavarietyofriskyadolescentbehav- iors, including marijuana and cocaine use, having sex with multiple partners, and academic underperformance.3

Peer influence has been shown to play a large role in adolescent alcohol initiation. Peer alcohol use and a best friend’s alcohol drinking behavior predict alcohol initia- tion among middle and high school students.

4,5 Middle

schoolers are more likely to begin using alcohol if they perceive a higher prevalence of alcohol use among other students in their grade.6,7

Social network analysis is the optimal research tool for studying the peer dynamics of adolescent alcohol use,8–10

because it maps out peer dyadic interdependencies in a larger social group (ie, network) context. In the social network analysis framework, adolescents and their friends indicate their friendships by naming a list of friends. The friendships are then represented by social tie connections on a graph.

As a result, the individuals (ie, egos) are directly linked to their friends (ie, alters) and indirectly to the friends of their friends. Thus, they form a large social network of relationships.

The unique feature of social network analysis is that it relies on friendship ties reported by each adolescent and not on perceived, reputational, or idealized social connections from one person’s point of view.

11,12 Adolescent social networks are ideally suited for social network analysis as they have naturally occurring friendship boundaries (ie, schools). There is no clear agreement on how social network

effects influence alcohol drinking among adolescents.8, 13–17

Some studies have found that isolates, teenagers who have relatively few or no friends, are more likely to consume alcohol.14,15,17 Other investigations indicate higher drinking prevalence among liaisons, school-age students who have friends but are not connected primarily to a single group of friends.

13 Popular adolescents, those who receive more friendship nominations from other students (ie, high indegree) appear to enjoy greater social status if they drink.

18 Individuals who are more centrally

Volume 11, Number 5 September–October 2011

ACADEMIC PEDIATRICS ADOLESCENT ALCOHOL USE IN PEER GROUPS 415

located in the social network (ie, higher centrality), by virtue of having more social ties and more thoroughly in- terconnected friends, are more likely to use alcohol if their friends drink.8,19 Social proximity (ie, 3-step reach) via friends of friends’ ties to other alcohol users is also linked to adolescent alcohol use.8 Dense social networks with more interconnected individuals are associated with alcohol use.8

Although adolescent social network characteristics are shown to play a role in adolescent alcohol consumption, little is known on how these factors lead to the onset of adolescent alcohol use.

To fill this gap in the literature, the present study will investigate the association between adolescent social network characteristics identified in the previous studies, such as social status, social embedded- ness, social proximity to alcohol users, and overall network interconnectedness, to adolescent alcohol initiation prospectively over time. The present study will explore the following research questions:

1. Is social status, as measured by indegree, associated with adolescent alcohol initiation?

2. Is social embeddedness in the social network, as measured by centrality, linked to adolescent alcohol onset?

3. Is proximity to alcohol users, as measured by 3-step reach, correlated with adolescent alcohol inception?

4. Is overall network connectedness, as measured by network density, related to the start of adolescent alcohol drinking?

METHODS

DATA SOURCE

The data source for this study is the National Longitu- dinal Study of Adolescent Health (Add Health).20 Add Health is a longitudinal, school-based study of adolescents in grades 7 to 12. Sample schools were selected through stratified sampling to be representative of high schools nationwide based on region of the country, urbanicity, school funding, and racial composition.

Middle school or junior high feeder schools for the participating high schools were also recruited. The Add Health study was approved by the institutional review board of the Univer- sity of North Carolina at Chapel Hill, and the current analysis was approved by the institutional review board at the University of Wisconsin–Madison.

An in-school survey was given to all 7th through 12th graders at the 132 participating schools. Students who re- sponded to the in-school survey (N ¼ 90 118), were eligible to be randomly selected for an in-home interview and parent survey (wave 1, n ¼ 20 745).

Wave 1 was con- ducted from April 1995 to December 1995. In-home inter- view respondents participated in wave 2 (n ¼ 14 738) of the study 1 year later, from April 1996 to August 1996.

A novel aspect of Add Health is the collection of peer social network data as a means of understanding adolescent health. Respondents named up to 5 male and 5 female friends from their school roster. To construct social

network variables, 21 Add Health excluded nominations

of students who did not complete the survey or whose name did not appear on the school roster. The Add Health publicly available social network data set provides individual and school-level network data on 75 781 adoles- cents.

SAMPLE

Wave 1 of Add Health surveyed 20 745 individuals. Of the wave 1 sample, 4039 (19%) were in 12th grade and excluded from the wave 2 sample. Of the remaining 16 706 students eligible for wave 2, 1968 students (11.8%) were lost to follow-up or refused participation, for a total wave 2 sample size of 14 738 students. The total sample size for the analysis was 2610 students.

The sample includes Add Health subjects who 1) completed in-home interviews at both wave 1 and wave 2, 2) had never drunk alcohol outside of the presence of a parent or adult family member prior to the wave 1 in- home survey, and 3) named at least 1 friend who also completed an in-home interview at wave 1.

A total of 1592 students (11%) were excluded from the analysis sample because they had already initiated alcohol use at wave 1, and 10 536 students (71%) were excluded because they did not name as a friend any student who completed a wave 1 survey.

MEASURES

ALCOHOL USE INITIATION

At both wave 1 and wave 2, adolescents were asked if they had ever drunk beer, wine, or liquor when they were not with their parents or other adult family member. Alcohol use initiators were defined as adolescents who had not consumed alcohol outside their family group at wave 1, but who at wave 2 had drunk alcohol without the presence of their relatives.

SOCIAL NETWORKS

Social network variables were based on friendship nomi- nations from the initial in-school survey and were provided in the Add Health dataset. The Add Health social network measures were calculated using social network analysis methodology.21 Indegree is the number of friendship nomi- nations received by the respondent from the other study participants.

Centrality (Bonacich b) is the relative number of connections that an individual’s friends have within the adolescent social network. 3-step reach is the degree to which a member of the peer social network can make contact with other members of the network through 3 steps of friendship connections.

A school-level measure, density, is the number of ties in the total school peer social network divided by the number of possible network ties.

DEMOGRAPHICS

Student gender, age, and race were collected in the in- home interview. Age was calculated to the nearest month. As a proxy for cognitive skills, participants completed a 5-minute picture vocabulary test. Respondents were

416 MUNDT ACADEMIC PEDIATRICS

also asked how often in the past week they participated in team sports.

FAMILY CHARACTERISTICS

Adolescents naming only 1 parent in their current house- hold roster were defined as living in a single parent house- hold. Study participants indicated how much fun they have with their family and whether they had gone shopping or to a movie or event with a parent in the past 4 weeks.

In the in- home interview at wave 1, parents offered information on how often they drink alcohol and how many times they had 5 or more drinks on a single occasion in the past month.

CENSUS BLOCK CHARACTERISTICS

The Add Health study used geocoding of addresses to link subject data to US census block data. Census block data included the percentage of families in the block who were at or below the poverty level, the percentage over age 25 who had completed a college degree, and the percentage in the block who reported themselves to be religious adherents.

SCHOOL CHARACTERISTICS

Schools were characterized as urban, suburban, or rural, and from the northeast, midwest, south, or east portion of the United States. Schools were listed as public or private, and either small (400 or less students), medium (401–1000 students), or large (1001 or more students).

School admin- istrators indicated if school staff had training in alcohol/ drug prevention.

FRIEND CHARACTERISTICS

The Add Health alcohol use of adolescents’ friends, grade point average, delinquency scale score, and parent alcohol use were derived directly from the friends’ answers to the Add Health in-home interviews. The delinquency scale was constructed in a manner similar to prior research using the Add Health data.22 Add Health respondents were provided with a 15-item delinquency scale and were asked to indicate how often they had engaged in each behavior in the past year. Items included vandalism, physical fighting, stealing, lying, joyriding, breaking and entering, and drug use, among others.

Responses to individual questions were coded as 0 being never, 1 being 1 or 2 times, 2 being 3 or 4 times, and 3 being 5 or more times. The delinquency scale score was created by summing together responses to each of the 15 items.

STATISTICAL ANALYSIS

Each observation in the data set represented a single adolescent-friend pair. A dichotomous variable indicated whether the adolescent initiated alcohol use in wave 2 of the study.

Multilevel modeling using generalized esti- mating equations (GEE) adjusted for a friend having multiple nominations. An independent working correlation structure was applied for the clusters.

First, the analysis estimated a reduced-form GEE model of adolescent characteristics associated with alcohol use

initiation. The model included demographics, parent and family characteristics, census block characteristics, area of the country, school size and funding, school staff training in alcohol prevention, and school-wide social network density. Second, the study estimated the impact of friend charac-

teristics on alcohol use initiation while including all of the reduced-form variables as control variables. Friend charac- teristics included grade point average, delinquency scale score, parent drinking, and friend drinking at wave 1. Third, a GEE model was constructed to test the influence

of friends’ social network characteristics on the adoles- cent’s alcohol initiation status, while excluding friend drinking from consideration. The model included social network parameters of indegree, centrality and 3-step reach. These social network measures were chosen a priori based on research findings regarding friend influence.

8

A fourth GEE model analyzed both friend drinking char- acteristics and friend social network characteristics while controlling for confounders. This model sought to deter- mine the independent main effect of social network charac- teristics on an adolescent’s alcohol use initiation after controlling for the friend’s drinking status.

Increased likeli- hood of alcohol initiation was calculated by exponentiating the beta coefficients in the model. Additional analyses tested various model interaction terms. Next, alcohol initiator social network characteristics

(in-degree, centrality, and 3-step reach) were compared with abstainer social network characteristics. T tests and chi-square tests contrasted the social network characteris- tics and the prevalence of alcohol use for the friends of initiators and abstainers. Friends were classified as being 1, 2, or 3 steps away from the initiator or abstainer based on the minimum number of friendship steps it took to reach the friend from the study participant.

For example, a directly named friend is 1 step away from an individual. A friend of a friend who is not directly named by the indi- vidual is 2 steps away. A friend of a friend of a friend who cannot be reached in the 1-step or 2-step manner described above is defined as being 3 steps away. All analyses were carried out with SAS statistical software (SAS 9.1.3, SAS Institute Inc, Cary, NC). Finally, the friendship networks for a sample initiator

and abstainer from the same grade were plotted using NetDraw software (Analytic Technologies, Lexington, KY). The diagram provides an indication of the 3-step reach of both adolescents and the degree of alcohol use within their networks.

RESULTS The study sample consisted of 2610 seventh through

eleventh grade students (Table 1). Subjects ranged from 12 to 19 years of age, with a mean age of 15. Forty-five percent of participants were minorities.

Over a quarter of respondents lived in single parent households. More than 40% had parents who consumed alcohol and 9% had a parent who consumed 5 or more drinks in a single sitting in the past month. A greater part of the respondents lived in a suburban area, were from the South, and attended a large

Table 1. Descriptive Statistics of School-Age Students Who Had

Not Initiated Alcohol Drinking in National Longitudinal Study of

Adolescent Health, Wave 1, 1995 (N ¼ 2610) Characteristic

Demographics

Male, % 48.8 Age, mean (SE) 15.0 (0.3) Age, range 12–19 Grade level, %

7th grade 19.9 8th grade 18.7 9th grade 17.2 10th grade 24.1 11th grade 20.1

Race, % Non-Hispanic white 54.6 Black 19.5 Native American 1.6 Asian 10.0 White Hispanic 12.5

Add Health picture vocabulary test, mean (SE) 100.9 (0.3) Participate in team sports, % 76.7 Family Characteristics

Parent consumed alcohol, past year, % 43.9 Parent consumed 5þ drinks, past month, % 9.0 Single parent household, % 25.4 Family has fun together (quite a bit/very much), % 70.1 Shopped together, past 4 weeks, % 75.6 Went to movie/event together, past 4 weeks, % 29.0 Census Block Characteristics

Families with income below poverty level, % 10.7 Population aged 25þ years with college degree, % 22.9 Proportion who are religious adherents, % 56.4 School Characteristics

Urbanicity, % Urban 24.0 Suburban 51.4 Rural 24.6

Region of residence, % Northeast 11.9 Midwest 28.4 South 31.7 West 28.0

School funding, % Public 87.1 Private 12.9

School size, % Small (1–400 students) 30.2 Medium (401–1000 students) 28.5 Large ($1001 students) 41.3

School staff training in alcohol/drug prevention, % 78.7 Friend Characteristics

Male, % 47.3 Age, y, mean (SE) 15.2 (0.3) Age, y, range 12–19 Drink alcohol, wave 1, % 35.1 Drink alcohol, wave 2, % 33.9

ACADEMIC PEDIATRICS ADOLESCENT ALCOHOL USE IN PEER GROUPS 417

public school (>1000 students). Over three quarters of the participants’ schools provided training for staff in alcohol and drug prevention.

Table 2 presents the results from the 4 multivariate GEE models for alcohol initiation. Twenty percent (n ¼ 523) of the 2610 adolescents who were nondrinkers in wave 1 initi- ated alcohol use by wave 2 of the study. The analysis data set comprised 5096 friendship pairs, for an average of 1.95 nominated friends for each individual in the sample. In the

reduced form model 1, significant predictors of alcohol use initiation were older age, white race, participating in team sports, heavy drinking by a parent, and higher social networkdensityin the school.Variablesthat wereassociated with a reduced likelihood of alcohol use initiation were having family fun together and being in a private school. Model 2 added friend characteristics to the model of

alcohol use initiation. Friend drinking at wave 1 increased the risk of alcohol use initiation. Having a friend with a higher delinquency score also increased likelihood of alcohol use initiation. Model 3 included the social network characteristics of the nominated friend while removing friend drinking from consideration. Having more popular friends, as measured by peer nominations (indegree) and being able to reach a greater number of friends (3-step reach), was highly predictive of alcohol use initiation. Model 4 presents the full model results, which include

both the friend’s network characteristics and the friend’s alcohol use. Two of the 3 friend social network character- istics (ie, indegree, 3-step reach) increased the risk for the student to initiate alcohol use. For every additional friend with high indegree, the likelihood that an adolescent initi- ated alcohol use increased by 13% (95% CI, 4%–22%). For every additional 10 friends within 3-step reach of a nomi- nated friend, risk of alcohol initiation by a nondrinker increased by 3% (95% CI, 0.3%–6%). Risk of alcohol use onset increased 34% (95% CI, 14%–58%) for each additional friend who drank alcohol. Additional analyses revealed that neither the interaction term between friend 3-step reach and drinking status nor the interaction between friend indegree and drinking behavior added significantly to the explanatory power of the model. Of note, friend centrality was not significant in the model. More network ties, as opposed to being highly embedded in a tight network, appeared to be the factor that had the strongest impact on alcohol initiation. Table 3 presents the social network characteristics of the

alcohol initiators’ friends up to 3 steps removed compared with the abstainers’ friends at 3 degrees of separation. These analyses were performed post hoc based on the significant social network variables found in the GEE models. The results indicate that drinking initiators have more friends within 3 steps of them (3-step reach) prior to starting drinking. The drinking initiator’s extended circle of friends also has more popular (indegree), more connected (3-step reach), and more alcohol drinking friends within 3 steps. The Figure provides a visual representation at wave 1 of

the 3-step networks of 2 adolescents at wave 1 from the same grade: one who initiates alcohol use by wave 2, and the other who remains an abstainer. The initiator has access to more social ties 3 steps removed from him/her and more alcohol drinking friends. The abstainer has fewer friends within 3-step reach.

DISCUSSION The objective of this study is to evaluate the association

between adolescent social network characteristics and

Table 2. GEE Model for Alcohol Use Initiation Among Adolescents in National Longitudinal Study of Adolescent Health (N ¼ 2610)†

Parameter

Model 1 Model 2 Model 3 Model 4

Beta SE P Value Beta SE P Value Beta SE P Value Beta SE P Value

Demographics Male 0.142 0.075 .058 0.077 0.075 .307 0.108 0.076 .155 0.080 0.076 .288

Age 0.086 0.031 .005 0.081 0.031 .008 0.089 0.031 .004 0.083 0.031 .007

Race, white 0.278 0.112 .013 0.297 0.112 .008 0.255 0.113 .024 0.257 0.112 .022

AH picture vocabulary‡ 0.001 0.003 0.660 0.002 0.003 .401 0.002 0.003 .472 0.002 0.003 .456

Team sports 0.084 0.033 .011 0.091 0.033 .006 0.084 0.033 .012 0.085 0.034 .011

Family characteristics

Parent drinking frequency 0.068 0.036 .060 0.069 0.036 .060 0.063 0.037 .086 �0.005 0.035 .893 Parent heavy drinking 0.113 0.056 .042 0.104 0.055 .060 0.112 0.056 .046 0.087 0.056 .122

Single parent household 0.011 0.088 .902 �0.020 0.089 .823 �0.010 0.088 .909 �0.017 0.089 .845 Family fun together �0.146 0.038 <.001 �0.136 0.038 <.001 �0.142 0.038 <.001 �0.138 0.038 <.001 Shopped together �0.101 0.086 .241 �0.072 0.088 .414 �0.090 0.087 .304 �0.073 0.088 .406 Movie/event together �0.012 0.084 .888 �0.002 0.086 .986 �0.010 0.086 .895 �0.012 0.086 .889

Census block characteristics

Below poverty percentage 0.421 0.429 .326 0.514 0.430 .232 0.548 0.426 .198 0.572 0.425 .179

College educated 0.387 0.323 .230 0.494 0.324 .127 0.522 0.323 .107 0.552 0.323 .087

Religious adherents 0.105 0.382 .784 0.143 0.380 .707 0.156 0.384 .684 0.136 0.381 .721

Geography

Urban 0.029 0.145 .841 0.082 0.147 .577 0.123 0.149 .408 0.173 0.150 .247 Suburban 0.010 0.121 .938 0.066 0.123 .593 0.085 0.121 .484 0.116 0.122 .344

West �0.262 0.168 .119 �0.216 0.169 .200 �0.143 0.170 .399 �0.136 0.169 .423 Midwest �0.101 0.139 .468 �0.079 0.140 .570 �0.119 0.140 .398 �0.106 0.140 .450 South �0.127 0.145 .383 �0.064 0.146 .664 �0.127 0.149 .397 �0.115 0.149 .441

School characteristics

Private �0.385 0.160 .016 �0.335 0.160 .036 �0.374 0.161 .020 �0.358 0.161 .027 Small �0.200 0.139 .151 �0.159 0.139 .252 �0.105 0.147 .478 �0.093 0.147 .530 Medium �0.127 0.126 .315 �0.103 0.127 .417 �0.129 0.126 .305 �0.111 0.127 .381 Staff training in alcohol prevention �0.040 0.112 .724 �0.053 0.111 .633 �0.091 0.112 .416 �0.081 0.112 .466 Social network density 0.912 0.383 .017 0.885 0.386 .022 1.083 0.389 .005 1.069 0.391 .006

Friend characteristics

Friend grade point average �0.082 0.055 .141 �0.116 0.056 .037 �0.105 0.056 .059 Friend delinquency 0.020 0.007 .004 0.027 0.007 <.001 0.019 0.007 .007

Friend parent drinking 0.000 0.035 .995 0.006 0.035 .873 �0.005 0.035 .893 Friend parent heavy drinking 0.071 0.057 .208 0.082 0.056 .142 0.087 0.056 .122

Friend drinking, wave 1 0.320 0.082 <.001 0.295 0.082 <.001

Friend social network characteristics

Friend indegree 0.034 0.011 .001 0.032 0.010 .002

Friend centrality �0.098 0.088 .266 �0.093 0.088 .288 Friend reach, per 10 friends 0.031 0.013 .014 0.030 0.013 .020

Intercept �2.814 0.712 <.001 �3.015 0.727 <.001 �3.252 0.731 <.001 �2.992 0.725 <.001 Deviance 4981.6 4943.0 4934.9 4925.7

df 5071 5066 5064 5063

Bold items are significant at the p < .05 level.

†GEE ¼ generalized estimating equations. ‡AH ¼ National Longitudinal Study of Adolescent Health.

4 1 8

M U N D T

A C A D E M IC

P E D IA T R IC S

Table 3. Social Network Characteristics of Adolescent Drinking Initiators’ Friends in National Longitudinal Study of Adolescent Health

(N ¼ 2610)

Social Network Variable, Wave 1

Drinking Initiator at Wave 2

(n ¼ 523) Mean (SD) Abstainer at Wave 2

(n ¼ 2087) Mean (SD) One step away friends†

Indegree 6.28** (0.12) 5.71** (0.06) Centrality (Bonacich b) 0.98 (0.02) 0.96 (0.01) Reach (3-step) 62.39** (1.25) 56.01** (0.60) Drank alcohol, wave 1, % 44.6** 32.0**

Two steps away friends‡ Indegree 6.89** (0.13) 6.41** (0.06) Centrality (Bonacich b) 1.02 (0.02) 1.01 (0.01) Reach (3-step) 71.57** (1.39) 62.08** (0.71) Drank alcohol, wave 1, % 45.9** 39.6**

Three steps away friends§ Indegree 6.95** (0.08) 6.54** (0.05) Centrality (Bonacich b) 1.04 (0.01) 1.04 (0.01) Reach (3-step) 78.38** (0.93) 71.33** (0.53) Drank alcohol, wave 1, % 46.3* 43.9*

*Significant at p < .05 level.

†Friend directly named by the individual.

‡Friend of a friend who is not directly named as friend by the individual.

§Friend of a friend of a friend who is not 1 or 2 steps away from the individual.

**Significant at p < .001 level.

ACADEMIC PEDIATRICS ADOLESCENT ALCOHOL USE IN PEER GROUPS 419

alcohol initiation prospectively over time. The study results demonstrate that both the friend’s alcohol use and the adolescent’s social network characteristics display an independent main effect on alcohol initiation. In line

Figure. 3-step reach at wave 1 of an alcohol initiator and an alcohol absta

wave 1. Alcohol initiator began using alcohol by wave 2.

with previous research,8,23 a best friends’ drinking at wave 1 was a significant predictor of alcohol initiation at wave 2. Similar to other investigations, the study findings demonstrate that, in addition to well-established

iner. Both alcohol initiator and alcohol abstainer were nondrinkers at

420 MUNDT ACADEMIC PEDIATRICS

demographic risk factors (eg, age, race, team sports), peers’ immediate drinking friends are risk factors for alcohol use inception.

Interestingly, having friends with more friends, regard- less of their drinking status, impacts the likelihood of alcohol initiation. For every additional 10 friends within 3-step reach of an adolescent, risk of alcohol initiation increases by 3%. The findings are in concordance with the results of the Framingham Heart Study, where adults up to 3 degrees removed from the individual influence weight gain, cigarette cessation, and alcohol use.

24–26

Similar clustering effects are demonstrated in studies of health behavior, such as vaccination decisions among college students who coordinate their flu shots with their friends.

27

The findings suggest that potentially limiting the size of adolescent groupings may have a positive effect on delay- ing alcohol initiation. In this case, the study results argue for smaller schools, as they provide a smaller number of peers an adolescent can reach on their own or through their friends. This reasoning may also explain why private schools show protective effects against alcohol initiation in the model. Interestingly, a new generation of on-line social networks (Path, GroupMe, Rally Up, Shizzlr) focuses on limiting the size of the friendship group.

28

In this study, adolescents in higher density school networks were more likely to initiate alcohol use. More dense networks exhibit more interconnected clusters that magnify the spread of influence. Notably, the results come to light in view of computer simulations showing that more dense networks amplify the dynamics of influ- ence cascades.

29,30 Future research may want to explore how the density of virtual social communities (eg, Facebook), which connect a great number of adolescents on-line, influence alcohol drinking among adolescents.

It should be noted that, in the current sample, alcohol initiators are closer through their friendship connections to more popular adolescents—defined here as individuals with more peer nominations (indegree)—than abstainers. For every additional friend with high-popularity status (in-degree), the likelihood that an adolescent initiates alcohol use increases by 13%. Our findings are in line with research showing popularity status and conforming to peer alcohol use are linked.

18 More desirable students with more social connections may serve as positive or negative opinion leaders who could influence the behavior of others. They may be critical in efforts to delay alcohol initiation. Studies on the immunization of complex networks (eg, sexual partnership Web, the Internet) confirm that immunization/intervention efforts targeting highly connected nodes (eg, most promiscuous individuals or high-traffic routers) will greatly reduce a networks’ vulnerability to virus outbreaks.

31–33

These data demonstrate that parental modeling of respon- sible alcohol use and having fun together as a family offer protective benefit against adolescent alcohol initiation. The results are similar to previous research showing that low family bonding and parental drinking are linked to the onset of alcohol consumption.34,35 Health care

professionals may wish to establish community partnerships for building stronger families that encompass spending quality time together. More research on fostering conditions for families to have fun together is warranted. Future studies may wish to explore how cascades

of influence to initiate drinking are driven, whether temporal patterns of the social network matter for alcohol initiation, and how adolescent social networks can be exploited to promote healthy choices with regard to alcohol. This study has several limitations. First, it relies on

participant self-report of alcohol initiation, although self-reported substance consumption is generally perceived as a valid measure.

36,37 Second, although the study design takes advantage of longitudinal data, it is not possible to distinguish between 2 potential causes of behavioral clustering: induction, or the direct influence of one individual on another, and homophily, the tendency of persons to choose to associate with similar individuals. This is left for future investigation. Third, this study is limited to individuals who provided school peer group data. For many adolescents, the peer network includes students outside of their particular school, which were not available for the analysis. Finally, the study results are susceptible to potential selection and sample biases. Subjects were excluded from the analysis if they had no friends who completed the Add Health wave 1 survey. This study does not attempt to draw conclusions about students who are isolated from their school peer group. Subjects were also excluded if they did not complete the wave 2 survey. Over 88% of eligible participants completed wave 2 in-home interviews, making it unlikely that the results suffer from significant response bias. Nonresponse has been investigated by the Survey Research Unit at the University of North Carolina, and findings showed that bias for measures of health and risk behaviors rarely ex- ceeded 1% in either wave 1 or wave 2.

38

CONCLUSION The findings suggest that, in addition to well established

demographic risk factors, adolescents are at heightened risk of alcohol use onset because of their position in the social network in relationship to their friends and the friends of their friends.

ACKNOWLEDGMENT I wish to thank Add Health, a program project directed by Kathleen

Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and

Kathleen Mullan Harris at the University of North Carolina at Chapel

Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver

National Institute of Child Health and Human Development, with cooper-

ative funding from 23 other federal agencies and foundations for the use of

the data. Special acknowledgment is due Ronald R. Rindfuss and Barbara

Entwisle for assistance in the original design. This research was funded by

grant P01-HD31921 from the Eunice Kennedy Shriver National Institute

of Child Health and Human Development, with cooperative funding from

23 other federal agencies and foundations. This work was supported by

a grant to Marlon Mundt from the National Institute on Alcohol Abuse

and Alcoholism, NIAAA 1K01 AA018410-01.

ACADEMIC PEDIATRICS ADOLESCENT ALCOHOL USE IN PEER GROUPS 421

REFERENCES 1. Eaten DK, Kann L, Kinchen S, et al. Youth risk behavior surveil-

lance–United States, 2005. MMWR CDC Surveill Summ. 2006;55:

1–108.

2. Hingson RW, Heeren T, Winter MR. Age at drinking onset and

alcohol dependence. Arch Pediatr Adolesc Med. 2006;160:739–746.

3. Grunbaum JA, Kann L, Kinchen S, et al. Youth risk behavior surveil-

lance–United States, 2003. MMWR Surveill Summ. 2004;53:1–96.

4. Ellickson PL, Hays RD. Antecedents of drinking among young

adolescents with different alcohol use histories. J Stud Alcohol.

1991;52:398–408.

5. Urberg KA, Degirmencioglu SM, Pilgram C. Close friend and group

influence on adolescent cigarette smoking and alcohol use. Dev

Psychol. 1997;33:834–844.

6. Graham JW, Marks G, Hansen WB. Social influence processes

affecting adolescent substance use. J Appl Psychol. 1991;76:291–298.

7. Marks G, Graham JW, Hansen WB. Social projection and social

conformity in adolescent alcohol use. Pers Soc Psychol Bull. 1992;

18:96–101.

8. Ennett ST, Bauman KE, Hussong A, et al. The peer context of adoles-

cent substance use: findings from social network analysis. J Res Ado-

lesc. 2006;28:159–186.

9. Valente TW, Gallaher P, Mouttapa M. Using social networks to under-

stand and prevent substance use: a transdisciplinary perspective. Subst

Use Misuse. 2004;39:1685–1712.

10. Valente TW. Social network influences on adolescent substance use:

an introduction. Connections. 2003;25:11–16.

11. Moody J, White DR. Structural cohesion and embeddedness:

a concept of social groups. Am Sociol Rev. 2003;68:103–127.

12. Crosnoe R. Friendships in childhood and adolescence: the life course

and new directions. Soc Psychol Quart. 2000;63:377–391.

13. Henry DB, Kobus K. Early adolescent social networks and substance

use. J Early Adolesc. 2007;27:346–362.

14. Fang X, Li X, Stanton B, Dong Q. Social network positions and

smoking experimentation among Chinese adolescents. Am J Health

Behav. 2003;27:257–267.

15. Ennett ST, Bauman KE. Peer group structure and adolescent cigarette

smoking: a social network analysis. J Health Soc Behav. 1993;34:

226–236.

16. Bauman KE, Ennett ST. Peer influence on adolescent drug use. Am

Psychol. 1994;49:820–822.

17. Pearson M, Sweeting H, West P, et al. Adolescent substance use in

different social and peer contexts: a social network analysis. Drugs

Educ Prev Policy. 2006;13:519–536.

18. Balsa AI, Homer JF, French MT, Norton EC. Alcohol use and popu-

larity: social payoffs from conforming to peers’ behavior. J Res Ado-

lesc. 2011;21:317–326.

19. Crosnoe R, Needham B. Holism, contextual variability, and the study

of friendships in adolescent development. Child Dev. 2004;75:

264–279.

20. Add Health Web site. Available at: http://www.cpc.unc.edu/

addhealth. Accessed February 23, 2011.

21. Wasserman S, Faust K. Social network analysis: methods and appli-

cations. New York, NY, USA: Cambridge University Press; 1994.

22. Beaver KM, Wright JP. Biosocial development and delinquent

involvement. Youth Violence Juv Justice. 2005;3:168–192.

23. Fisher LA, Bauman KE. Influence and selection in the friend-

adolescent relationship-findings from studies of adolescent smoking

and drinking. J Appl Soc Psychol. 1988;18:289–314.

24. Christakis NA, Fowler JH. The spread of obesity in a large social

network over 32 years. N Engl J Med. 2007;357:370–379.

25. Christakis NA, Fowler JH. The collective dynamics of smoking in

a large social network. N Engl J Med. 2008;358:2249–2258.

26. Rosenquist JN, Murabito J, Fowler JH, Christakis NA. The spread of

alcohol consumption behavior in a large social network. Ann Intern

Med. 2010;152:426–433.

27. Rao N, Mobius MM, Rosenblat T. Social networks and vaccination

decisions. FRB of Boston Working Paper No. 07–12, 2007. Available

at: http://ssrn.com/abstract ¼ 1073143. Accessed February 23, 2011. 28. Wortham J, Miller CC. Social networks offer a way to narrow the field

of friends. The New York Times. May 10, 2011;B1.

29. Pastor-Satorras R, Vespignani A. Epidemic spreading in scale-free

networks. Phys Rev Lett. 2001;86:3200–3203.

30. Watts DJ, Dodds PS. Influentials, networks, and public opinion

formation. J Consum Res. 2007;34:441–458.

31. Rowe DC, Rodgers JL. Adolescent smoking and drinking: are they

“epidemics”? J Stud Alcohol. 1991;52:110–117.

32. Cruz IY, Dunn ME. Lowering risk for early alcohol use by chal-

lenging alcohol expectancies in elementary school children. J Consult

Clin Psychol. 2003;71:493–503.

33. Kosterman R, Hawkins JD, Guo J, et al. The dynamics of alcohol and

marijuana initiation: patterns and predictors of first use in adoles-

cence. Am J Public Health. 2000;90:360–366.

34. Kuntsche EN, Kuendig H. What is worse? A hierarchy of family-

related risk factors predicting alcohol use in adolescence. Subst Use

Misuse. 2006;41:71–86.

35. Fisher LB, Miles IW, Austin B, Camargo CA, Colditz GA. Predictors

of initiation of alcohol use among US adolescents–findings from

a prospective cohort study. Arch Pediatr Adolesc Med. 2007;161:

959–966.

36. Post A, Gilljam H, Rosendahl I, et al. Validity of self-reports in

a cohort of Swedish adolescent smokers and smokeless tobacco

(snus) users. Tob Control. 2005;14:114–117.

37. Campanelli PC, Dielman T, Shope JT. Validity of adolescents’ self-

reports of alcohol use and misuse using a bogus pipeline procedure.

Adolescence. 1987;22:7–22.

38. Kalsbeek WD, Morris CB, Vaughn BJ. Effects of nonresponse on the

mean squared error of estimates from longitudinal study. 2001; ASA

Proceedings of the Joint Statistical Meetings, American Statistical

Association, Alexandria, VA.