We preseni an empirical exploration oJpeer network mechanisms that encourage nntisocial behavior in early adolescents. We apply SIENA, u network modeling n~eihodology that uddresses developmentully changing, statistically dependent interperronaljriendships.

Early Adolescent Antisocial Behavior and Peer Rejection: A Dynamic Test of a Developmental Process

John M . Light, Thomas J . Dishion

Evidence supports the hypothesis that adolescer~t peer groups play a signif- icant role in the genesis of youth antisocial behavior (Dishion & Patterson. 2006). A longstanding interest in research focused on individual clilferences in teen exposure to deviant peer groups is the notion that high-risk youth aggregate because of their comrnon rejection within social contexts, such as the school environment (Cairns, Neckerman, & Cairns, 1989; Dodge et al., 2003; Girford-Smith, Dodge, Dishion, & McCord, 2005; Olson, 1992). We refer to this clynamic process as the confluence hypothesis (Dishion, Patter- son, & Griesler, 1994). The Funclamental dy~~ain ics are generaled by two interrelated processes, selection and influeixce:

I. Youtli at risk for antisocial behavior tend to affiliate differei~tially wit11 each other (seleciion).

2. These affiliations then increase the risk for higher frequency and sever- ity of such behavior (influence).

Two sets of findings provided a m ~ j o r impetus to the surge in peer rela- tionship research in the 1980s. First, investigators recognized that peer

N I k I grants DAOlR760-02 and DA13773-05 supported the work on this chapter. We thank Christine Cody [or editorial support.

78 SO<;lAl. lu'El\VORK Avaivsis AND CFIILEKEN'I PIlER RE1 AIIONSIIIPS

rejection and isolation in childhood were associated with poor adult out- comes (Hartup, 1992). Second, studies showed that peer rejection was a reti- able outcome or childhood aggressive behavior (Coie & Kupersmidt, 1983; Dodge, 1983). which furthermore predicted coiltinued aggression several years later (Dodge et al.; 2003). Yet this plethora of work provided no con- vincing case that peer rejection had a causal role in the etiology and coui-se oiproblem bel~avior. lf peel- rejection is causal, what are the mechanisms?

Advarices on this issue occurred by crossing boundaries between two disciplines: criminology and developmental psychology Those studying ado- lescent delinquency recognized such behavior as a "tearn activity" (Gold, 1970). Later, systematic longitudinal studies of iiacionally representative sam- ples revealed deviant peers to be a strong correlate oC adolescent delinq~~ency (Elliott, Huizinga, & Ageton, 1985). Considering both developnlentai and sociological literature Icd to a par-adox: If delinquent youtll face peer rejec- tion and lack social skills, why do they belong to normal-sized peer groups (see Farmer & Ilollowell, 1994; Farmer, Van Acker, Pearl. & Rodkin, 1999) ellfouraging their behavior?

The confluence llypothesis emerged to account for the paradox. The idea is simple: deviant peer groups form arnoilg mutually rejected youth, who adapt by forming a unique group with a deviant set of murually influ- ential norms (Dishion et a]., 1994). In earlier studies consistent with this hypothesis, we found that peer rejection covaried with more extreme mea- sures ol deviant peer iuvolvernenr in early adolescence. Dishion, Nelson, and Yasui (2005) fourld sixth-grade peer rejection combined with academic failure to predict eighth-grade male gang ii~volvernenr, and Rodkin, Farmer, I'earl, and Van Ackcr (2000) Founcl antisocial niales to be most typically (though not invariably) disliked by others.

These studies, while compelling, fell short of a dynamic test of the con- fluence hypothesis. Primaril~t the analyses dicl not consider network enlhed- ding. Various factors besides sociai influence among rejected youth could explain that rejection rates in one year correlate with gang Cormation two years later. For example, neighborhood could afrect both rejection and future gang affiliation. However, analyses by Dishion, Nelson, and Kavanagh (2003) suggested an important principle: each school creates its own peer ecology. Hence, rejection coniluence dynamics inay vary arnoilg schools and need examination within schools. We seek to advance this work by apply- ing a relatively new modeling technique that applies panel-data logic (Kessler & Greenberg, 1981) to peer sociai fcologies (Snijders, Steglich, Schweinberger, &r Fluisman, 2007), which we describe in Inore detail here.

The study o i peer group selection and influence effects has benefited from increased reliance an social network methodology 111 network stud- ies, we might nleasure individual-level peer arfiliations by asking each niem- her of a relatively closed social group (members of one grade at a school) who their friends are. A number of variations on this procedure are possi-

ble (Marsdm, 2005). This yields a complete network, that is, a map show- ing precisely who links to whom. The network approach has two major advantages over other less-specializecl metbodoiogies. 'The first stems from evidence that youth overestimate similarity between themselves and their friends with respect to aritisocial behavior (Bauinan & Fisher, 1986; Prin- steiii & Wang, 2007). This car1 lead to biased estimates of selection arm! influence effects (Ennett & Baurnan, 1994). In network studies, the data point to the friends and the friends' self-reports give a truer assessrnent of behavior. A second advantage is tl ia~ networks properly represent "corn- plete" embedding of individuals-that is, their indirect as well as direct rela- tionships-thus reflectiilg not only an actor's own frier~ds but also the larger context ill which this peer group is located.

However, social network inodeling presents some methodological chal- lenges. The object is to predict changes in an array of interrelated links, which is beyond the scope of standard statistical methods. Moreover, because network-ernbedding measures for individuals usually have their basis in overlapping subgroups (Snijders, 20051, individual-level network data are inherently clepenclent. Ignoring data dependency, even of ostensi- bly rninor magnitude, can produce unpredictable biases in analyses (Kenny & Judd, 1986; Murray & Hannaii, 1990). Multilevel modeling is of limited value for longitudinal analysis of such data, moreover, because dependen- cies change between waves.

Snijders and colleagues developed a statistical modeling methodology, SIENA, for retaining advantages of network methodology while addressing these challenges (Snijders, 2005; Snijders, Stegiich & Schweinberger, 2006) SIENA is an open-source program, available free with documentatioll at http://stat.gamiim.rug.nl/stocnet. We describe it further below.

We applied SIENA to test a set of specific hypotheses derived lrom the social leari~ing-based account of antisocial deviant peer processes outlined above. These hypotheses focused on the parts of aritisocial peer dynamics that lnake the process, by virtue of creating a causal feedback loop.

I . Antisocial behavior predicts peer social marginalization. 2. Marginalized (I-ejected) youth tend to affiliate dispr-oportionately with

each other. 3. Changes in antisocial beliavior are proportional to the level of anti-

social behavior- 01' one's direct peers.

In lhis model, aritisocial behavior and social status enter as both cause and eilect; thus, the model has "endogenous dynamics" (Butner, Amazeen, & Mulvey, 2005); that is, change in system properties over time in the absence of external forces. Together, the above hypotheses posit that early a~liisocial behavior affects peer affiliations and those afliliations subsequeiltly aECect antisocial behavior, further influencirig peer affiliations, and so on.

80 SOCIA1 NET\VOKK ANALYSIS AND CHILDREN'S PEEK RFIKI~IOUSHII?

SIENA Models of Social Ecologies

We may represent lriendship linkages at any time as a social network (Wasser- man & Faust, 1994). To capture change, we imagine a series of snupshots rep- resenting the state of each possible friendship tie at rac11 point. Cor~ceptuai and analytical problems posed by time series of social networl<s are immedi- ately apparent in, for examnple: the sheer 11uvnher of possible ties. in a typical U.S. iniddle school of 300 students, some 90,000 relationships are possible (counting ties in both directions). Modeling their collecti\~e change over time is the rather daunting prirnary objective of lorlgitudinal network analysis.

In contrast to typical statistical modeling, SIENA takes a decision- simulation approach by randomly choosing network members in simulated continuous time. After an actor's selection, the actor may choose to change a network tie: acldirig a new link or deleting an existing one. The probabil- ity of making a specific choice depends on a set of user-speciliecl effects. Many effecrs are possible, as discussed below The decision process rnay fur- ther include simulated choices to change a particular behavior, and hehav- ioral decisions niay be eliect-driven. Each efrect has a parameter associated with it, the size and sign o l which determine the effect's magnitude and direction. Statistically sophisticated readers will recognize the decision sim- ulation as a Markov process; hence, Monte-Carlo (MCMC) procedures can provide parameter estimates. Each run takes pararneter values from the last one, incrementally changing thc~n to fit the data better until further changes to parameter values do not lnaterially impvove model fit. SIENA supports several fit criteria, hut we used Method Of Movnents because of its superior execution speed, Iterative p~.ocediires also provide estimated standard errors (Snijdcrs, 2001, 2005; Snijders et al., 2007).

SlENA allows three types or predictors of link cl~anges. The first is structural, such as out-degrees (unidirectional ties), reciprocity (2-way ties), transitivity (3-way link closure), among others (Wasserman & 17aust, 1994). The second comprises rneasured properties of edges or pairs of individ~rals in the network, such as geographical distance between individuals or mem- bership in one organization. Finally characteristics of individuals themselves may serve as predictors of link change, These diilerent classes of predictors can interact. The result is a model of tie formation arid dissolution probahil- ity. We interpret the rriodel much like an ordinary regression model with main and interaction effects. Significance of individual effects are obtained with 1-ratios.

SIENA also models change in user-specified characteristics (usually a behavior) hypothesized to depend on aspects of an actor's network embed- ding (number of direct or reciprocated linliages, the average behavior of those linkages, and others; Stlijders et al., 2007). This way, the SIENA-based actor decision model reflects a mutual relationship between behavior and social relationships posited by the rejection confluence idea. that is, they affect each other over time.

EARLY ADOI.ESCENT ANTISOCIAI. B1:tlAVIOR AND I'FER RFjFCIION 81

Sample and Methods

Subjects. Subjects were 1,289 public middle-school students from eight schools in orle district in suburban Oregon. The Sinai sarr~ple represents 74 percent recruitment, with 82 percent colnpleting all three waves of assess- ment, conducted during the Sall of sixth, seventh, and eighth grades. The sample was 45 percent male, 55 percent female, primarily European Amer- ican (79 percent), and with a small percentage of Latino students (4.5 per- cent), representative of the study community The final recruitment rate by wave and school varied from 58 to 81 percent, with an unweighted mean of 70 percent. Further proceclural details are available elsewhere (Stormshak, Dishion, Light. & Yasui. 2005).

Measures

Social Nefworhs. We defined the networks based on friendship nomi- nations. Upon request, students named three best Sriends kom a class list. We treated a nominatioil as a direct tie, one that another studei~t may or may not reciprocate.

We performed analyses on direcfecl (asymmetrical) network data, using twenty-four adjacency matrices to represent the peer networks of our respon- dents over time. For each school k = 1, . . . , 8 and each wave I = 1 , 2 , 3 (cor- responding to sixth through eighth grades), nominations were converted to asymmetric n,, (ilui~ll>er of students per class) X n,, matrices, with the (ij)th entry = 1 iE individual i nominated individual j on the relationship criterion, and 0 otherwise. Table 6.1 shows the wave-specific numbers Sor each school.

Rejection (Sociul Stufus). Students also nominated those iildividuals in their grade as those "with whom you would not like to be in a group." We took the total number of such rlominations as a proportion (X LOO, rounded to the nearest integer, a requirement for fitting SIENA models) of the total number in the network (total participants) as a rejection score for each individual.

Antisocial Behcivior The Oregon Healthy Teens Survey (Biglan, Metzler, & Ary, 1994) provided a measure of problem behavior. We used the Anti- social Behavior scale for these analyses (alpha = 0.84), including self-reports of lying, stealing, vandalism, violence, and others. Respondents reply on a 0-5 scale ("never" to "more than 20 times").

Model Testing Protocol. SIENA models normally include standard network structure ellects: outdegree, reciprocity. and soiue form oS uansi- tivity Outdegree is the number of indivicluals the respondent selects as a friend. Reciprocity is the tendency for individuals to extend more ties if they are reciprocatecl (that is, if individual j also selects irtdividual i). Transitiv- ity involves "closure" o l triads iu the network over time; if A links with B and B links with C , then more likely C will eveniually link with A. Several specific network statistics reflect transitivity (see Snijders, 2005). We found that "actors at distance 2" (DL) was the most reliable triad clostire effect (the fewer such structures there are, the more likely it is that an A-8-C linkage

Z Table 6.1. Descriptive Statistics For School Networks 5 .. 2

k More #Less "

g Network Size Netwoi I? Totu! Antisoiia! Bc!zavior ilntisocicil B~havinr Antisocial liehavior' Antisocial Behavior* School Waxre Subj~cts Grndc Perirnlage Mean (SD) ~Mnvirnulii (I- I)-! ( t - l)+t

= -- - .- n

1 1 111 182 61 I i ( .27) 2.0 - .- > 2 119 196 61 .20(.39) 2.3 21 14 5 3 119 190 63 .32(.35) 1.9 51 10

2 1 94 144 62 .14(.35) 2.6 - - K - 2 104 143 71 19(.40) 2.1 17 10 %

i 3 101 I57 64 .34(.44) 2.4 50 39

'Number of individuals increasing or decicasing ilifir antisocial behavior between pairs oi waves.

'Scale range 0-5.

I EARLY Al>OLPSCINS AUSISOCIAL BEHAVIOR AND PEER RFJIICI.IOU 83

chain has "closed" through the addition of an A-C link, or the chain itself has disintegrated; hence D2 is essentially an inverse-transitive structure). In some analyses, network structure effects are substantively important. but here they were simply control covariates, included in all niodels reported below, along with gender.

Tesring Confluence. Using SIENA, we estimated separate models Sor each of the eight schools, based on our experience that school social ecolo- gies may have unique dyr~amics (Stormshak et al., 1999). We examined three confluence-based hypotheses. With three waves ol dam, we estimated lag-one eSEectspooled across the two consecutive two-year periods (grades 6 thvough 7 and grades 7 through 8). l w o of these hypott~eses pertain to friendship forma[ion and dissolution (selection): ( I) antisocial youth tend to lace rejection by their peers, and (2) rejected youth tencl to affiliate with each other. lo test for these statistical patterns, we included two model effects: Antisocial Behavior Alter (AB-Alter) and Rejection Similarity (llej Sim). We hypothesized friendship ties were less likely when the Alter had higher levels of antisocial behavior; tilerelore, we expected this effect to be negative. 111 contrast, we expected the Rej Sirn to be positive; indivicluals are more likely to form ties with others of similai- rejection status. The third hypothesis predicted a positive associatioil between a students antisocial behavior at time t and that of nominateil friends at time t - 1 (influence)

Results

Table 6.1 shows mean antisocial scores by school and wave that increase generally from sixth through eighth grades. The increase from seventh to eighth grade is greater than from sixth to seventh. There are obviously dif- ferences in average antisocial behavior between schools (largest in school 5, smallest in school 7).

Table 6.1 also shows changes in ailtisocial behavior between the two pairs of waves. A subs~antial percentage (about 25 to 50 percent) report behavior changes between waves, with most increasing rather than decreas- ing. Such change is cornrnon in early adolescence. Overall, about 1 to 1.5 percent of ties change between waves, the low percentage reflecting mainly the (typical) sparseness of tiiendship networks. As a network of 150 indi- viduals includes over 20,000 possible one-way linlcages; there are hundrecls of link changes evident in the data across each pair of waves. Our models were designed to account for these changes.

Table 6.2 presents final SIENA models estimated for each of the eight schools. All models included four basic structural effects: outdegree, reci- procity, distance-2, and gender si~nilarity (rate efiects were modeled bur are not presented). For the purpose of this study, we ueated these effects as nui- sance variables, included to ensure confluence-related effects were not con- founded with other well-lznown influeilces on iriendship tie formation. Table 6.2 shows nearly all these efSects to be statistically significant for all

- . z - e h? . Table 6.2. SIENNA Model Results for "3 Best Friend Linkages .

E f i c t s School 1 School 2 School 3 School 4 School 5 School 6 School 7 School 8 g a .

c Tie Formation covariatcs 2 Outdegree -2.4'fO.L) -2.0-(0.1) -2.0*(0.2) -1.6*0.L) -l.Y*(O.l) -2.3'(0.6) -1.7"(0.3) -1.9*(0.3) 5 I<eciprociiy 2.4'(0.2) 2.3'(0.1) 2.4'(0.1) 2.7*(0.l) 2.Yh(0.2) 2.7-(0.5) 2.6*(0.3) 2.2"(0.4) 3 Dislance 2 -0.6"(0.1) -0.6"(0.1) -!.0"(0.1) -1.2*(0.2) -1.4*(0.3) -0.Yf0.5) - 1 . l O . l ) -1.3*(0.4)

Sex sirniidriiy I . l " f 0 3 ) 0.9"fO.l) l.I"(O.3) 0.6*(0.!) 0.6*(0.2) 0.9(0.5) OY"(0.2) 1.1"(0.5) x 3 . Tie fonnarion confucnce 8 4 B altc? 0.06(0.06) 0.05(0.06) -0.13*(0.06) 0.01f0.05) 0.34Fi0.07) 0.0Yf0.09) -0.21(0.20) -0.29"fO.l)

Rejectionh similarily 0.6'(0.3) OH"(0.3) 1.3"(0.3) 0.6f0.7) 1.4"(0.4) 0.8"(0.3) 1 .6f l .0) 0.7i0.6) z AK average alter' ,c l.O(O.6) O.l(O.4) 0.2f0.3) -0.1(0.6) -0.2 (0.5) O.H(O.5) 1.3"(0.5) 0.3f0.7) $ L

rot',: Pai-amctcr cstiiiiates can hi. iiiterprctcd like coefficients in s inultinomial logii regrcssioii, refircling ihr iiiciea,c (+! or decrease (-! in log-odds o i a tie form- ing for ihe link-foriilarion part of the rnodci (aii cilecis but i l ~ r iast) aiiil of a oiicxnit iiicicitsc (decici~sc! in aniisocial hehaiior ioi the lbelia\~ioi p i t oi the model (the AB Average 1?1tei effeci only).

"11 (+!, implies ties arz inorc likciy to f<,mi wiili inore siitisocial alters: if i--!. ihcy are less likeiy 10 [orin. blf (+). ties are niore likely io form heiwcen individuals who have a siruilar levcl of rcjccrion; if (-), similar rejeciiori iinplies lies are less iikely to forin.

CLi (+), an indiniiiiial :ends io increase his or her level oiantisocial hchavior at time t i 1 the more antisocial his direct-connect alters were at time t: i f (-!, the ten- dency is to decrease such behavior tiie inore antisocial one's direct coiiiieci airers.

schools. Outdegree was typically negative and slightly smaller in ahsolure vaine than reciprocity; together these parameters suggest that reciprocated (but not asymmetric) ties are likely to form at a rate sliglltly highe* than chance. The negative distance-2 elfects inclicate a tendency toc\~arcls "net- work closure" lor ties to Link back on themselves to Corm cycles. Final13 the well-known preference for same-gender Frieilds a n l o ~ ~ g early adolescents is reilected in the gcnei-ally significant "sex similarity" efrects; only one school was found to have a nonsignificant gender effect. and in that scliool i t was trend-significant (12 <.lo).

The last three rows of-kble 6.2 address confluence hypotheses 1, 2, and 3. Hypothesis 1 stateil that antisocial youth should be socially marginalized, which we tested by including a11 effect of AB-alter in each model. This pararri- cter should he significant and negative if hypothesis 1 is correct. In schools 3 and 8, we see this conlirmed. but erfects are nonsignificant in 1, 2 ,4 , 5, and 7. 111 school 5, the effect is significant but positive: ties are more likely to extend to the antisocial youth. This is clearly the opposite of hypothesis 1 but consistent with other findings related to ecological variation in how antisocial youth affiliate with peers (lor example, Stormshak et al., 1999).

Hypothesis 2 implies that rcjectfd youth (often nolnirlated as disliked) tend to affiliate disproportionately with each other. A significant reiectioil confluence effect appears in five schools, with positive trends in the other three, providing compelling support for rejection confluence in that stu- dents hi-m into subgroups based on similar experiences with respect to mar- ginalization within their schools.

Finally, hypothesis 3 states that yonth who affiliate with rnore AB-alters will tend to have higher antisocial behavior scores in the next wave. How- ever, the effect appeared in only school 7.

Discussion

The confluence rnodel proposes that rejection by peers iniluences peer net- works by rncans of high risk youth aggregating into cliques, which then later influences the development o l antisocial behavior, especially in ado- lescence. This simple hypothesis has been dillicult to test comprehensively, largely because of the complex methodological issues inherent in analyzing changing networks over time.

The agent-oriented SIENA modelirlg approach, however, provided a nat- ural framework for such a test. lil our analyses, we exainined each school as its own ecology We found universal support for the hypothesis that rejected children tend to affiliate over time. Antisocial youth, howevel., were riot always peer re~ected. Consistent with the lindings of others, in some schools, anti- social behavior was associated with acceptance and even popularity (for example, Kodkin ct al., 2000; Rodkin, Farmer, Pearl, &I Van Acker, 2006) Finally, we found evidence that antisocial affiliation predicted subsequent growth in antisocial behavior-I hat is. influence-in only one middle school.

Clearly it was useful to analyze each school separately, as it seems that each school presents with unique network dynamics. From a substantive perspective, it is evident that the confluence hypothesis can only account for peer-level antisocial homopliily and the growth of a~ltisocial behavior over time in one o l these eight schools. What is to be niade of these resillts?

We suggest three potentially productive directioils for iuture research. First, of course it may turn orit that the endogenous growth of arltisocial behavior is relatively iare, occurriilg only under certain special conditions. If true, this would riot be uninteresting-hundreds or thousands of middle schools nationwide may be unwittingly structured to allow this type of parhological process. Therebre, one fruitful approach is to study antisocial peer processes within and across specific school ecologies. Multilevel designs can potentially disenvangle school-level effects (sleighborhood, coni- position, adminisrrative procedures) vcrsus cohort composition effects, either or both of which might account for this variation. For example, we might find that schools with low overall levels of peer rejection confluence also exhibit less peer influence on antisocial behavior. Perhaps in ttiose ecologies, the only sources of grow111 in such behavior are exogenous.

Second, given ancillary evidence o l the ubiquity of social influence among adolescents, i t is altogetl~er possible that to detecr such effecu in nai- ural settings, much betvr data are requirecl. Frequent assessments of peer relationships and behaviors are required to acciirately observe which changes seem to be causes (having occurred earlier in time) and which are effects (having occurred later). As the interval between observations increases, more of these events will appear to be "ties," that is; both occurring in the same time period. Ties arc useless as evidence for rs~hether affiliation caused behav- ior (influence) or behavior caused affiliation (selection). The yearly assess- ment strategy used in this study may he i~~adequate to the task. The same point may apply to other network-based studies that found weak or absent influence effects (see Lngels, Vitaro, Dell Exter Blokland, de Kemp, 61 Scholt, 2004; Ennett & Banman, 1994; Jaccard, Blailton, & Dodge, 2005).

Tliird, we suggest that social ~let~vorking researcli be linked at some point wirh direct observation research. We know froin direct observatioll stuclies that deviancy training as observed with a "friend" at age thirteen pre- dicts growth in multiple lorms of problem behavior well into adulthood (for example, Dishion, Nelson, Winter & Bullock, 2004). Olten youth identified as friends are not part of the school environment or even especially close. Rather, rbey are rrrore iiearly "regular associates," peel-s with whoin the child spends a great deal of time. Thus, some peer environments may promote influence, but not ~lecessarily from friendships as defined in previous stud- ies. Adapting the types o l observatio~lal methods used with smaller groups of high-risk children (for example, Dishion, Poulin, 61 Burraston, 2001) to larger peer social ecologies could help identify some of these mechanisms ant1 the contexts in which they arise.

EARI,Y ADOLLSCI~N~ ANI.ISOCIAL BEIIAVIOK AND PEI:II REJECTION 87

Nevertheless, even il only some school social ecologies unwirtingly promote the growth of antisocial behavior or other behavior problems, it is important to continiir to develop our res~arch methods to help engineer less contagion-prone environn~ents. Every teacher and principal knows that cohorts o l students can vary dralr~atically in their levels of problem behavior even within one school. Giving adults better tools to mailage peer environments would help students with learning and character develop- ment ancl scllool stall with teaching environment and stress. Festinger. Schachter, and Back (1950) foui~d that relationships between i~ldividuals are less likely when the opportunity for chance encounters is lower; sys- ternatically separating at-risk youth (contrary to mr1c11 current pracrice) or lessening their opportunities for unsupervised interaction provide possi- ble examples. Nevertheless, we must be able to reliably predict the absence and the presence of key dynamics before this strategy can be reliably implc- mented, underscoring thc ci-itical importance of Inore powerl'ul research designs.

References

liauriian, K. E., & Fisher, I.. A. (1986). Ori thc measrlrenient o l liierid behavior in research on friend influe~nce and selection: IFiniliiigs SI-om Iongitiiclinal stnilies of ado- lescenl smokiiig and drinking. journal ,$Yoicth and Adolrsceiicc, 15, 345-53.

liiglan, 4.. Metzlcr, C. W., 6r Ary. T j . V. (1994). Increasing thc prevalence of successiul children: 'The case lor co,iimunity inierveniion researcii. l3ehavior Analyst, 17,335-51

Iiuirii.r, J. , Aii~aieen, P. C;., 6z Mulvey. G. M. (2005). Multileuel n~odelirtg of two cycli- cal processes: Extending dil'ferential structural equation modeling to ~ioiilincar coil- pled systems. Psychological Mf,rhods, 10(2), 159-1 77.

Cairns, K. R.. Nrckerman, H. J., & Cairns, B. D. (1989). Social networks and tile shad- ows of synclirony. In G . P. Adams, R. Moiitemayor, & f . P. Gulloua (Eds.), Biology ofadolcicenl behavior uiid developrneni (p1). 275-301). ?housanil Oaks. CA: Sage.

Coic, j. D., & Kupersrnidt, J . B. (1983). 4 behavioral analysis of erncrging social status in boys groups. Ciiild Development, 54, 1400-1416.

Dishion, '1'. I . , Kcison. 5 , E.; & Kavanagh. K. (2003). The family check-up with high- risk young adolescents: I'reuenting early-onset subsiaricc use by parent lnoi~itoriiig. Behavior Therctpy, 34, 553-571.

Dishioil, T. J. , Nelsori, 5. F., 'd'i~lter, C., & Bullock, B. M. (2004). A~l~lescerit friendship as a dynamic system: Entropy and deviance in the etiology arid course ol ai~tisocial hehavior,Joirrnnl ofAbnorrnal Child i'sycholo'v, 32, 651-663.

Dishion, I, J . , Nelson, 5. E., & Yasui. M. (2005). Predicting early adolrscent gaiig involvement lrorn middle school adaptatioii. Journal ($ Clinical Child and Adol~scvnr Psycholojg, 34(1), 62-73.

Dishion, '1'. J . , & Patterson, G . R. (2006). Thc developiiient and ccology of aritisiicial behavior, 111 D. Cicchetii & D. 1. Cotlet1 (Ecls.), ljel,eloprnental psychopitthology Val. .3: Risk, disuriler, und a~laplatioi~ (pp. 503-541). Hohoken, SJ: Wiley.

ijistiion, T. J.. Patlerson. G. R., 61 Cricslcr, P. C. (1994). Peer adaptations in the devcl- opmeiit of antisocial behavior: A coriniiencc model. In L. R. Huesinann (Ed.), Agqres- sivc bcha~,ior: Currimt perspectives (lip, 61-95). New York: Plcnum.

Dishion, T. J. , Poulin, F., 6z Burraston, 13. (2001). Peer group dynamics associated with iatrogenic erSecs in grilup interventions with high-risk yoiitli (pp. 79-92). In C. Erdly

Nrn I>mr<rtois ioli Ciiil u Avo baol i:SciNI 13cVil OPblEu 1 . 1>01 10 ii>O2cd

88 SOClAi, NtTWORl< Arn,41.~Sis , iN t i CHILDREN'S PiliK RtiI.alIONs1itPs

& D. W. Nangle (Eds.), ljamon's new directions in cilild dcveli,pment: Thi. role i,i/rienrl- ship in lisychologicnl adjusiment. Sail Francisco: lussey-Bass.

Dodge. K. A. (1983). Behavioral antecedents: A peer social status. Child ljevclopm~~nt, 54, 1386-1399.

Dodge, K. A, Lansiord, J . E.. Buiks. V. S., Batcs.J. E., Pcttit, G. S., Fontaine, R., et al. (2003). Pecr rejection and social infiirmation-processing factors in tile development o ? aggressive behavior problems in children. Child Development, 74(2), 374--393.

Elliott, I)., iiuiririga, U. , & Ageion, 5. S. (1985). Explaining de1inqucnc)fl and drug usc. Thousand Oaks. CA: S a p .

Encels. 1iC.M.. Vitani. F.. Den Exler Blokland, E.. de Keinn. R., & Scholt, R.Il.1. (2004).

-. Ennett, S. T., h- iiauman, K. E. (1994). ' fhr contribution of influence and selection to

adiilesceni peer group l~omogeneity: The case o i adolescent cigarette smoking. Jo~ir- nai ~Ji'rrionality and Sociul Psycholog)', 67(4), 652-663.

Farmer, T. W., & Hollou~ell, J . H. (1 994). Social networks in ti~ainstreain classrooins: sock ,I 1. ail~liaiio~rs "' '

. and behavioral characteristics of students wit11 FBD. Joinnai ofEmo--

tionul arid Behavioral Disorders. 2, 143-156. Farmer, i. W., Van Acker. R. M.. Pcavl, R., & Roiikin, P. C. (1999). Siicial networki and

peer-assessed problem behavior in clemfntary classrooins Remcdial and Special Frlu- cnrion, 20(4). 244-256.

Fesliingcr. L., Sclrachter, 5.. &Back, K. (1950). Social prcssurvs in iidonrial groups: A study oJhumnnJaciors in housing. New Yorli: HarperCollins.

Gifford-Smith, M., Dodge, K. A . Dishion, T. J . , & McCorti, J . (2005). Pcrr inijttence it?

children and adolescents: Crossing the bridgeJrom developmenial to intervention science. Journnl ~Abnorrnril Child Psychology, 33(3), 255-265.

Gold, M. (1970). i)elinq!iunt bchavior in an Amcrican city. San Francisco: Brooks & Colcman.

Wartup, W. W. (1992). Fiicnilships and their dcvelopn~cntal signiiicancc. In H. McGurl< (lid.), Childhood soria! development: Conteiiipor~1r.y perspective.i (pp. 175-205) Mah- uraii, NJ: Erlhanm.

Jaccard,J., Blanton, H., & Dodge, 1'. (2005). I'eer ir~iluences on risk hchavior: An arraly- sis of the effccts of a close lriend. Developniental Ps.ychology, 41(1), 135-147.

Kenny, D. A , & Judd. C. M. (1986). Consequences of violating the independence assumptioii in analysis of variance. Psychoii>gical Bulleiin, 99; 422-431

IZessler, It. C., & Greenberg, D. T. (1981). Linparpancl unalysi3: Modeis ojyuantitatiiv changu. Orlando. FL: Acadeiiiic Press.

Marsdcn, P. V. (2005). Recent developments in network measurement. In 1'J. Cairing- toii , j . Scoti, & S. Wasscvman (Eiis.), Models and methods in social nciwork nnul~~sis (pp. 8-30). Cambridge: Carribridge University Press.

Murray, i). M., & Hanoari, I'.J. (1990). PLannirig for the appropriate analysis in school- based drug use prevention studies, Journal of Consulting and Clinical I'sychoiogy, 58(4), 458-468.

Oison. S. L. (1992). tlevelopmcnt of conduct problems and peer icjcction iil pre-school childreil: A social systems analysis. Journal ojAbnormal Child Psychology, 20, 327-350.

Prinstein, M. J . , & LVang, S. 5. (2007). False consensus arid aiiolesceiit peer contagion: Examining discrepancies between percrptions and actual reported levels o? friends' deviant anti hcalth risk behaviors. Jo~irlial of Ahnortnai Child Psychology, 33(3), 29 3-306.

Rodkin. P. C., Farmer, T. W., Pearl, R., & Van Acker, R. M. (2000). Ileterogcneityof pop- ular boys: A~ltisocial and prosocial corrfigurations. Dvvelopmentcil Psyrhoiog)', 16, 14-24.

Iiodkin, P. C., Farmer, T. \V, Pearl, R., & V a i ~ Ackrr, R. M. (2006). I'lrey're cool: Social status and peer group supports for aggressive boys and girls. Social Devulopment, JS(2). 175-204.

I\, w liilil CIli'N, r0'( CHILD 4YD A"OLli<l\l O i V I l O i ' h r i N l . i>,ii 10 1002,cd

E.4RI.Y ,iDOLFSCI:Ni ANTISOCI~II. BEIHAVIOR AND PEEK RIJECTION 89

Snijdcrs, T.A.B. (2001). 'The statistical evaluation of social network dynarnics. In M. E. Sobel & M. 1'. liecker (Eds.), Sociological rncthodulogy, Boston: Basil Blacliu-ell.

Siiijders, T.A.B. (2005). Models ior lo~~giiuilinal network data. In P. J . Carritlgton, J . Scott, & 5. Wasserman (Eds.), Modcis and mcthods in social network aiialysis (pp. 215-247). Cambridge: Catnbriilgc University Press.

Sniiclers. T.A.B.. Sleelich. C.E.C., hr Scl~weinbcipei. M. (2006). Modeling the co- I , I . : I I i l , . : .\ ,.'.I I . i I \ I I x . \ ' l t . ..

. . . . , / . I * I ' . , . , ., ; \. I . I . \ I I 8 . . . u

Snijders, T.A.B.. Steglicli, C.E.G., Schweinberger, M.. & Huisn~an; M . (2007). Manual for SIENfI version 3. Gronii~pen, The Nelhcrlands: University of Gioningen, ICS, Dcpartmeni o i Siicioiogy.

Stormsliak, E. A,, Dishion, T. J . ; l.iglit, J. , & Yasui, M. (2005). liliplementing ii~mily- centered intcrvcntions within the public iniddie school: Linking srivice delivery to clvange in prohleni behavior. journal of Abi~orrnal Child Psycl~oiogy, 33, 723-33.

Stormshak, E. A,, Biennan, K. L., Bruschi, C., Dodge, K. A , Coie, J., & the Conduct Problems Resfarch Group (1999). The reiation bctween behavior prohlems aiid peer preferelice iii difkrent classroom contexts. Child Uevrloprnt~nl, 70(1), 169-182.

Wassernrail, S., & Taiist. K. (1994). Social nctivork aniil~~sis: Methods anrl applications. Cambridge: Cambridge University Press

joii~v M. Lic i r r is associate scieniisi ai ihe Oregon Research Institute

I ' i - i o ~ ~ s J. D~SFI~ON is professor of psychology and school psychology and the dit-ector ojthe Child and Fatnily Center a1 ihr University of Oregon.

Ni*. I>IIIICIIONI ioll Cill3.D .Ah" All01 i \ r i X T DEVLLOPIZYT . DO1 10 1002/cil