A longitudinal social network analysis of peer influence, peer selection, and smoking behavior among adolescents in British schools

A Longitudinal Social Network Analysis of Peer Influence, Peer Selection,and Smoking Behavior Among Adolescents in British Schools

Liesbeth MerckenCardiff University and Maastricht University

Christian SteglichUniversity of Groningen

Philip SinclairCardiff University and Manchester Metropolitan University

Jo Holliday and Laurence MooreCardiff University

Objective: Similarity in smoking behavior among adolescent friends could be caused by selection of friendson the basis of behavioral similarity, or by influence processes, where behavior is changed to be similar to thatof friends. The main aim of the present study is to disentangle selection and influence processes and studychanges over time in these processes using new methods of longitudinal social network analysis. Methods:The sample consists of 1716 adolescents (mean age at baseline � 12.17 years, SD � .38) in 11 British schoolsparticipating in the control group of the ASSIST (A Stop Smoking in School Trial) study. The design waslongitudinal with three observations at one-year intervals. At each observation, participants were asked toreport on their smoking behavior and friendship networks. An actor-based model of friendship network andsmoking behavior coevolution (a statistical model for the simultaneously occurring changes in friendshipnominations and smoking) was analyzed, capable of modeling possible changes occurring between observa-tions, allowing alternative influence and selection mechanisms to be investigated, and avoiding the violationof assumptions of statistical independence of observed data. Results: Adolescent’s tendency to select friendsbased on similar smoking behavior was found to be a stronger predictor of smoking behavior than friends’influence. The proportion of smoking behavior similarity explained by smoking-based selection of friendsincreased over time, whereas the proportion explained by influence of friends decreased. Conclusions:Smoking prevention should not solely focus on social influence but also consider selection processes andchanges in both processes over time during adolescence.

Keywords: influence, selection, friendship networks, adolescent smoking behavior, social network analysis

Tobacco use is the second highest cause of mortality and thefourth most common risk factor for disease worldwide (WorldHealth Organization [WHO], 2007). In most Western countries,there is an increased prevalence of smoking during adolescence

(Chassin, Presson, Rose, & Sherman, 1996; U.S. Department ofHealth & Human Services, 1994). The ease with which youngpeople become addicted to nicotine means that once adolescentsstart to smoke, it is very hard to quit and they are more likely tobecome regular smokers than if they were to experiment later inlife (Fergusson, 1995; Prokhorov, Pallonen, Fava, Ding, & Niaura,1996; Stanton, 1995). Peer groups play a crucial role in adolescentsmoking behavior (Bauman, Fisher, Bryan, & Chenoweth, 1984;Eiser, Morgan, Gammage, Brooks, & Kirby, 1991; Ennett, Bau-man, & Koch, 1994; Sussman et al., 1990), as evidenced byprevious research suggesting that smoking behavior tends to besimilar among friends (Eiser et al., 1991; Ennett et al., 1994).Although it is typically assumed that peer influence (or peerpressure) is the key process driving the creation of this observedsimilarity, it is equally plausible that it is caused by smoking-basedselection processes (Cohen, 1977; Engels, Knibbe, Drop, & deHaan, 1997; Ennett & Bauman, 1994; Fisher & Bauman, 1988;Hoffman, Monge, Chou, & Valente, 2007; Kandel, 1978; Mer-cken, Snijders, Steglich, & De Vries, 2009; Pearson, Steglich, &Snijders, 2006; Steglich, Snijders, & Pearson, 2010; Urberg, Luo,Pilgrim, & Degirmencioglu, 2003). Whereas influence occurswhen adolescents assimilate to the smoking behavior of theirfriends, smoking-based selection occurs when adolescents selectnew friends based on similarities in their smoking behavior. Thetwo processes are very hard to disentangle because of methodolog-

This article was published Online First January 16, 2012.Liesbeth Mercken, Cardiff Institute of Society and Health, Cardiff

University, Cardiff, UK, and Department of Health Promotion, MaastrichtUniversity, Maastricht, the Netherlands; Christian Steglich, Department ofSociology, University of Groningen, Groningen, the Netherlands; PhilipSinclair, Cardiff Institute of Society and Health, Cardiff University, Car-diff, UK, and Computing and Mathematics, Manchester MetropolitanUniversity, Manchester, UK; Jo Holliday and Laurence Moore, CardiffInstitute of Society and Health, Cardiff University, Cardiff, UK.

We thank the editor and reviewers for their helpful suggestions. Thisstudy was funded by the Medical Research Council (grant no. G0501806).Data were collected during the Medical Research Council-funded StopSmoking in Schools Trial (grant no. G9900538). The study protocol wasapproved by the Wales Multi-Centre Research Ethics Committee. Thiswork was performed with help of Dr. James Osborne using the computa-tional facilities of the Advanced Research Computing @ Cardiff (ARCCA)Division, Cardiff University, Cardiff, UK.

Correspondence concerning this article should be addressed to Dr.Liesbeth Mercken, Maastricht University, Department of Health Promo-tion, P.O. Box 616, 6200MD Maastricht, the Netherlands. E-mail:[email protected]

Health Psychology © 2012 American Psychological Association2012, Vol. 31, No. 4, 450–459 0278-6133/12/$12.00 DOI: 10.1037/a0026876

450

ical challenges, but it is important to do so as the design ofprevention programs may be based on a false assumption that peerinfluence is the prime mechanism that needs to be interrupted byprevention efforts.

The main goal of the present article is to study changes overtime in selection and influence processes during adolescence. Aschildren grow older they have fewer environmental and parentalconstraints on their social contacts, gain a longer perspective ontheir friendships, and become more selective about their friends(Aboud & Mendelson, 1996). Behavioral and egocentric criteriafor selecting and maintaining friends might become less importantwhen social–cognitive skills mature, while sharing social andemotional experiences, and selecting friends based on criteria thatpredict friendship durability gain importance (Aboud & Mendel-son, 1996; Berndt & Hoyle, 1985; Brown, 1981). Smoking-basedselection and influence may therefore decrease when adolescentsgrow older.

Only two studies have examined changes in smoking-basedselection and influence processes over time (Mercken, Candel,Willems, & De Vries, 2009; Simons-Morton & Chen, 2006), butthe results were conflicting and not completely in line with expec-tations based on the previously mentioned studies (Aboud &Mendelson, 1996; Berndt & Hoyle, 1985; Brown, 1981). Mercken,Candel, Willems, & De Vries (2009) found that selection de-creased over time but influence of friends increased. Simons-Morton and Chen (2006) found that influence was stable over timefrom 6th to 9th grade, while selection was not significant exceptfrom 7th to 8th grade. Both studies, however, suffered from threeimportant methodological flaws. First, even the more advancedstatistical techniques such as structural equation modeling used forthis type of data assume incorrectly that there are no dependenciescaused by the network structure. However, a given individual’svalue on smoking behavior could appear within more than oneobservation, for example, as the smoking behavior dependentvariable for one case, and as smoking behavior of one of thefriends supplying data for the independent variables in other cases.Second, researchers did not account for the continuous changes ofnetwork structure and smoking behavior over time happeningbetween observations. Longitudinal data were gathered at only afew discrete moments in time, which makes it impossible tounequivocally identify the processes responsible for a network orbehavior change. In between two observations, a change may haveoccurred as well as a change back to the original value before thenext observation moment. Not accounting for the possibility ofother intervening changes may be misleading, and it is preferableto use a technique that does take this possibility into account.Third, although both studies included important alternative influ-ence processes, they did not control adequately for alternativeexplanatory selection mechanisms. A smoking adolescent canchoose a smoker as a friend because this individual already nom-inated the adolescent to be his friend (reciprocity) or because thisparticular individual was already a friend of the adolescent’s otherfriend(s) (transitivity) (Burk, Steglich, & Snijders, 2007; Mercken,Snijders et al., 2009). Additionally, the selection of this smokingfriend may also be based on similarities in gender and age insteadof similarities in smoking behavior (McPherson, 2001). In bothstudies, these methodological shortcomings may have led to in-correct conclusions about the presence of selection and influenceprocesses over time.

The present study will be the first to examine changes over timein smoking-based selection and influence processes taking thepreviously mentioned methodological flaws into account. Al-though data are gathered at discrete moments in time, the possibleoccurrence of changes in friendships and smoking behavior inbetween two observations will be accounted for using advancedlongitudinal social network analyses techniques (Snijders, Steg-lich, & Schweinberger, 2007; Steglich, Snijders, & West, 2006).This method uses a more complete representation of repeatedmeasures data on friendship networks and smoking behavior andcan take dependencies caused by the network structure and alter-native explanatory selection mechanisms into account. Using thisnovel approach, we will address three main research questions(RQ). RQ1: Do adolescents select friends based on similar smok-ing behavior? RQ2: Are adolescents influenced by friends to adjusttheir smoking behavior? RQ3: Does the strength of these processeschange over time?

Method

Participants

The sample consists of 1716 adolescents in 11 British secondaryschools that participated as part of the control group in ASSIST (AStop Smoking in Schools Trial), a cluster (group) randomized trialof a school-based, peer-led adolescent smoking prevention pro-gram (Campbell et al., 2008; Starkey, Moore, Campbell, Sidaway,& Bloor, 2005). The intervention, which was theoreticallygrounded in diffusion theory, involved influential Year 8 studentsbeing trained to encourage their peers, through informal conver-sations, to be smoke-free. Because in the intervention group therelationships between variables of interest might have been af-fected as a consequence of young people being involved in theintervention, only control group schools were included in thepresent study. From the 29 control schools in the trial, 11 wereselected based on location (six schools in Wales and five inEngland), size (five large schools with more than the mean numberof 182 students in Year 8, and six small schools), level of entitle-ment to free school meals (five with �19% of students entitled tofree school meals and six with �19% of students entitled to freeschool meals), and urbanization (six urban and five more rurallylocated schools).

Procedure

Questionnaires were administered in the school setting to allparticipating Year 8 (T1, age 12–13 years) students by ASSISTresearch staff (Starkey et al., 2005). Follow-up was conducted at1-year intervals when students were in Year 9 (T2, age 13–14years) and Year 10 (T3, age 14–15 years). All students were askedto complete questionnaires on smoking behavior and social net-works. To minimize reporting bias (Murray & Perry, 1987), self-reported data on smoking behavior were validated for all smokingor nonsmoking children in 24 schools (39% of the total population)by measuring and analyzing salivary cotinine concentrations ateach data sweep. These 24 schools represented a broad range ofdifferent types of participating schools. Data were only used toassess the amount of misreporting, not to correct self-reported data(Campbell et al., 2008). Discordance between self-reported smok-

451SOCIAL NETWORK ANALYSIS OF SMOKING BEHAVIOR

ing behavior and salivary cotinine findings was 1% at 1-yearfollow-up and 3% at 2-year follow-up among nonsmokers.

Measurements

Friendship ties were assessed by a social network questionnairein which respondents could name up to six closest friends of anyage, from inside or outside school (Wasserman & Faust, 1994).Only friends inside the same school year were included in thecurrent analyses, because only those also participated as respon-dents and filled out the questionnaire.

Adolescent smoking behavior was assessed by one questionwhich asked adolescents to report their own smoking behavior(1 � never smoked, 2 � I have only tried smoking once, 3 � Iused to smoke but never smoke cigarettes now, 4 � less than onecigarette a week, 5 � 1–6 cigarettes a week, 6 � more than 6cigarettes a week).

A number of variables known to be associated with adolescentsmoking were included as covariates: gender (1 � boy, 0 � girl),age (in years), and parental smoking (0 � none of my parentssmoke, 1 � at least one of my parents smoke). Socioeconomicstatus (SES) was measured using the Family Affluence Scale(Mullan & Currie, 2000) which is an index derived from threequestions: ‘Does your family have a car or van?’ (0 � no, 1 � yes,2 � yes, two or more), ‘Do you have your own bedroom foryourself?’ (0 � no, 1 � yes), and ‘During the last 12 months, didyou travel away on holiday with your family?’ (0 � not at all, 1 �once 2 � twice, 3 � more than twice). This Family AffluenceScale is completed more accurately by children than alternativemarkers of SES such as parental occupation or education (Wardle,Robb, & Johnson, 2002) and has been validated favorably againstsuch other markers of SES (Currie et al., 2008). To control forfriendship selection resulting from tutor group membership, tutorgroup was also included.

Plan of Analysis

A longitudinal social network model. An actor-based modelfor network-behavior coevolution will be analyzed, which is alongitudinal social network model that can simultaneously modelnetwork changes (e.g., whether an adolescent selects a new friendor returns a friendship) and behavior changes (e.g., whether anadolescent smokes more or less) in a complete network (e.g., alladolescents within a school). Each adolescent is assumed to makedecisions to change their friendship ties or their smoking behaviorin response to the current state of the adolescent’s network neigh-borhood and the distribution of smoking behavior in this neigh-borhood. The changes in friendship ties are modeled by what wewill call the ‘Friendship network evolution’ part of the model,whereas the changes in smoking behavior will be modeled by whatwe will call the model’s ‘Smoking behavior evolution’ part. Thefriendship network evolution part captures the rules that governadolescents’ friendship changes by a list of variables on whichfriendship choice probabilities depend. The smoking behaviorevolution part of the model captures, analogously, the rules thatgovern adolescents’ change in smoking behavior, again by a list ofvariables on which smoking behavior change probabilities depend.

Actor-based models for network-behavior coevolution use acontinuous-time Markov process that repeatedly imputes likely

behavioral and network change trajectories for the unobservedperiods between observation moments. The parameters expressingthe rules of network and behavior change are identified in aniterative, computationally intensive, simulation-based algorithmthat optimizes the match between the model-imputed trajectoriesand the observed data. The mathematical specification and statis-tical estimation procedures are described in detail by Snijders et al.(Snijders, 2001; Snijders, Steglich, & Schweinberger, 2007; Sni-jders, van de Bunt, & Steglich, 2010). Table 1 presents a descrip-tion of each of the included variables for both parts of the modeland a brief outline of the model is as follows.

Friendship network evolution. The friendship network evo-lution part of the model specifies the preferred direction of net-work change by a list of variables on which friendship choiceprobabilities may depend. These include current network structureand attributes of adolescents. Four smoking-related variables wereincluded: the effect of adolescent’s smoking behavior on numberof friends chosen (smoking behavior ego) to model whether ado-lescents smoking at higher rates choose more friends themselvescompared with nonsmokers or adolescents smoking at lower rates,the effect of adolescents’ smoking behavior on the probability ofthem being chosen as a friend by others (smoking behavior alter)to model whether adolescents smoking at higher rates have ahigher probability to be chosen as a friend compared with adoles-cents smoking at lower rates or nonsmokers, and the effect ofsimilar smoking behavior on friendship selection (smoking behav-ior similarity) to model whether adolescents select friends thatsmoke at a similar rate as they themselves do. The inclusion of thelatter effect will address RQ1. We also included the effect of thesquared value of adolescents’ smoking behavior on the probabilityof them being chosen as a friend (smoking behavior squared alter)to control for possible nonlinearities (Snijders et al., 2010), toexamine whether the probability of a smoker being chosen as afriend increases in linear relation to the extent of their smoking orwhether it reduces beyond an optimal point. It is possible thatadolescents tend to select friends who smoke but not those smok-ing at very high rates.

Several characteristics of the current network as well as variousindividual attributes were included as control variables becausenetwork dynamics have important endogenous components(Schaefer, Light, Fabes, Hanish, & Martin, 2010; Snijders, 2001;Van de Bunt, Van Duijn, & Snijders, 1999). These were thegeneral tendency to select friends (outdegree), the number ofreciprocal friends chosen (reciprocity), the tendency to selectfriends of friends (transitive ties and transitive triplets), the ten-dency to stay indirectly instead of directly (distance 1) connectedto other adolescents (number of actors at distance 2), and to formnonhierarchical groups (3-cycles). Similarity on age, gender, SES,and the effects of these adolescent attributes on the number offriends they choose themselves (age ego, gender ego, SES ego)and on the propensity to be chosen as a friend by others (age alter,gender alter, SES alter) were included as control variables. Theeffect ‘tutor group similarity’ controlled for selection based onbeing in the same tutor group within school. In Table 1 all includedeffects are explained in detail.

Smoking behavior evolution. The smoking behavior evolu-tion part of the model specifies a list of variables on whichsmoking behavior change probabilities may depend. The lowerpart of Table 1 gives a list of variables relating to the network,

452 MERCKEN, STEGLICH, SINCLAIR, HOLLIDAY, AND MOORE

smoking behavior, and other attributes on which probabilities ofchanges in smoking behavior may depend. The model contains onemain friendship-related influence component: the tendency tochange smoking behavior to become similar to current friends, toexamine whether adolescents are influenced by the smoking be-havior of their friends (RQ2). Control variables are age, gender,SES, parental smoking, and two shape effects: a linear interceptterm modeling average smoking, and a quadratic term capturingresidual variance in smoking.

Statistical Analysis

The developed actor-based model for smoking-friendship co-evolution will be analyzed for each school separately in twosuccessive periods using SIENA (Simulation Investigation forEmpirical Network Analysis) (Snijders, Steglich, Schweinberger,& Huisman, 2007; Steglich et al., 2010). Period 1 includes datafrom T1 and T2, and period 2 data from T2 and T3. In each schoolnetwork, all respondents were included. Respondents who entered

the study at a later time point than the baseline measurement orwho left the study at an earlier time point than the final measure-ment were included for the duration of their membership of theschool (Snijders, Steglich, Schweinberger et al., 2007). The in-cluded effects were tested on the basis of t ratios defined asestimate divided by standard error, with an approximate standardnormal null distribution (Snijders, 2001). Subsequently, the resultsof all separate school network analyses were combined in onemeta-analysis for each time period. The t ratios were combinedseparately for each of the effects described in Table 2. It wasdesirable to use a combination method with a good power to detectvarious patterns of nonzero parameter values across the 11 schoolswith a minimum of assumptions. For each effect, the overall nullhypothesis that the effect was 0 in all schools was tested byFisher’s combination procedure (Hedges & Olkin, 1985) with twoone-sided tests. The right-sided test examines the null hypothesisthat in all schools the effect was nonpositive. The alternativehypothesis is that in at least one school the effect was positive. The

Table 1Effects for Modeling the Probability of Changes in Friendship Ties and Smoking Behavior

Description

Friendship network evolutionSmoking behavior ego Association between smoking level and the tendency to nominate friendsSmoking behavior alter Association between smoking level and the tendency to be nominated as a friendSmoking behavior squared alter Marginal association between smoking level and the tendency to be nominated, controlling for the

previous effectSmoking behavior similarity (RQ1) Tendency to select a friend based on similar smoking behavior

Included control effects:Outdegree General tendency to select friends (density of the network)Reciprocity Tendency to return friendshipsTransitive ties Tendency to select a friend that is already friends with one of the adolescent’s other friendsTransitive triplets Tendency to befriend additional friends of friends, on top of the first such friend which the

previous effect (transitive ties) measuresNumber of actors at distance 2 Tendency to be indirectly (through one of your friends) instead of directly tied to others3-cycles Tendency to stay indirectly tied to others within a closed triad (adolescent ‘a’ has a friendship tie

directed to adolescent ‘b’, ‘b’ to ‘c’, and ‘c’ to ‘a’)Age ego Tendency for older adolescents to select more friends compared to younger adolescentsAge alter Tendency for older adolescents to be select more often as a friend compared to younger adolescentsAge similarity Tendency to select a friend based on similar ageGender ego Tendency for male adolescents to select more friends compared to female adolescentsGender alter Tendency for male adolescents to be select more often as a friend compared to female adolescentsGender similarity Tendency to select a friend based on similar genderSES ego Tendency for higher SES adolescents to select more friends compared to lower SES adolescentsSES alter Tendency for higher SES adolescents to be select more often as a friend compared to lower SES

adolescentsSES similarity Tendency to select a friend based on similar SESTutor group similarity Tendency to select a friend based on being in the same tutor group

Smoking behavior evolutionSmoking behavior friendsa (RQ2) Tendency to change smoking behavior to become similar to current friends

Included control effects:Tendency to smoke General tendency to smokeTendency to smoke squared Marginal tendency to smoke, controlling for the general tendencyAge adolescent Effect of an adolescent’s age on adolescent’s own smoking behaviorGender adolescent Effect of an adolescent’s gender on adolescent’s own smoking behaviorSES adolescent Effect of an adolescent’s SES on adolescent’s own smoking behaviorSmoking behavior parents Effect of parental smoking behavior on adolescent’s own smoking behavior

Note. The friendship evolution part of the model includes a higher number of effects compared with the smoking behavior evolution part. This differenceis attributable to the multidimensional nature of selection processes. The probability to select a friend in my example depends on the age of the adolescentthat selects, the age of the friend that gets selected, and on similarities in age of both. The effect of age on smoking behavior can be modeled by includingonly the effect of an adolescent’s age on their own smoking behavior.a Total similarity effect.

453SOCIAL NETWORK ANALYSIS OF SMOKING BEHAVIOR

left-sided test examines the null hypothesis that in all schools theeffect was non-negative. The alternative hypothesis is that in atleast one school the effect was negative. The test statistic inFisher’s procedure is minus twice the sum of the natural loga-rithms of the p values of the one-sided tests for the individualschools, with under the combined null hypothesis a chi-squareddistribution having, for 11 schools, 22 degrees of freedom. Tocontrol for multiple (right and left) testing, there was deemed to besignificant support for an effect if either of these combination testswas significant at level p � .025. The Fisher’s combination pro-cedure of one-sided tests is preferred above the Snijders–Baerveldtmethod for meta-analysis (two-sided test) (Snijders & Baerveldt,2003) as it does not make the assumption that estimated standarderrors and estimated parameter values are uncorrelated.

The null hypothesis that effect parameters are constant acrossschools was tested by the method of Cochran (1954) adapted fornetwork dynamics by Snijders and Baerveldt (Cochran, 1954;Snijders & Baerveldt, 2003).

In addition, network autocorrelations were examined to explorethe relative contribution of selection, influence, and alternativemechanisms that lead to observed smoking behavior similaritiesamong friends. By calculating the average similarity of linkedindividuals during each period in simulated models with coeffi-cients estimated under different model specifications, the relativecontributions of selection, influence, and control effects to ob-

served smoking similarity between friends can be expressed. As asimilarity measure of individuals linked in a network we usedMoran’s I, a spatial autocorrelation coefficient (Cliff & Ord,1981). This method is explained in detail elsewhere (Mercken,Snijders, Steglich, Vartiainen, & de Vries, 2010; Steglich et al.,2010). Two networks were excluded (n � 252) because of verylow smoking rates (mean smoking behavior � 2), which may biasthe results of these simulations.

Attrition Analysis

The sample consisted of 1716 adolescents. At the first measure-ment 1677 respondents participated. At the second measurement,1614 of 1677 respondents completed the questionnaire (96.24%).At the third measurement, 1540 of those 1614 respondents partic-ipated (95.42%). At the second measurement, 39 respondentsjoined the study. Logistic regression analysis showed that dropoutwas not significantly predicted by any of the included variables.

Results

Descriptive Statistics

Table 2 presents school characteristics, the average networkcharacteristics, and adolescent characteristics. Over time, the num-ber of nominated friends as well as the smoking behavior ofadolescents increased. At T1, 7.2% of the respondents reportedthat they were weekly smokers. At T2 and T3, respectively 14.7%and 19.0% reported to be weekly smokers.

Friendship Network Changes: Selection of Friends

The results of the first part of the model, examining variables onwhich friendship selection depends, are shown in the upper part ofTable 3 for time period 1 and the upper part of Table 4 for period2. In both successive periods, adolescents had a significant ten-dency to choose friends with similar smoking behavior, as indi-cated by the significant ‘smoking behavior similarity’ effects(RQ1). Alongside this, during period 1, an overall positive linearassociation between an adolescents’ smoking behavior and his orher tendency to be chosen as a friend was found (significantpositive effect ‘smoking behavior alter’), but the marginal increasein attractiveness with smoking level was smaller for heavy smok-ers than for light smokers (significant negative effect ‘smokingbehavior alter squared’). Both effects were not significant in period2. During period 1, none of these results varied significantly acrossthe 11 included schools. During period 2, differences betweenschools were found regarding selection based on similar smokingbehavior (chi-squared � 22.326, d.f. � 10, p � .014, estimatedtrue SD � 0.321).

The results for the included control effects revealed that ado-lescents in both time periods significantly tended not to selectarbitrary friends, but tended to have reciprocal friendships and tobe friends with their friends’ friends, as indicated by the significantnegative outdegree effect, and significant reciprocity and transi-tivity effects. Adolescents did not select friends based on similarage within the year group. However, in period 2, adolescents hadthe tendency to select friends who were older in the year group.Adolescents tended to select friends based on similar gender and

Table 2Descriptive Statistics

School yeargroup characteristics Mean Min. Max.

Number of adolescents 158 80 236Number of tutor groups within school 6 4 9

Average network characteristics T1 T2 T3

Densitya 0.028 0.030 0.030Number of nominations 4.028 4.388 4.434Missing fractionb 0.049 0.000 0.068Reciprocity indexc 0.609 0.597 0.623Transitivity indexd 0.354 0.353 0.344Number of adolescents not connected

to any others 7.000 0.455 3.818

Adolescent characteristics T1 T2 T3

Mean age 12.17 — —Mean % of boys 51.495 — —Mean % parental smoking 45.873 45.873 —Mean SES 3.869 3.895 —Mean smoking behavior 2.011 2.376 2.723% adolescents Never smoked 46.0 41.9 32.2% adolescents Tried once 19.2 20.6 19.6% adolescents Used to smoke 12.4 15.2 13.9% adolescents � 1 a week 6.4 6.1 07.4% adolescents 1 to 6 a week 3.1 5.8 04.8% adolescents � 6 a week 4.1 8.9 14.2% missing 8.9 1.5 7.8

a The proportion of ties relative to total number of possible ties. If adoles-cents share only a few friendship ties with each other, density is low, if theyshare many friendship ties, density is high. b The proportion of missingnetwork data. c The proportion of friendship ties that are mutual. d Theproportion of ties to friends that also have a tie to the adolescent’s otherfriends in the network.

454 MERCKEN, STEGLICH, SINCLAIR, HOLLIDAY, AND MOORE

based on being in the same tutor group in both periods. Selectionbased on similar SES was found only in the second time period.

Smoking Behavior Changes: Influence Processes

The results of the smoking behavior evolution part of the modelare reported in the lower parts of Table 3 and Table 4 for periods1 and 2, respectively. Adolescents had a significant tendency tochange their smoking behavior to that of their friends, as isindicated by the significant ‘smoking behavior friends’ effects inboth periods (RQ2).

The results for the included control effects furthermore showthat, at period one, adolescents had the general tendency not tosmoke, however this effect was not significant. The ‘tendency tosmoke squared component’ was found to be significantly positive,which implies that smoking was self-reinforcing, that is, that anadolescent’s tendency to further increase smoking increased withhis or her smoking level. In the second period adolescents dem-onstrated a significant tendency to smoke as well as a significant

self-reinforcing effect which could be a reflection of the addictivenature of smoking. Regarding the other included variables, resultsindicated that adolescents smoked more when at least one of theirparents smoked and that female adolescents had a higher tendencyto smoke compared with male adolescents. Both effects dimin-ished in the second period.

The Relative Contribution of Selection and InfluenceProcesses to Smoking Behavior Similarities

Figure 1 shows that the mean proportion of similarity attributedto smoking-based friendship selection was 23% in period 1 and33% in period 2. The mean proportion attributed to influence offriendship networks 19% in period 1 and 17% in period 2. Of thetotal proportion explained by both processes during period 1, 45%was attributable to influence of friends, 55% to smoking-basedselection. During period 2, 34% of the total proportion explainedby both processes was attributable to influence of friends, 66% tosmoking-based selection.

Table 3Results Meta Analysis Period 1

Snijders–Baerveldt method Fisher combination test one-sided�

b SE OR �2 p value df

Friendship network evolution

Smoking behavior ego 0.015 0.011 1.015 24.495 0.322 22Smoking behavior alter 0.079 0.022 1.082 54.040 0.000 22Smoking behavior alter squared �0.027 0.010 0.973 50.694 0.000 22Smoking behavior similarity (RQ1) 0.425 0.101 1.530 73.367 0.000 22

Control effects:Outdegree �2.469^ 0.050 0.085 12493.537 0.000 22Reciprocity 1.786^ 0.044 5.966 12505.818 0.000 22Transitive ties 0.496 0.042 1.642 199.048 0.000 22Transitive triplets 0.090 0.012 1.094 98.100 0.000 22Number of actors at distance 2 �0.637 0.026 0.529 5822.437 0.000 223-cycles �0.220 0.031 0.803 103.192 0.000 22Age ego �0.002 0.001 0.998 31.159 0.093 22Age alter 0.000 0.001 1.000 18.166 0.696 22Age similarity 0.022 0.056 1.022 21.562 0.486 22Gender ego 0.028 0.034 1.028 25.340 0.281 22Gender alter �0.027 0.033 0.973 31.351 0.089 22Gender similarity 0.522^ 0.031 1.685 358.983 0.000 22SES ego �0.002 0.009 0.998 19.966 0.585 22SES alter 0.002 0.010 1.002 20.684 0.540 22SES similarity 0.073 0.059 1.076 29.005 0.145 22Tutor group similarity 0.427^ 0.053 1.533 2968.568 0.000 20

Smoking behavior evolution

Smoking behavior friends (RQ2) 0.612 0.140 1.844 57.298 0.000 22Control effects:

Tendency to smoke �0.092 0.102 0.912 32.433 0.070 22Tendency to smoke squared 0.262 0.054 1.300 134.966 0.000 22Age adolescent �0.004 0.006 0.996 32.148 0.075 22Gender adolescent �0.191 0.086 0.826 37.848 0.019 22SES adolescent �0.014 0.032 0.986 24.698 0.312 22Parental smoking behavior 0.508 0.170 1.662 78.833 0.000 22

� Only one-sided p values in the direction of the sign of the coefficients according to the Snijders–Baerveldt method (2003) are reported, tests in the otherdirection all gave nonsignificant results; b � unstandardized coefficients according to the Snijders–Baerveldt method (2003); SE � standard error; df �degrees of freedom for the one-sided tests. Bold values represent significant results. ^ Significant differences found between schools according to theSnijders–Baerveldt method (2003).

455SOCIAL NETWORK ANALYSIS OF SMOKING BEHAVIOR

Alternative influence and selection mechanisms (i.e., selectionbased on gender, age, and the influence of gender, parental smok-ing) are jointly represented by the slices labeled ‘control.’ Theproportion of similarity attributed to these control processes was12% in period 1 and 11% in period 2. The proportion of smokingbehavior similarity explained by the previous behavior and previousnetwork status (trends) was 42% in period 1 and 38% in period 2.

The proportion of smoking behavior similarity explained bysmoking-based selection increased over time, whereas the propor-tion of similarity explained by influence of friends decreased. Alarge proportion of smoking similarity was found to be carriedover from previous observation moments by mechanisms otherthan smoking-based influence and selection. These parts are ex-pressed as ‘trend’ and ‘control,’ which quantify the amount ofsimilarity that could already be explained by the limited amount ofchange in the network (trend) and change mechanisms that corre-late with similarity (control).

Discussion

The main goal of the present study was to examine the role ofsmoking-based selection and influence processes in adolescentsmoking and changes over time in these processes for the first timeusing newly developed longitudinal social network analysis tech-niques.

Our findings demonstrate that selection and influence processesboth played an important role in creating and maintaining smokingbehavior similarity within friendships. In line with previous re-search that identified the importance of selection processes (Co-hen, 1977; De Vries, Candel, Engels, & Mercken, 2006; Ennett &Bauman, 1994; Fisher & Bauman, 1988; Kandel, Kessler, &Margulies, 1978), adolescents tended to select friends with similarsmoking behavior. Support for peer influence within friendships,which is often suggested in the literature (Aloise-Young, Graham,& Hansen, 1994; Kandel et al., 1978; Mercken et al., 2010;Sussman et al., 1990), was also found. The results of these anal-

Table 4Results Meta Analysis Period 2

Snijders–Baerveldt method Fisher combination test one-sided�

b SE OR �2 p value df

Friendship network evolution

Smoking behavior ego �0.002 0.009 0.998 18.642 0.667 22Smoking behavior alter 0.015 0.021 1.015 30.823 0.100 22Smoking behavior alter squared �0.006 0.010 0.994 27.955 0.177 22Smoking behavior similarity (RQ1) 0.605^ 0.119 1.831 148.002 0.000 22

Control effects:Outdegree �2.717 0.059 0.066 15197.062 0.000 22Reciprocity 1.876^ 0.076 6.527 13866.299 0.000 22Transitive ties 0.550 0.040 1.733 242.737 0.000 22Transitive triplets 0.143 0.014 1.154 168.514 0.000 20Number of actors at distance 2 �0.509 0.032 0.601 478.285 0.000 223-cycles �0.241^ 0.056 0.786 123.788 0.000 22Age ego �0.001 0.001 0.999 23.951 0.350 22Age alter 0.004 0.002 1.004 56.775 0.000 22Age similarity 0.057 0.054 1.059 27.721 0.185 22Gender ego �0.040 0.032 0.961 31.560 0.085 22Gender alter 0.024 0.031 1.024 27.045 0.210 22Gender similarity 0.490^ 0.061 1.632 299.134 0.000 22SES ego �0.018 0.009 0.982 33.956 0.050 22SES alter 0.012^ 0.018 1.012 43.128 0.005 22SES similarity 0.124 0.061 1.132 38.711 0.015 22Tutor group similarity 0.264^ 0.061 1.302 194.223 0.000 20

Smoking behavior evolution

Smoking behavior friends (RQ2) 0.797 0.148 2.219 77.338 0.000 22Control effects:

Tendency to smoke 0.273^ 0.268 1.314 54.321 0.000 22Tendency to smoke squared 0.306 0.127 1.358 169.074 0.000 22Age adolescent �0.001 0.004 0.999 17.794 0.718 22Gender adolescent �0.156 0.114 0.856 33.094 0.033 20SES adolescent �0.068 0.040 0.934 36.056 0.030 22Parental smoking behavior 0.144 0.139 1.155 34.262 0.046 22

� Only one-sided p values in the direction of the sign of the coefficients according to the Snijders–Baerveldt method (2003) are reported, tests in the otherdirection all gave nonsignificant results; b � unstandardized coefficients according to the Snijders–Baerveldt method (2003); SE � standard error; df �degrees of freedom for the one-sided tests. Bold values represent significant results. ^ Significant differences found between schools according to theSnijders–Baerveldt method (2003).

456 MERCKEN, STEGLICH, SINCLAIR, HOLLIDAY, AND MOORE

yses differ from the results of a previous study that also includedadolescents within U.K. schools (Mercken, Snijders et al., 2009), inwhich no support was found for influence among friends. However,this previous study assumed selection and influence processes wouldbe stable over time, which could have resulted in a failure to detectinfluence if only present in one of the periods studied.

In the current study, possible changes in selection and influenceprocesses over time were examined in two successive time periods.The findings showed that smoking-based selection as well asinfluence of friends played important roles in both periods. Rela-tive to smoking-based selection, influence of friends slightly de-creased over time. During period 1, 45% of the proportion smok-ing similarity explained by both processes was attributable toinfluence of friends. During period 2, this percentage decreased to34%. This is not entirely in line with findings of two previousstudies arguing that influence of friends was stable (Simons-Morton & Chen, 2006) or increased (Mercken, Candel et al., 2009)over time. However, the decrease in explained similarity was smallin the present study, and in these previous studies other statisticalmethods were used which do not fully account for the dependen-cies caused by the network structure, potentially leading to biasedestimations of influence and selection. Our findings are in line withprevious research which argues that as adolescents grow older,they form a more autonomous sense of self, group affiliationsbecomes less important, and conformity decreases (Newman &Newman, 1976). Our findings furthermore show that the propor-tion of smoking similarity explained by smoking-based selectionof friends increased over time. This again contradicts earlier re-search using conventional statistical analysis techniques that showa decrease in smoking-based selection over time from 7th to 9thgrade (Mercken, Candel et al., 2009; Simons-Morton & Chen,2006). During early adolescence, smoking-based selection offriends possibly first increases before a decrease becomes visible.Table 1 shows that the number of never smokers decreased whilethe number of adolescents that smoked at a high rate (i.e., weekly

smokers) increased. If those adolescents who smoke at a high rateare mostly forming close groups, an increase in smoking-basedselection can become visible. Future research should study this andaim to replicate our findings in other samples using actor-basedmodels for behavior network coevolution to overcome the meth-odological flaws of conventional statistical techniques.

Furthermore, our findings indicated that the magnitude of se-lection effects differed across the 11 school networks in the secondperiod, although the effect was found to be consistently positive.Besides the importance of examining successive time periods ofadolescence one by one, this finding also implies that a multilevelapproach should be considered to investigate the role of schoolcharacteristics such as school ethos and smoking bans on smoking-based selection and influence processes among adolescents.

The present study included several alternative mechanisms tocounter potentially biased estimations of influence and selectionprocesses. In line with previous studies (Burk et al., 2007; Snijders& Baerveldt, 2003), adolescents tended not to choose arbitraryfriends, but there was a strong tendency toward reciprocal friend-ships and triadic closure in both periods. Being of the same genderand being in the same tutor group also determined friendshipselection. Only in the second time period did adolescents have agreater tendency to choose older friends and friends of similarSES. However, we measured adolescents’ SES only with theFamily Affluence Scale (Mullan & Currie, 2000). While the Fam-ily Affluence Scale is recognized as an appropriate self-reportmethod for identifying different levels of affluence (Currie et al.,2008), it is possible that other measurements of SES could resultin different findings. Regarding the alternative influence mecha-nisms, support was found for the feedback effect of current smok-ing behavior in both periods, which could reflect the addictivenature of smoking. In the first period, girls smoked more than boysand in line with previous studies, adolescents were influenced byparental smoking behavior (Avenevoli & Merikangas, 2003; Hov-

Figure 1. Relative contribution of smoking-based selection and influence on similarities in smoking.

457SOCIAL NETWORK ANALYSIS OF SMOKING BEHAVIOR

ing, Reubsaet, & De Vries, 2007; West, Sweeting, & Ecob, 1999).These effects diminished during the second period.

This study is subject to some limitations. First, a direct measureof parental smoking was not available, which might have biasedestimates of parental smoking effects. However, previous researchhas demonstrated that adolescents aged 13–17 years can be reliablesources for assessing their parents’ smoking status (Harakeh, En-gels, De Vries, & Scholte, 2006). Second, only friendships withinthe same school grade were examined. Older friends may have animportant influence on smoking behavior of younger adolescents,and as adolescents grow older, friends are more likely to comefrom various settings. Future social network studies should aim toinclude all friends inside school as well as outside. Third, in thisstudy smoking behavior of adolescents was measured with onlyone question. However, reported smoking behavior was biochem-ically validated among all respondents (smokers and nonsmokers)and showed very low discordance (Campbell et al., 2008). Finally,by conducting longitudinal social network analysis, it was not yetpossible to examine the specific influence of those adolescentswho quit or rejected smoking. Future research should explorepossibilities to model the specific role of adolescents who quitsmoking or smoke at a specific level in smoking-based selectionand influence processes.

Our findings have important implications for practice and futureresearch. Smoking prevention should not solely focus on socialinfluence but should also consider selection processes. Previousresearch has already emphasized that peer network structure needsmore attention within prevention programs besides the promotionof social influence skills (Audrey, Holliday, & Campbell, 2006;Campbell et al., 2008; Dishion & Owen, 2002; Pearson & West,2003; Valente, Hoffman, Ritt-Olson, Lichtman, & Johnson, 2003).Prevention could benefit from creating nonsmoking majoritieswithin groups, working with popular peers as role models (Camp-bell et al., 2008), or increasing self-awareness regarding imitationand selection. Our findings also demonstrated the importance ofexamining change in selection and influence processes over timeand the need for a multilevel approach to study school effects onpeer selection and peer influence.

References

Aboud, F. E., & Mendelson, M. J. (1996). Determinants of friendshipselection and quality: Developmental perspectives. In W. M. Bukowsky,A. F. Newcomb & W. W. Hartup (Eds.), The company they keep.Friendship in childhood and adolescence (pp. 87–112). Cambridge, UK:Cambridge University Press.

Aloise-Young, P. A., Graham, J. W., & Hansen, W. B. (1994). Peerinfluences on smoking initiation during early adolescence: A compari-son of group members and group outsiders. Journal of Applied Psychol-ogy, 79, 281–287. doi:10.1037/0021-9010.79.2.281

Audrey, S., Holliday, J., & Campbell, R. (2006). It’s good to talk: Ado-lescent perspectives of an informal, peer-led intervention to reducesmoking. Social Science & Medicine, 63, 320 –334. doi:10.1016/j.socscimed.2005.12.010

Avenevoli, S., & Merikangas, K. R. (2003). Familial influences on ado-lescent smoking. Addiction, 98, 1–20. doi:10.1046/j.1360-0443.98.s1.2.x

Bauman, K. E., Fisher, L. A., Bryan, E. S., & Chenoweth, R. l. (1984).Antecedents, subjective expected utility, and behavior: A panel study ofadolescent cigarette smoking. Addictive Behaviors, 9, 121–136. doi:10.1016/0306-4603(84)90050-9

Berndt, T. J., & Hoyle, S. G. (1985). Stability and change in childhood andadolescent friendships. Developmental Psychology, 21, 1007–1015. doi:10.1037/0012-1649.21.6.1007

Brown, B. B. (1981). A life-span approach to friendship: Age-relateddimensions of an ageless relationship. Research in the Interweave ofSocial Roles, 2, 23–50.

Burk, W. J., Steglich, C. E. G., & Snijders, T. A. B. (2007). Beyond dyadicinterdependence: Actor-oriented models for co-evolving social networksand individual behaviors. International Journal of Behavioral Develop-ment, 31, 397–404. doi:10.1177/0165025407077762

Campbell, R., Starkey, F., Holliday, J., Audrey, S., Bloor, M., Parry-Langdon, N., . . . Moore, L. (2008). An informal school-based peer-ledintervention for smoking prevention in adolescence (ASSIST): a clusterrandomised trial. The Lancet, 371, 1595–1602. doi:10.1016/S0140-6736(08)60692-3

Chassin, L., Presson, C. C., Rose, J. S., & Sherman, S. J. (1996). Thenatural history of cigarette smoking from adolescence to adulthood:Demographic predictors of continuity and change. Health Psychology,15, 478–484. doi:10.1037/0278-6133.15.6.478

Cliff, A., & Ord, J. K. (1981). Spatial Processes: Models and Applications.London: Pion.

Cochran, W. G. (1954). The combination of estimates from differentexperiments. Biometrics, 10, 101–129. doi:10.2307/3001666

Cohen, J. M. (1977). Sources of peer group homogeneity. Sociology ofEducation, 50, 227–241. doi:10.2307/2112497

Currie, C., Molcho, M., Boyce, W., Holstein, B., Torsheim, T., & Richter,M. (2008). Researching health inequalities in adolescents: The develop-ment of the Health Behaviour in School-Aged Children (HBSC) FamilyAffluence Scale. Social Science & Medicine, 66, 1429–1436. doi:10.1016/j.socscimed.2007.11.024

De Vries, H., Candel, M., Engels, R., & Mercken, L. (2006). Challenges tothe peer Influence Paradigm: Results for 12–13 year olds from sixEuropean countries from the European Smoking Prevention FrameworkApproach study. Tobacco Control, 15, 83– 89. doi:10.1136/tc.2003.007237

Dishion, T. J., & Owen, L. D. (2002). A longitudinal analysis of friendshipand substance use: Bidirectional influence from adolescence to adult-hood. Developmental Psychology, 38, 480 – 491. doi:10.1037/0012-1649.38.4.480

Eiser, J. R., Morgan, M., Gammage, P., Brooks, N., & Kirby, R. (1991).Adolescent health behaviour and similarity-attraction: Friends sharesmoking habits (really), but much else besides. British Journal of SocialPsychology, 30, 339–348. doi:10.1111/j.2044-8309.1991.tb00950.x

Engels, R. C. M. E., Knibbe, R. A., Drop, M. J., & de Haan, Y. T. (1997).Homogeneity of cigarette smoking within peer groups: Influence orselection? Health Education & Behavior, 24, 801–811.

Ennett, S. T., Bauman, K. E., & Koch, G. G. (1994). Variability in cigarettesmoking within and between adolescent friendship cliques. AddictiveBehaviors, 19, 295–305. doi:10.1016/0306-4603(94)90031-0

Ennett, S. T., & Bauman, K. E. (1994). The Contribution of Influence andSelection to Adolescent Peer Group Homogeneity: The Case of Adoles-cent Cigarette Smoking. Journal of Personality and Social Psychology,67, 653–663. doi:10.1037/0022-3514.67.4.653

Fergusson, D. M., Lynskey, M. T., & Horwood, L. J. (1995). The role ofpeer affiliations, social, family and individual factors in continuities incigarette smoking between childhood and adolescence. Addiction, 90,647–659. doi:10.1111/j.1360-0443.1995.tb02203.x

Fisher, L. A., & Bauman, K. E. (1988). Influence and selection in thefriend-adolescent relationship: Findings from studies of adolescentsmoking and drinking. Journal of Applied Social Psychology, 18, 289–314. doi:10.1111/j.1559-1816.1988.tb00018.x

Harakeh, Z., Engels, R., De Vries, H., & Scholte, R. H. J. (2006). Corre-spondence between proxy and self-reports on smoking in a full family

458 MERCKEN, STEGLICH, SINCLAIR, HOLLIDAY, AND MOORE

study. Drug and Alcohol Dependence, 84, 40 – 47. doi:10.1016/j.drugalcdep.2005.11.026

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis.New York, NY: Academic Press.

Hoffman, B. R., Monge, P. R., Chou, C.-P., & Valente, T. W. (2007).Perceived peer influence and peer selection on adolescent smoking.Addictive Behaviors, 32, 1546–1554. doi:10.1016/j.addbeh.2006.11.016

Hoving, C., Reubsaet, A., & De Vries, H. (2007). Predictors of smokingstage transitions for adolescent boys and girls. Preventive Medicine: AnInternational Journal Devoted to Practice and Theory, 44, 485–489.doi:10.1016/j.ypmed.2007.02.011

Kandel, D. B., Kessler, R. C., & Margulies, R. Z. (1978). Antecedents ofadolescent initiation into stages of drug use: A developmental analysis.Journal of Youth and Adolescence, 7, 13–40. doi:10.1007/BF01538684

Kandel, D. B. (1978). Homophily, selection and socialization in adolescentfriendship. American Journal of Sociology, 84, 427–436. doi:10.1086/226792

McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a Feather:Homophily in Social Networks. Annual Review of Sociology, 27, 415–444. doi:10.1146/annurev.soc.27.1.415

Mercken, L., Candel, M., Willems, P., & De Vries, H. (2009). Socialinfluence and selection effects in the context of smoking behavior:Changes during early and mid adolescence. Health Psychology, 28,73–82. doi:10.1037/a0012791

Mercken, L., Snijders, T. A. B., Steglich, C., & De Vries, H. (2009).Dynamics of adolescent friendship networks and smoking behavior:Social network analyses in six European countries. Social Science &Medicine, 69, 1506–1514. doi:10.1016/j.socscimed.2009.08.003

Mercken, L., Snijders, T. A. B., Steglich, C. E. G., Vartiainen, E., & deVries, H. (2010). Dynamics of adolescent friendship networks andsmoking behavior. Social Networks, 32, 72– 81. doi:10.1016/j.socnet.2009.02.005

Mullan, E., & Currie, C. (2000). Socioeconomic inequalities in health. InC. Currie & K. Hurrelmann (Eds.), Health and Health Behaviour amongYoung People (pp. 65–72). Copenhagen: WHO Regional Office forEurope.

Murray, D. M., & Perry, C. L. (1987). The measurement of substance useamong adolescents: When is the ‘bogus pipeline’ method needed? Ad-dictive Behaviors, 12, 225–233. doi:10.1016/0306-4603(87)90032-3

Newman, P. R., & Newman, B. M. (1976). Early adolescence and itsconflict: Group identity vs. alienation. Adolescence, 11, 127–142.

Pearson, M., Steglich, C., & Snijders, T. (2006). Homophily and assimi-lation among sport-active adolescent substance users. Connections, 27,47–63.

Pearson, M., & West, P. (2003). Drifting smoke rings: Social networkanalysis and Markov processes in a longitudinal study of friendshipgroups and risk-taking. Connections, 25, 59–76.

Prokhorov, A. V., Pallonen, U. E., Fava, J. L., Ding, L., & Niaura, R.(1996). Measuring nicotine dependence among high-risk adolescentsmokers. Addictive Behaviors, 21, 117–127. doi:10.1016/0306-4603(96)00048-2

Schaefer, D. R., Light, J. M., Fabes, R. A., Hanish, L. D., & Martin, C. L.(2010). Fundamental principles of network formation among preschoolchildren. Social Networks, 32, 61–71. doi:10.1016/j.socnet.2009.04.003

Simons-Morton, B., & Chen, R. S. (2006). Over time relationships betweenearly adolescent and peer substance use. Addictive Behaviors, 31, 1211–1223. doi:10.1016/j.addbeh.2005.09.006

Snijders, T. A. B., & Baerveldt, C. (2003). A multilevel network study ofthe effects of delinquent behavior on friendship evolution. Journal ofMathematical Sociology, 27, 123–151. doi:10.1080/00222500305892

Snijders, T. A. B., Steglich, C., & Schweinberger, M. (2007). Modeling the

co-evolution of networks and behavior. In K. Van Montfort, H. Oud &A. Satorra (Eds.), Longitudinal models in the behavioral and relatedsciences (pp. 41–71). Mahwah, NJ: Lawrence Erlbaum.

Snijders, T. A. B., Steglich, C. E. G., Schweinberger, M., & Huisman, M.(2007). Manual for SIENA version 3.1. University of Groningen: ICS;University of Oxford: Department of Statistics.

Snijders, T. A. B., van de Bunt, G. G., & Steglich, C. E. G. (2010).Introduction to stochastic actor-based models for network dynamics.Social Networks, 32, 44–60. doi:10.1016/j.socnet.2009.02.004

Snijders, T. A. B. (2001). The statistical evaluation of social networkdynamics. Sociological Methodology, 31, 361–395. doi:10.1111/0081-1750.00099

Stanton, W. R. (1995). DSM-III-R tobacco dependence and quitting duringthe adolescence. Addictive Behaviors, 20, 595–603.

Starkey, F., Moore, L., Campbell, R., Sidaway, M., & Bloor, M. (2005).Rationale, design and conduct of a comprehensive evaluation of aschool-based peer-led anti-smoking intervention in the UK: The ASSISTcluster randomised trial [ISRCTN55572965]. BMC Public Health, 5, 43.

Steglich, C., Snijders, T. A. B., & Pearson, M. (2010). Dynamic networksand behavior: Separating selection from influence. Sociological Meth-odology. doi:10.1111/j.1467-9531.2010.01225.x

Steglich, C., Snijders, T. A. B., & West, P. (2006). Applying SIENA: Anillustrative analysis of the co-evolution of adolescents’ friendship net-works, taste in music, and alcohol consumption. Methodology: Journalof Research Methods for the Behavioral and Social Sciences, 2, 45–56.

Sussman, S., Dent, C. W., Stacy, A. W., Burciaga, C., Raynor, A., Turner,G. E., . . . Flay, B. R. (1990). Peer-group association and adolescenttobacco use. Journal of Abnormal Psychology, 99, 349 –352. doi:10.1037/0021-843X.99.4.349

Urberg, K. A., Luo, Q., Pilgrim, C., & Degirmencioglu, S. M. (2003). A two-stagemodel of peer influence in adolescent substance use: Individual andrelationship-specific differences in susceptibility to influence. Addictive Behav-iors, 28, 1243–1256. doi:10.1016/S0306-4603(02)00256-3

U.S. Department of Health and Human Services. (1994). Preventing to-bacco use among young people: A report of the Surgeon General.Atlanta, GA: US Government Printing Office (Publication No. S/N017-001-00491-0).

Valente, T. W., Hoffman, B. R., Ritt-Olson, A., Lichtman, K., & Johnson, A.(2003). Effects of a social-network method for group assignment strategies onpeer-led tobacco prevention programs in schools. American Journal of PublicHealth, 93, 1837–1843. doi:10.2105/AJPH.93.11.1837

Van de Bunt, G., Van Duijn, M. A. J., & Snijders, T. A. B. (1999).Friendship networks through time: An actor-oriented dynamic statisticalnetwork model. Computational & Mathematical Organization Theory,5, 167–192. doi:10.1023/A:1009683123448

Wardle, J., Robb, K., & Johnson, F. (2002). Assessing socioeconomicstatus in adolescents: The validity of a home affluence scale. Journal ofEpidemiology and Community Health, 56, 595–599. doi:10.1136/jech.56.8.595

Wasserman, S., & Faust, K. (1994). Social network analysis: Method andapplication. Cambridge: University Press.

West, P., Sweeting, H., & Ecob, R. (1999). Family and friends’ influences on theuptake of regular smoking from mid-adolescence to early adulthood. Addiction,94, 1397–1411. doi:10.1046/j.1360-0443.1999.949139711.x

WHO (World Health Organization). (2007). http://www.who.int/tobacco/health_priority/en.

Received August 16, 2010Revision received August 1, 2011

Accepted August 8, 2011 �

459SOCIAL NETWORK ANALYSIS OF SMOKING BEHAVIOR