Differences in the Relations Between Antisocial Behavior and Peer AcceptanceAcross Contexts and Across Adolescence

Jeff KiesnerUniversita di Padova

Massimiliano PastoreUniversita di Cagliari

This study tests the hypothesis that, during adolescence, antisocial behavior becomes positively associated withpeer acceptance. This hypothesis was tested considering both classroom and out-of-class peer relations. Datafrom a previously published study, with a cross-sectional sample of 577 Italian 11- to 13-year-olds, were used.Analyses showed that in the 6th grade antisocial behavior was negatively related to classroom peer preference,but not significantly related to out-of-class peer inclusion. By the 8th grade, antisocial behavior was positivelyrelated to out-of-class peer inclusion, but not significantly related to classroom peer preference. Similar resultswere found for males and females. The higher level of peer acceptance among the 8th grade antisocial indi-viduals was primarily due to nominations received by other antisocial individuals.

Moffitt’s (1993) theory of adolescent-limited and life-course-persistent antisocial behavior predicts that,during adolescence, antisocial behavior comes to beviewed as desirable because it represents adultstatus and access to adult opportunities. If this istrue, it could be argued that antisocial behavior willnot consistently be related to low peer acceptancethroughout adolescence. Instead, the relation be-tween antisocial behavior and peer acceptanceshould become positive as antisocial behavior comesto be viewed more positively by peers. Althoughpast research has shown that antisocial and aggres-sive youth tend to be rejected by their peers(Brendgen, Vitaro, Turgeon, & Poulin, 2002; Kiesner,2002; Kiesner, Cadinu, Poulin, & Bucci, 2002; New-comb, Bukowski, & Pattee, 1993), little is knownabout the age-related differences in these relationsthat are predicted by Moffitt’s theory. This study wasconducted to test for changes in the relation betweenantisocial behavior and peer acceptance, consideringpeer relations across two different contexts, with asample of Italian middle school students.

Changes in the Relation Between Peer Acceptance andAntisocial Behavior

Few studies have tested for age-related differ-ences in the relation between antisocial behavior and

peer acceptance, and the few studies that have testedfor such differences primarily have focused on ag-gressive behaviors. For example, Haselager, Cilless-en, Van Lieshout, Riksen-Walraven, and Hartup(2002) followed 274 children 6–11 years old, andfound that the relation between aggressive behaviorand peer rejection remained nearly unchangedacross this age range (r5 .50 at 6 years, and r5 .44 at11 years). Similarly, in a cross-sectional study, La-Fontana and Cillessen (2002) found stable relationsbetween aggressive behavior and peer acceptancefrom the fourth grade to the eighth grade: Across allage groups (except the sixth grade) physical ag-gression was moderately and negatively associatedwith social preference.

Whereas the above studies suggest that the rela-tion between aggressive behavior and peer accept-ance is stable (at least through the eighth grade),other research has come to a different conclusion. Forexample, Cillessen and Mayeux (2004) found that,from the fifth grade to the ninth grade, the relationbetween physical aggression and social preferencedropped from b5 ! .280 to ! .068, controlling forgender, perceived popularity, and relational aggres-sion. Importantly, although the negative relation be-tween physical aggression and social preferencedecreased over time, there was no evidence that ag-gressive youth actually became liked by their peers.

Bukowski, Sippola, and Newcomb (2000), on theother hand, found that the transition from elemen-tary school to middle school was associated with anincrease in attraction to aggressive peers. This find-ing was true for girls’ attraction to aggressive boys,

r 2005 by the Society for Research in Child Development, Inc.All rights reserved. 0009-3920/2005/7606-0011

Data used for this study come from a larger project from whichdata have previously been published.Correspondence concerning this article should be addressed to

Jeff Kiesner, Dipartimento di Psicologia DPSS, Universita di Pad-ova, via Venezia 8, 35131 Padova, Italy. Electronic mail may be sentto jeff.kiesner@unipd.it.

Child Development, November/December 2005, Volume 76, Number 6, Pages 1278 – 1293

and for boys’ attraction to aggressive boys and girls.These authors concluded that, consistent with Mof-fitt’s (1993) theory, during early adolescence ag-gressive behavior becomes associated with higherlevels of peer acceptance rather than rejection.

Although the studies discussed above do notconsistently lead to the same conclusion, the studiesby Cillessen and Mayeux (2004) and Bukowski et al.(2000) provide a basis for expecting that antisocialyouth, or at least aggressive youth, achieve higherlevels of peer acceptance during adolescence thanthey had during childhood. However, further workis needed before clear conclusions can be drawn. Indoing so, there remain two important issues thatmust be addressed.

The first issue regards the context in which peerrelations are studied. In all of the studies discussedabove, peer relations were examined within theschool context. It is important to recognize, however,that peer acceptance may differ across contexts at thesame developmental stage. For example, previousstudies have shown generally weak relations be-tween classroom peer acceptance and out-of-schoolself-reported peer acceptance (Ladd, 1983; Ray, Co-hen, & Secrist, 1995). However, these relations ap-pear to be somewhat stronger (r " .45) when usingpeer reports of classroom peer acceptance and out-of-school peer acceptance (e.g., sports clubs, churchgroups; Durrant & Henggeler, 1986).

In a recent study, Kiesner, Poulin, and Nicotra(2003) examined additive peer homophily acrosspeer contexts, using a sample of sixth to eighth grad-ers (the same sample that is used in this study).Although not central to their research questions,Kiesner et al. (2003) observed that classroom prob-lem behavior was negatively correlated with class-room social preference (r5 ! .26) but unrelated toout-of-school peer inclusion, whereas delinquencyshowed a weak positive correlation with out-of-school peer inclusion (r5 .12) but no relation withclassroom social preference. These differences in therelation between antisocial behavior and peer ac-ceptance across contexts likely depend on a numberof variables. For example, the type of behavior maybe important: Substance use and stealing may bemore attractive to adolescent peers than disruptiveclassroom behaviors. Another variable is the degreeto which individuals are able to select and avoid peercontacts. For example, school personnel typicallydecide classroommembership, and thus the studentshave no choice with whom they spend their schooldays. Therefore, weaker relations between antisocialbehavior and peer rejection could be expected out-side of the classroom, where youth are more free to

select peer affiliates and ignore and avoid youth whoare antisocial. This could be expected at least duringpre- and early adolescence.

During the course of adolescence, however, therelation between antisocial behavior and peer ac-ceptance should change. As Moffitt’s (1993) theorysuggests, during adolescence antisocial behaviorshould be viewed as a sign of maturity, and thus pos-sibly related to higher levels of peer acceptance.Moreover, we could expect that this change would bestronger outside of the classroom, for three reasons.First, outside of the classroom, the noxious and aver-sive aspects of antisocial behavior may be more easilyavoided by peers, and thus will be less likely to resultin negative effects on peer acceptance by those peerswho still do not approve of such behavioral tenden-cies. Second, nonclassroom settings allow one todemonstrate antisocial behaviors that are more sym-bolic of maturity (smoking, drinking, stealing), andthat are likely to be more social in nature (done withpeers). Third, during after-school hours, adolescentsgain increasing freedom from their parents, provid-ing opportunities to associate with individuals whowould not be approved of by parents or other adults.This study will specifically test for contextual differ-ences in the relations between antisocial behaviorand peer acceptance, and how these relations changeacross early adolescence.

A second issue regards the type of antisocial be-havior considered. The studies by Cillessen andMayeux (2004) and Bukowski et al. (2000) examinedthe relation between peer acceptance and aggressivebehavior. However, Moffitt’s (1993) theory focuseson a general construct of antisocial behavior, notspecifically aggressive behaviors. It could be hy-pothesized that a general antisocial lifestyle, ratherthan aggressive behaviors specifically, would bemore strongly linked to increases in peer acceptance.For example, French and Conrad (2001) found thenegative relation between a general measure of an-tisocial behavior and peer preference to be stablefrom the eighth grade (r5 ! .26) to the tenth grade(r5 ! .39). However, examining the items that con-tributed to the antisocial behavior scale suggests thatthis measure primarily addressed aggressive be-haviors and interpersonal conflict. In a study using alongitudinal design, Maggs, Almeida, and Galambos(1995) found that the correlation between peer ac-ceptance and a general construct of antisocial be-havior increased from r5 .05 at 11.6 years to r5 .24 at14 years. Although these researchers examined ageneral construct of antisocial behavior, and the re-sults are consistent with predictions made in thepresent study, the measure of peer acceptance was a

Antisocial Behavior and Peer Acceptance 1279

self-report of how well the individual felt acceptedby his or her peers. Therefore, these results may notreflect an actual increase in peer acceptance.

In this study we test for age-related differences inthe relation between antisocial behavior and peeracceptance considering a general construct of anti-social behavior. To do this we use multiple inform-ants (self-report, peer report, and teacher report) andstructural equation modeling (SEM) to create a latentconstruct of antisocial behavior.

By Whom is Delinquency Accepted?

If antisocial adolescents are more positivelyaccepted by their peers during middle adolescence,as compared with early adolescence, we must ask bywhom they are becoming more accepted. Accordingto Moffitt’s (1993) theory, during adolescence anti-social youth are admired and imitated by averageadolescents. Thus, it could be hypothesized thatnon-antisocial, or at least average, adolescents wouldbegin to have more positive relations with anti-social youth as they move from early to mid-ado-lescence. Therefore, in this study, we expect that anincreased acceptance of antisocial youth would beattributable, specifically, to an increase in acceptanceby average peers.

Alternatively, Dishion’s peer confluence model(Dishion, French, & Patterson, 1995; Dishion, Pat-terson, & Griesler, 1994) postulates that delinquentpeers form homogeneous groups who reinforce an-tisocial behavior among themselves. Thus, accordingto Dishion’s confluence model, we should expectthat increases in peer acceptance of antisocial youthwould be attributable to increases in affiliation amongantisocial youth, rather than increased acceptance byaverage or non-antisocial youth. Consistent with thismodel, a great deal of research has demonstrated peerhomophily when considering a wide range of anti-social behaviors (Cairns, Cairns, Neckerman, Gest, &Gariepy, 1988; Espelage, Holt, & Henkel, 2003;Kiesner et al., 2002, 2003; Poulin et al., 1997; Urberg,Degirmencioglu, & Pilgrim, 1997).

On the other hand, there also appears to be a highdegree of mixing of aggressive and nonaggressiveyouth, even as early as elementary school (Farmeret al., 2002). If Moffitt’s theory is correct, then such‘‘mixing’’ could become more common during ado-lescence as ‘‘average’’ youth become more attractedto antisocial youth. Therefore, in this study we willtest whether increases in peer acceptance of antiso-cial youth are attributable to increased affiliationsamong antisocial youth or to increased acceptance byaverage youth.

This Study

In this study we examine age-related differencesin the relations between antisocial behavior and in-and out-of-classroom peer acceptance during earlyadolescence, with a sample of Italian middle schoolstudents. Peer acceptance is measured within theclassroom, at the schoolwide level, and outside of theschool at the neighborhood level. It is expected thatthe relation between antisocial behavior and class-room peer acceptance will be negative in the sixthgrade and will diminish in strength throughout theseventh and eighth grades. On the other hand, it isexpected that antisocial behavior will become in-creasingly and positively related to peer acceptanceoutside of the classroom throughout the middleschool years. Moreover, we will test whether suchage-related differences are attributable to an increasein acceptance by average peers or to an increase inassociations among antisocial youth.

This study comes from a relatively large-samplecross-sectional study conducted in Milan, Italy. Datafrom this sample have been previously published(Kiesner et al., 2003; Kiesner, Nicotra, & Notari, 2005),and although most of the measures used in this studywere also included in one of those previous publi-cations (Kiesner et al., 2003), the research questionsaddressed in these two studies are very different.Specifically, whereas the earlier paper examinedindividual-group homophily across contexts, and thelink between peer acceptance across contextsand depressive symptoms, this study tests whetherthe relations between antisocial behavior and peeracceptance across contexts differ across age groups.

Gender Differences

Antisocial behavior may be judged differentlywhen performed by males and females, and thusdifferently related to peer acceptance. For example,Cillessen and Mayeux (2004) found that the negativerelation between physical aggression and socialpreference was stronger for girls than for boys in thesixth grade. Therefore, in this study we will comparemales and females with regard to age differences inthe relation between antisocial behavior and peeracceptance across contexts.

Considerations Regarding an Italian Sample ofAdolescents

Previous research has shown both cross-nationaldifferences and similarities between Italian andNorth American youth regarding family and peer

1280 Kiesner and Pastore

relations (Attili, Vermigli, & Schneider, 1997; Ca-siglia, Lo Coco, & Zappulla, 1998; Claes, Lacourse,Bouchard, & Luckow, 2001; Eslea et al., 2004; Kiesneret al., 2002; Schneider, Fonzi, Tani, & Tomada, 1997;Tomada & Schneider, 1997). Considering the existingliterature, we have concluded that the overall patternof relations between individual behavior, peer rela-tions, and social adjustment appears to be similaracross Italian and North American cultural contexts.

There is, however, one important aspect of theItalian educational system that is worth noting: Ital-ian middle school students remain with the sameclass of peers for all classes across all three years ofmiddle school (the same class of 25 students spendall day together for 3 years of middle school). As aresult, it may be more difficult in Italian schools tochange one’s level of peer acceptance from one yearto the next.

Finally, it should also be noted that in Italianmiddle schools: (1) students are not tracked into spe-cific types of schools based on tests or past perform-ance; (2) students typically attend the neighborhoodmiddle school, although they are able to attend anyother middle school if space is available; (3) studentsreturn home around 2:00 p.m.; and (4) Italian middleschools typically do not offer extracurricular activi-ties such as sports activities.

Methods

Much of the material in the following section has beenpresented in an earlier publication (Kiesner et al.,2003). However, as a service to the reader, we havealso included this information here. More detailedinformation regarding neighborhood, sample char-acteristics, and procedure can be found in the earlierpublication using this data set (Kiesner et al., 2003).

Participants

This study was conducted in a neighborhood ofMilan, Italy. All three middle schools serving thatneighborhood agreed to participate. All students(sixth to eighth grades) from all three middle schoolswere asked to participate. A total of 798 studentswere enrolled in these schools, of which 30 did notregularly attend school. Therefore, the total possiblesample was 768. Parental permission to participatewas obtained for 593 students, of whom 577 (75% ofthe total possible sample) actually participated. Thepercentage of students who participated, within eachof the 40 classrooms, ranged from 18% to 100%, with80% of the classes having a participation rate of 60%

or higher, and only one class with a participation ratebelow 40%. Of the 577 participants (288 girls, 289boys), 215 were sixth graders, 171 were seventhgraders, and 191 were eighth graders. Four partici-pants had missing data and were not included in thefollowing analyses. Based on the year of birth (exactdates were not asked), the mean age was approxi-mately 11.5 years for the sixth graders, 12.5 years forthe seventh graders, and 13.5 years for the eighthgraders. Of the 577 participants, 540 (93.6%) identi-fied themselves as being ethnically Italian (457 re-ported being only Italian, and 83 reported beingItalian and some other ethnicity such as Albanian,French, Jewish). Thirty-seven participants (6.4%) re-ported belonging only to a non-Italian ethnic group(e.g., Albanian, French, German).

Measures

Self-report of antisocial behavior. A self-reportquestionnaire (Kiesner, 2002; Kiesner et al., 2003) wasused as one measure of the participants’ antisocialbehavior. All of the items measuring antisocial be-haviors (18 items) were included in the self-reportscore of antisocial behavior. These items included,for example, ‘‘lied to parents,’’ ‘‘hit someone,’’ ‘‘stolesomething from a store,’’ ‘‘did graffiti on publictransportation or property,’’ ‘‘used alcohol,’’ and‘‘used drugs.’’ Participants were asked to indicatehow often they were involved in these behaviorsthinking about the last week, using a 4-point scale:with 05 never, 15 rarely, 25 sometimes, 35 frequently.The Cronbach’s alpha for these items was a5 .87,with all item-to-total correlations greater than .30.The self-report antisocial behavior score is the meanof the nonstandardized scores on the 18 items. Aone-way analysis of variance (ANOVA) showed asignificant difference across age groups (F(2, 570)510.14; po.001), with levels of self-reported antisocialbehavior increasing from the sixth grade to theeighth grade (means are presented in Table 1).

Peer reports of problem behavior. Peer nominationswere conducted within each classroom, providingeach participant with a list of all classroom peers(participants and nonparticipants were included onthe list). Unlimited and cross-gender peer nomina-tions on three behavioral questions were used as ameasure of problem behavior. These questions were:‘‘Who are the kids that tease (in a mean way) otherkids?,’’ ‘‘Who are the kids that hit other kids?,’’ and‘‘Who are the kids that get into trouble?.’’ Thenumber of nominations received by classmates oneach of these questions was computed and thenstandardized within each classroom to control for

Antisocial Behavior and Peer Acceptance 1281

differences in class size. The Cronbach’s alpha forthese items was a5 .91, with all item-to-total corre-lations greater than .80. These standardized scoreswere then averaged together to create the peer reportof antisocial behavior score.

Teacher report of problem behavior. An adaptedversion of a teacher-report questionnaire (Kiesner,2002) was used to measure problem behavior in theclassroom during the past week. Six items from thequestionnaire were used to calculate the problembehavior score. These questions included, for exam-ple, ‘‘was argumentative?’’ and ‘‘disturbed class-mates?’’ All questions required a response using a6-point scale, ranging from no, not at all to yes, fre-quently. A separate form was used for each student.The Cronbach’s alpha for the 6 items was a5 .92,with all item-to-total correlations greater than .70.Individual scores are the mean of the nonstandard-

ized items. These scores were not standardizedwithin the classroom because each child’s score wasindependent of the number of students within eachclassroom. A one-way ANOVA showed a significantdifference across age groups (F(2, 570)5 5.12; po.01),with levels of teacher-reported antisocial behaviordecreasing from the sixth grade to the eighth grade(means are presented in Table 1).

Classroom peer preference. Peer nominations wereconducted within each classroom, providing eachparticipant with a list of all classroom peers (partic-ipants and nonparticipants were included on thelist). Unlimited and cross-gender peer nominationsfrom classmates on the liked-most (LM) and liked-least (LL) questions were used to assess each ado-lescent’s level of social preference. The number ofnominations received on these items was computedfor each participant; these scores were then stan-

Table 1

Correlations Among all Measured Variables, by Grade Level

1 2 3 4 5 6 7

Sixth grade (n5 214)

1. Antisocial: self 1

2. Antisocial: teacher 0.37!!! 1

3. Antisocial: peer 0.42!!! 0.67!!! 1

4. Like most ! 0.09 ! 0.29!!! ! 0.39!!! 1

5. Like least 0.14! 0.45!!! 0.59!!! ! 0.77!!! 1

6. Group: in-school 0.12 ! 0.07 ! 0.16! 0.45!!! 0.38!!! 1

7. Group: out-of-school 0.18!! ! 0.02 ! 0.05 0.35!!! 0.25!!! 0.68!!! 1

M 0.83 1.93 ! 0.02 0.03 0.07 3.35 2.18

SD 0.49 1.21 0.89 0.94 0.93 2.38 2.02

Seventh grade (n5 170)

1. Antisocial: self 1

2. Antisocial: teacher 0.38!!! 1

3. Antisocial: peer 0.34!!! 0.66!!! 1

4. Like most 0.10 ! 0.05 ! 0.07 1

5. Like least ! 0.04 0.12 0.20!! ! 0.79!!! 1

6. Group: in-school ! 0.01 ! 0.06 ! 0.03 0.57!!! 0.49!!! 1

7. Group: out-of-school 0.01 ! 0.07 ! 0.02 0.41!!! 0.31!!! 0.63!!! 1

M 0.78 1.70 ! 0.05 0.04 0.07 4.36 2.72

SD 0.43 0.98 0.87 0.94 0.98 2.90 2.31

Eighth grade (n5 189)

1. Antisocial: self 1

2. Antisocial: teacher 0.44!!! 1

3. Antisocial: peer 0.52!!! 0.67!!! 1

4. Like most 0.21!! 0.04 0.04 1

5. Like least ! 0.06 0.06 0.17! ! 0.80!!! 1

6. Group: in-school 0.18! 0.25!! 0.23!! 0.49!!! 0.39!!! 1

7.Group: out-of-school 0.24!! 0.21!! 0.26!!! 0.38!!! 0.21!!! 0.64!!! 1

M 1.01 1.59 ! 0.04 0.05 0.06 4.47 2.40

SD 0.54 0.92 0.86 0.93 0.93 3.20 2.51

!po.05. !!po.01. !!!po.001

1282 Kiesner and Pastore

dardized within each classroom and across gender.The bivariate correlation between LM and LL wasr5 ! .78 (po.0001, n5 577).

The LM and LL scores were used to create a latentconstruct of classroom peer preference. For the SEManalyses using this latent construct, the LL score wasreverse coded (multiplied by ! 1) so that high scoreson the latent construct indicate high peer preferenceand low scores indicate low peer preference.

Nominations of in-school network members and after-school network members. Nominations of in-schooland after-school peer networks were used to havemeasures of each individual’s level of in-school andafter-school peer network inclusion (Kiesner et al.,2003). The procedure for obtaining these nomina-tions involved two steps. First, students were given adefinition of a ‘‘group’’ that included two key char-acteristics: (1) there must be at least three children inthe group (including the target child), and (2) thesechildren must spend time together. Thus, a dyadwould not qualify as a group and a set of indepen-dent friends who do not spend time together wouldnot qualify as a group. In the second step, partici-pants were asked to list the names of their peers whowere in their group. This procedure was repeated foran in-school group and an after-school group. Par-ticipants were allowed to nominate only one groupfor each context. For the in-school group, partici-pants were able to list any students from the school,whether or not those students were participating inthe study. For the after-school group, participantswere allowed to nominate anybody, whether or notthose individuals attended the same school, andwhether or not those individuals were participating inthe study. All participants were asked to nominate thein-school group first and the after-school group second.As previously reported (Kiesner et al., 2003), 17 par-ticipants (2.9%) reported having no in-school group,and 21 participants (3.6%) reported not having an after-school group. Four of these participants had neither anin-school nor an after-school group. The mean pro-portion of overlapping members was M5 0.33(SD5 0.30), with 125 (23%) having no overlappingmembers and 37 (6.8%) having complete overlap.

A 2 (gender) # 3 (grade) multivariate analysis ofvariance (MANOVA) was used to test for differencesin the size of the groups for both the in-school groupand the out-of-school group. The overall MANOVAindicated a main effect for grade (Wilk’s lamb-da5 .94, F(4, 1072)5 8.5, po.001), but no effect ofgender and no interaction between grade and gen-der. Univariate tests confirmed a significant effect ofgrade for both the in-school group (F(2, 537)5 9.15,po.001) and the out-of-school group (F(2, 537)5 13.0,

po.001). Examination of the means indicated ageneral tendency for older children to have slightlylarger in-school groups (sixth grade M5 4.96, sev-enth grade M5 6.07, eighth grade M5 5.71), andout-of-school groups (sixth grade M5 4.59, seventhgrade M5 5.35, eighth grade M5 5.60). For both in-school and out-of-school groups, group size was posi-tively, although generally weakly, related to the threemeasures of antisocial behavior (correlations rangingfrom r5 .08 to r5 .21), suggesting that antisocialyouth tended to nominate slightly larger groups.

Because individuals could nominate study partic-ipants and nonparticipants as group members, wedid not have data on all nominated group membersand a precise measure of reciprocity was not possi-ble. However, to estimate the level of reciprocity foreach participant, the number of nominations thatwere reciprocated was divided by the number ofnominated group members who were study partici-pants. This was done separately for the in- and after-school groups. For the in-school group the meanproportion of reciprocated nominations was M5 0.59(SD5 0.34), and for out-of-school groups the meanproportion of reciprocated nominations wasM5 0.45 (SD5 0.37). These proportions likely un-derestimate actual reciprocity because individualsmay belong to multiple groups in each context, butwere restricted to nominating only one group foreach context. Reciprocity was slightly negativelycorrelated with the three measures of antisocial be-havior (correlations ranging from r5 ! .02 to ! .18),suggesting that antisocial individuals receivedslightly fewer reciprocated nominations.

For the out-of-school group, the average proportionof members who attended the same school as thetarget individual was M5 0.61 (SD5 0.35). The pro-portion of out-of-school groupmembers who attendedthe same school as the target child was uncorrelatedwith all three of the measures of antisocial behavior.

To obtain separate measures of a child’s peer in-clusion across contexts, we computed two separatescores. First, we computed the number of times thateach child was nominated by his or her peers as an in-school network member (in-school peer networkinclusion). Second, we computed the number oftimes that each child was nominated by his or herpeers as an after-school network member (after-school peer network inclusion). The mean number ofnominations received as an in-school network mem-ber was M5 4.0 (SD5 2.9) and as an after-schoolnetwork member was M5 2.4 (SD5 2.3). Becausethe reference group for the in-school peer networkinclusion measure was the same-school peers, thisscore was standardized within each school.

Antisocial Behavior and Peer Acceptance 1283

The in-school and after-school peer inclusionscores were used to create a latent construct of peerinclusion outside of the classroom. The bivariatecorrelation between the standardized in-school peernetwork inclusion score and the after-school peernetwork inclusion score was r5 .62 (po.0001, n5577). Thus, individuals who are accepted by peers atthe school level also tend to be accepted by peersoutside of the school.

Group-delinquency score. A group-delinquency scorewas calculated to measure the average level of delin-quency demonstrated by the group members within(1) the in-school group and (2) the after-school group.These scores are based on a subscale of 6 items fromthe self-report measure of antisocial behavior thatassessed more severe forms of antisocial behaviorsuch as vandalism, stealing, and substance use. TheCronbach’s alpha for these 6 items was a5 .77, withall item-to-total correlations greater than .48.

To obtain this score we calculated the average levelof delinquency within each peer network, includingall nominated network members except the targetindividual. This was done separately for the in-school network and the after-school network. Forindividuals who had no peer network (either in- orafter-school) or for whom no network members wereparticipants (thus, for whom we had no behavioraldata), we assigned missing data for the correspond-ing network score. Note that network members whowere not study participants (i.e., attended one of theschools but did not participate; lived outside of theneighborhood and attended some other school)could not be included in the calculation of networkbehavior scores. The mean proportion of peer net-work members who were participants, and thus forwhomwe had data, wasM5 0.75 (SD5 0.24, n5 560)for the in-school peer network and M5 0.50 (SD50.32, n5 556) for the after-school peer network.

Use of a latent construct for antisocial behavior. Asstated in the introduction, in the primary analyses ofthis study we use a latent construct of antisocialbehavior, including various types of problem be-havior that occur in a variety of contexts. Although itarguably would be better to consider different typesof antisocial behavior separately (substance use, ag-gressive behavior, property crimes), in the presentdata set we did not have multiple informants foreach of these behavior types, and thus behavior typewould have been confounded with informant, anddifferences in effects would not have been unam-biguously attributable to behavior type or informantbias. Note that, in a previous publication using thesedata (Kiesner et al., 2003), in-school problem be-havior and after-school delinquency were analyzed

separately. However, because those earlier analysesfocused primarily on the group-delinquency scores,for which we had multiple informants, responsebiases of specific respondents did not represent athreat to the internal validity. Also, it should benoted that, although the three reports of antisocialbehavior considered different time periods (self- andteacher reports asked about behavior over the pastweek, whereas peer reports asked about generalbehavioral tendencies), the use of a latent constructallowed us to have a general measure of antisocialbehavior, free from bias introduced by any of thespecific reporters or time periods. Therefore, we de-cided to use a latent construct of antisocial behavior,which we believe provides the most conservativeapproach for testing the present hypotheses.

Results

To test the hypothesis that the relations between anti-social behavior and peer acceptance, across contexts,differ across the sixth, seventh, and eighth grades,we used SEM multigroup analyses (LISREL 8, Jore-skog & Sorbom, 1996). The correlation matrices,means, and standard deviations for all measures arepresented in Table 1, separately for each grade level.Examining these bivariate correlations, a generaltrend can be seen for antisocial behavior to be neg-atively related to in-class peer acceptance in the sixthgrade but positively related to out-of-class peer ac-ceptance in the eighth grade. Importantly, it shouldbe noted that this overall pattern is very similaracross all three measures of antisocial behavior.

Measurement Models

The first step of these analyses was to test whetherthe two separate measurement models (the antisocialconstruct on the left side of Figure 1, and the peer-relations constructs on the right side of Figure 1)were plausible, and to test for measurement invari-ance across the three age groups. After testingwhether each measurement model was appropriatefor the overall sample, we used procedures outlinedby Joreskog (1971) to test for age-related differencesin the covariance matrices for each of the twomeasurement models, and then to test for differencesin the factor loadings.

We first examined the measurement model for theantisocial behavior construct. Because this modelwas saturated, it was not possible to test for a goodfit; however, the factor loadings indicated that allthree measures were significantly correlated with thelatent construct (standardized factor loadings were

1284 Kiesner and Pastore

.48 for self-report; .75 for teacher report; .88 for peerreport; all ts411.0). A multigroup analysis testing forage-related differences in the covariance matrices ofthe three measures of antisocial behavior demon-strated that significant differences did exist (w2533.62, df5 12, po.001). Moreover, the multigroupanalysis testing for invariance of the factor loadingacross the three age groups also resulted in signifi-cant differences (w25 13.05, df5 6, p5 .05). Although

significant differences were found in this multigroupanalysis, an examination of the factor loadings, sep-arately for each class, indicated that the differenceswere very small, and that the same general patternheld across all three age groups (standardized factorloadings for self-, teacher-, and peer-reports were .48,.76, and .88 for the sixth grade, .44, .86, and .77 for theseventh grade, and .59, .75, and .88 for the eighthgrade, respectively).

Self

Peer

Teacher

Antisocial

ClassroomPreference

Out-of-ClassInclusion

ClassroomLike-Most

ClassroomLike-Least

Group InclusionIn-School

Group InclusionOut-of-School

!.62**

.13

.78†

1.0**

.92†

.73**

.48**

.44**

.95**

.72**

Self

Peer

Teacher

Antisocial

ClassroomPreference

Out-of-ClassInclusion

ClassroomLike-Most

ClassroomLike-Least

Group InclusionIn-School

Group InclusionOut-of-School

!.17*

!.01

.79†

1.0**

.99†

.64**

.49**

.43**

.80**

.84**

Self

Peer

Teacher

Antisocial

ClassroomPreference

Out-of-ClassInclusion

ClassroomLike-Most

ClassroomLike-Least

Group InclusionIn-School

Group InclusionOut-of-School

!.14

.28**

.80†

1.0**

.98†

.64**

.44**

.58**

.90**

.75**

(a)

(b)

(c)

Figure 1. Structural equation models conducted separately for the sixth grade (a), seventh grade (b), and eighth grade (c). Fit indexes arepresented in the text. wSignificance not estimated for these parameters, !po.05, !!po.001.

Antisocial Behavior and Peer Acceptance 1285

Second, we examined the measurement model forthe peer-relations constructs. The first step was tocompare a one-factor model (with all four measuresof peer-relations loading onto the same single factor)with a two-factor model (including a path goingfrom classroom preference to out-of-class inclusion,as presented on the right side of Figure 1). The one-factor model did not fit the data well (w25 230.64,df5 2, po.001, GFI5 .85, CFI5 .76, NNFI5 .27,RMSEA5 .41). For the two-factor model, althoughthe w2 was significant, the other fit indexes show-ed that this model did fit the data reasonablywell (w25 8.63, df5 1, p5 .01, GFI5 .99, CFI5 .99,NNFI5 .95, RMSEA5 .12), and demonstrated asignificant improvement to the one-factor model(Dw25 222.01, Ddf5 1, po.001).

We next conducted a multigroup analysis com-paring the covariance matrices (across age groups) ofthe four peer-relations measures. Differences wereagain found (w25 40.82, df5 20, p5 .004). However, amultigroup analysis testing for differences specifi-cally in the factor loadings showed that the factorloadings were invariant across the three age groups(w25 12.21, df5 7, p5 .09).

Relations Between Antisocial Behavior and Peer Relations

To test our hypotheses regarding differences in therelations between antisocial behavior and peer rela-tions, across the different age groups, we used themodel presented in Figure 1. In this model we usedthree measures of antisocial behavior to create a la-tent construct for antisocial behavior; we used twopeer-report measures to create a latent constructfor classroom peer preference; and we used twopeer-report measures to create a latent construct forout-of-class peer inclusion. We then specified pathsgoing from antisocial behavior to each of the peer-relations constructs. These are the paths used to testthe hypotheses regarding changes in the relationsbetween antisocial behavior and peer relations, andare therefore the path coefficients most importantwith regard to theory testing and interpretation. Wealso allow a path from classroom peer preference toout-of-class peer inclusion.

To test for differences across the three age groups,we followed a three-step procedure. First, we speci-fied a single ‘‘baseline’’ model separately for all threegroups (sixth, seventh, and eighth grades) to deter-mine whether the overall model was plausible foreach of the three age groups. Second, we tested fordifferences in w2 when each of the paths of interestwas constrained to be equal across age groups, andthen when it was allowed to vary across age groups.

Finally, for the sixth and eighth grades, we tested forchanges in w2 when each of the paths of interest wasfirst constrained to be zero, then when it was released.

In all of the models presented below, the errorvariance for the classroom LL nominations was con-strained to be zero, and the error variance for LMnomination was allowed to correlate with the errorvariance of the two out-of-class nominations mea-sures. For simplicity, these correlations will not bepresented in the figures, but it should be noted thatthey were significant in all cases.

In the first step, we tested the specified modelseparately for each of the three age groups. The fitindexes were as follows: w25 26.96, df5 11, p5 .005,GFI5 .97, CFI5 .98, NNFI5 .96, RMSEA5 .08 forthe sixth graders (note that the degrees of freedomfor this group are different because we were requiredto constrain one additional error variance term);w25 12.27, df5 10, p5 .27, GFI5 .98, CFI5 .99,NNFI5 .99, RMSEA5 .04 for the seventh graders;and w25 37.17, df5 10, po.001, GFI5 .95, CFI5 .95,NNFI5 .89, RMSEA5 .12 for the eighth graders. Al-though the w2 was significant for sixth and eighthgraders, this index has been widely criticized, and it isgenerally recommended to also consider other good-ness-of-fit indexes (Mulaik et al., 1989). Examining theother fit indexes suggests that this model provided areasonably good fit for all three age groups. Based onthese results, we concluded that, in order to test forage-related differences in the structural coefficientsof interest, it was reasonable to apply the samestructural model to all three age groups.

The structural coefficients for each of the three agegroups are presented in Figures 1a– c. The first im-portant finding in these analyses is that, as predicted,the relation between antisocial behavior and class-room peer preference is negative and significant inthe sixth grade, but then drops in magnitude andsignificance in the seventh and eighth grades. Thesecond important finding in these analyses is that therelation between antisocial behavior and out-of-classpeer inclusion is not statistically significant in thesixth and seventh grades, but becomes significantand positive in the eighth grade. Thus, in the eighthgrade, but not the sixth and seventh grades, antiso-cial behavior is significantly associated with higherlevels of peer acceptance outside of the classroom.

It should be noted that all of the other coefficientswere remarkably stable across the 3 years. Thisstability was evident for the measures contributingto each of the constructs, as well as for the path goingfrom classroom preference to out-of-class inclusion.This is important because it demonstrates that, al-though the factor structures and the relation between

1286 Kiesner and Pastore

peer acceptance in- and out-of-class remain stable, therelation between antisocial behavior and peer accept-ance across different contexts changes dramatically.

Next, we tested whether it was plausible to con-strain the paths from the antisocial behavior con-struct to each of the peer-relations constructs to beinvariant across the three age groups. To do this weexamined the change in w2 when each of these pathswas first constrained to be equal across age groups,and then when it was allowed to vary across agegroups. Because measurement invariance had beendemonstrated for the factor loadings of the peer-re-lations measures, these factor loadings were fixed tobe equal across age groups using the loadings ob-tained by the multigroup measurement model. Onthe other hand, because differences were found forthe factor loadings of the antisocial behavior con-struct, these factor loadings were fixed to be equal tothe loadings obtained in the baseline measurementmodel for each age group. Allowing each of thepaths of interest to vary across groups resulted in asignificant improvement in the model (Dw25 9.01,Ddf5 2, p5 .01 for the path going from antisocialbehavior to out-of-class inclusion; Dw25 32.51,Ddf5 2, po.001 for the path going from antisocialbehavior to classroom preference). These resultsdemonstrate that the paths from antisocial behaviorto each of the peer constructs are significantly dif-ferent across age groups.

Finally, considering the sixth and eighth gradesseparately, we tested whether the paths from theantisocial construct to each of the peer-relationsconstructs could be constrained to be equal to zero.For the sixth grade, releasing the path from antisocialbehavior to classroom preference resulted in a sig-nificant improvement in the model (Dw25 89.31,Ddf5 1, po.001), whereas releasing the path fromantisocial behavior to out-of-class inclusion did notresult in a significant improvement in the model(Dw25 1.83, Ddf5 1, p5 .18). For the eighth grade,releasing the path from antisocial behavior to class-room preference did not result in a significant im-provement in the model (Dw25 2.99, Ddf5 1,p5 .084), whereas releasing the path from antisocialbehavior to out-of-class inclusion did result in asignificant improvement in the model (Dw25 13.74,Ddf5 1, po.001). Thus, as expected, in the sixthgrade, the path from antisocial behavior to classroompreference, but not to out-of-class inclusion, wasfound to be important for the overall fit of the model.On the other hand, in the eighth grade, the path fromantisocial behavior to out-of-class inclusion, but notto classroom preference, was found to be importantfor the overall fit of the model.

We also tested for differences between males andfemales across the three age groups. Although dif-ferences were found in the covariance matrices formales and females in both the sixth and eighthgrades (but not the seventh grade), multigroupanalyses comparing the structural model requiredthat different error terms be constrained for malesand females across age groups. Moreover, because ofsmall sample sizes, these analyses appeared sensi-tive to disturbances of correlated errors and a fewmoderate outliers. Therefore, formal comparisons ofthe structural model were not conducted. However,when the model was fit separately for males andfemales, for each grade level, a similar overall pat-tern was observed for both males and females. Fromthe sixth grade to the eighth grade, for both malesand females, there was a decrease in the negativerelation between antisocial behavior and classroompeer preference, and an increasing and positive re-lation between antisocial behavior and out-of-classpeer inclusion.

By Whom is Delinquency Accepted?

The above analyses show that, throughout themiddle school years, antisocial behavior becomesassociated with increasing peer success outside ofthe classroom and a diminished negative effect onpeer preference within the classroom. To testwhether the age-related differences are attributableto an increase in affiliations between average indi-viduals and antisocial peers, or to an increase inaffiliations among antisocial peers, we conducted a2 # 2 # 3 factorial MANOVA, using gender, grade,and individual antisocial behavior as the indepen-dent variables. The dependent variables were thegroup-delinquency scores for both the in-school andout-of-school groups. As noted in the Methods sec-tion, these group-delinquency scores were simplythe average level of delinquency among the groupmembers (excluding the target individual), for boththe in-school and out-of-school groups, on a 6-itemdelinquency subscale (see the Methods section for adescription of how these scores were calculated).

The specific items included in the group-delin-quency scores focused on vandalism, stealing, anddrug use. These delinquent behaviors were used forthree reasons. First, as noted in the introduction, itwould be most consistent with Moffitt’s theory toexpect these types of behaviors to be most stronglyassociated with increases in peer acceptance becausethey are likely to be viewed as adult behaviors (as com-pared with disruptive classroom behaviors or otherless severe forms of problem behaviors). Second,

Antisocial Behavior and Peer Acceptance 1287

because the SEM models presented above showedthat antisocial behavior was positively associatedwith only out-of-class peer inclusion, we believedthat antisocial behaviors likely to occur outside of theclassroom would be most relevant to age-relateddifferences in peer affiliations. Finally, because thegroup-delinquency scores were based on multiplerespondents (of the various group members), indi-vidual response bias was not a significant threat andwe were not constrained to use a latent construct ofantisocial behavior. Therefore, at both a theoreticallevel and an empirical level, considering more severedelinquent behaviors seems most appropriate.

To create three antisocial behavior groups (low,average, high), based on individual antisocial be-havior, we conducted a tertile split on a combinedself-, peer-, and teacher-report antisocial-behaviorscore. Because the most evident differences betweenantisocial behavior and peer acceptance/inclusionacross contexts was between the sixth graders andthe eighth graders, only the sixth and eighth graderswere included in this analysis.

Before addressing the stated research question, wetested for mean level differences of delinquency acrossthe two peer groups using a 2 (group: in-school group,out-of-school group) # 2 (gender)# 2 (grade) repeat-ed measures ANOVA. The measures of in-school andout-of-school group delinquency were standardizedacross the sixth and eighth grades to have an overallmean of zero and a standard deviation of one. Ourinterest was whether differences existed between thein-school and out-of-school group-delinquency scores(the within-subjects factor), and whether this factorinteracted with the between-subjects factors. Therewas nomain effect for peer group (F(1, 315)5 2.50, ns),but a significant interaction was observed betweenpeer group and grade (F(1, 315)5 4.27, po.05). Ex-amination of the means shows that in the sixth gradedifferences were minimal (M5 ! 0.09 for the in-school group;M5 ! 0.10 for the out-of-school group),whereas in the eighth grade the out-of-school groupshowed a higher mean level of delinquency (M5 0.26)as compared with the in-school group (M5 0.18).

For testing the hypotheses regarding by whomdelinquency is accepted, the measures of in-schooland out-of-school group delinquency were stan-dardized across the sixth and eighth grades to have anoverall mean of zero and a standard deviation of one.Significant multivariate effects were found for grade(Wilk’s lambda5 .90, F(2, 306)5 17.70, po.0001),individual level of antisocial behavior (Wilk’s lamb-da5 .83, F(4, 612)5 15.02, po.0001), and the inter-action between grade and individual antisocialbehavior (Wilk’s lambda5 .95, F(4, 612)5 4.08, po.01).

No other effects were significant. When separateanalyses were conducted for the measures of in-school group delinquency and out-of-school groupdelinquency, the same pattern of results was foundfor both dependent measures. Of particular impor-tance is the significant interaction found betweengrade and individual antisocial behavior for both thein-school group’s delinquent behavior (F(2, 307)5 7.15,po.001) and the out-of-school group’s delinquentbehavior (F(2, 307)5 6.76, po.001; see Table 2 for cellmeans and standard deviations). It should be notedthat, although there were significant differences inthe group-delinquency scores across grade levels,the interaction cannot be attributed to this differencebecause the interaction is tested after controlling forthe main effect of grade.

To further help interpret the interaction betweengrade and individual antisocial behavior, a bar graphfor the out-of-school group’s delinquency scores ispresented in Figure 2 (the plot for the in-schoolgroup’s delinquent behavior is almost identical andleads to the same conclusions, so is not presentedhere). The means presented in Table 2 and the bargraph presented in Figure 2 show that individualswho themselves show relatively high levels of anti-social behavior (upper tertile) consistently nominatemore delinquent individuals as group members,both in the sixth and eighth grades. Moreover, acrossall three levels of individual antisocial behavior,participants demonstrated higher group-delinquen-cy scores in the eighth grade as compared with thesixth grade (also verified testing for grade effectsseparately at each level of individual antisocial be-havior, i.e., simple effects). However, the differencebetween the sixth grade and the eighth grade was

Table 2

Cell Means and Standard Deviations (in Parentheses) for Out-of-School

and In-School Group Delinquency by Grade and Individual Level of

Antisocial Behavior

Individual antisocial behavior

Low Average High

Out-of-school group delinquency

Grade

Sixth ! .43 (.59) ! .42 (.48) ! 0.07 (0.77)

Eighth ! .12 (.87) ! .01 (.77) 1.06 (1.52)

In-school group delinquency

Grade

Sixth ! .51 (.51) ! .37 (.64) 0.08 (0.82)

Eighth ! .15 (.81) ! .07 (.78) 1.04 (1.46)

Note. All ns $ 50 and % 71.

1288 Kiesner and Pastore

clearly greater for the individuals who themselveswere relatively high on antisocial behavior. Thus, itappears that the increase in peer acceptance of anti-social individuals is attributable to increased nomi-nations received by other adolescents with relativelyhigh levels of antisocial behavior.

It should be noted that these analyses were re-peated using teacher and peer reports of antisocialbehavior for calculating the group behavior scores,and in both cases the interaction between grade andindividual antisocial behavior was nonsignificant,suggesting that these results are specific to moresevere types of delinquency.

Discussion

This study was conducted to test the hypothesis thatthe relation between antisocial behavior and peeracceptance would differ across 3 years during ado-lescence (from the sixth grade to the eighth grade).Results demonstrated that in the sixth grade antiso-cial behavior was negatively and significantly relat-ed to classroom peer preference, but not significantlyrelated to out-of-class inclusion. In the eighth grade,however, antisocial behavior was not significantlyrelated to in-class peer preference but was positivelyand significantly related to out-of-class peer in-clusion. Thus, these data support the idea that,during adolescence, antisocial individuals experi-ence an increase in peer acceptance (and a decreasein peer rejection).

One important contribution of this study was totest simultaneously for age-related differences in therelations between antisocial behavior and peer ac-ceptance across both classroom and out-of-classcontexts. As noted by Kiesner et al. (2003), peer-

relations research has focused almost exclusively onclassroom contexts, and when research in this fieldhas included nonclassroom or nonschool peers (see,e.g., Dishion, Andrews, & Crosby, 1995; Pettit, Bates,Dodge, & Meece, 1999) it typically has not done socomparatively. In this study we were able to simul-taneously test for age-related differences in bothclassroom and nonclassroom/out-of-school contexts.The fact that clear differences were found in the re-lation between antisocial behavior and peer accept-ance across these different peer contexts providescompelling evidence that studying cross-contextpeer relations will provide important information forunderstanding the socializing influence of peers.

These results also suggest that our theories needto consider the context specificity of peer relationsand how they are affected by individual behavior.For example, based on Moffitt’s (1993) theory onecould hypothesize that, during adolescence, antiso-cial behavior should be positively related to peeracceptance. Although the present study providessome support for this hypothesis, the hypothesisneeds to be further refined to specify contextualdifferences in where the behavior is exhibited andwhich peers are considered. In the introduction weproposed that classroom and nonclassroom contextswould differ because classroom peers are not chosen,and individuals with different behavioral tendenciesare constrained to be together. Moreover, in theclassroom, individuals are less able to avoid thenoxious and aversive behaviors of their antisocialpeers. As a result, although antisocial behavior maybe more tolerated during adolescence, it does notappear to be associated with high levels of accept-ance within the classroom. The out-of-class context,however, appears to be very different. In this context,where individuals are able to avoid the noxious andaversive behaviors of antisocial youth, and are ableto select more actively with whom they pass theirtime, antisocial behavior becomes positively associ-ated with peer inclusion. Thus, theories regardingantisocial behavior and peer acceptance should con-sider both age and context effects.

Although the age-related differences in the rela-tion between peer acceptance and antisocial be-havior are consistent with Moffitt’s theory, thepresent data also showed that much of this differenceis attributable to increased affiliations among delin-quent youth and not to an increased acceptance ofdelinquent youth by average peers. Thus, rather thanbecoming popular with average individuals, antiso-cial individuals are becoming more popular witheach other. This suggests an increased organizationand grouping of delinquent youths during the mid-

!0.6

!0.4

!0.2

0

0.2

0.4

0.6

0.8

1

1.2

Sixth EighthGrade

Afte

r-S

choo

l Gro

up D

elin

quen

cy

Low Average High

Individual Level of Antisocial Behavior

Figure 2. Bar graph for the out-of-school group-delinquency score,presented separately by grade and individual level of antisocialbehavior (low, medium, high, based on tertile split).

Antisocial Behavior and Peer Acceptance 1289

dle school years, and is consistent with Dishion’sconfluence model (Dishion et al., 1994; Dishion,French, & Patterson, 1995) that delinquent individ-uals form homogeneous groups who reinforce anti-social behavior among themselves.

As noted in the introduction, the fact that Italianmiddle school students stay with the same class-mates all day, for all 3 years of middle school, maymake it more difficult for Italian middle school stu-dents to change their school-based peer relationsfrom one year to the next. If this is true, then the dif-ferences across age groups observed in the presentsample may be attenuated (at least when consideringthe school context) as compared with what would befound in other countries. Furthermore, research andreplication across different countries will be neededto better understand how cross-national differencesin school organization and culture may influencethese effects.

In a previous publication based on the same dataset (Kiesner et al., 2003), it was shown that in-schoolpeer inclusion and after-school peer inclusion inter-acted in explaining variance in depression. Specifi-cally, for youth with high levels of peer inclusionoutside of the school, there was no relation betweenin-school peer inclusion and depressive symptoms.Thus, high peer acceptance in one context may bufferthe negative effects of low peer acceptance in othercontexts. Considered together with the presentfindings, it could be hypothesized that the relationbetween antisocial behavior and depression (Cap-aldi, 1991; Kiesner, 2002) also changes across this ageperiod, and that such relations would be differen-tially mediated through peer relations across differ-ent contexts. Such context-specific mediation mayhelp explain why the mediational model proposingthat antisocial behavior predicts peer failure, whichthen predicts depressive symptoms, has not receivedstrong support (Kiesner, 2002).

It is also important to note that in the previouspublication using this data set (Kiesner et al., 2003), itwas suggested that specific behavior types, and thecontexts in which they occur, should be consideredin research focusing on antisocial behavior and peerrelations. This point is further emphasized by theresults from the present study showing that the in-creased networking among the eighth-grade antiso-cial youth was found for more severe forms ofdelinquency, but not for teacher and peer reports ofclassroom problem behavior. However, in the pri-mary analyses of this study we used a general con-struct of antisocial behavior. This was done becausewe did not have multiple informants for varioustypes and contexts of antisocial behavior, and re-

sponse biases unique to a specific source of infor-mation would have represented a serious threat tothe internal validity of this study (which was not thecase for the earlier publication, or the analyses in thisstudy using the group-delinquency scores). Thus, itwill be the task of future research to further test thecurrent findings considering different behavior typesthat are specific to the different contexts.

A related point is that, whereas mean levels of self-reported antisocial behavior increased across the threeage groups, mean levels of teacher-reported antisocialbehavior showed a significant decrease across thethree age groups. Although the relations among thesevariables remained similar across age groups (seefactor loading for the antisocial behavior construct),the different pattern of mean differences across mea-sures further suggests that, to understand fully thelinks between antisocial behavior and peer relations,we need to consider multiple behavioral contexts.

Findings from this study suggest that the overallpattern of results held for both males and females.This is different from the results of the earlier pub-lication based on the same data set (Kiesner et al.,2003). However, because the research questions werevery different and because different analytic strate-gies were used (latent construct of antisocial be-havior vs. separate measures of antisocial behaviorin different contexts), these seemingly contradictoryresults are difficult to interpret together.

There are several limitations in this study thatshould be noted. First, the construct of peer inclusionoutside of the classroom was based solely on nomi-nations of group membership. It would be verypossible, however, that an individual with no groupmembership would have several out-of-class friends.Because the nominations process in this study con-sidered only group memberships, these dyadic peeraffiliations would not have been detected. However,the focus of this study was specifically on groups,and friendship nominations were not used. Thus wehad no information on dyadic friendships and howsuch nominations may have added information tothe present analyses and conclusions.

A second limitation concerns the calculation of thegroup-delinquency scores used to test with whomantisocial youth are becoming more accepted. Thesescores are based on each individual’s nominations oftheir group members, without requiring that nomi-nations be reciprocated or verified by another per-son. As a result, some group members may beincluded in a group even though they did not re-ciprocate the nomination. Moreover, two individualswho belong to the same group (by nominating manysimilar members) may not nominate the exact same

1290 Kiesner and Pastore

group members. As a result, even though these twoindividuals may nominate each other, and severalcommon members, they may have different group-delinquency scores. However, creating a singleuniform group for all members of the same group,although conceptually appealing, likely does notrepresent the reality and complexity of groupmemberships. For example, previously used meth-ods for identifying cliques and social networks(NEGOPY, Richards & Rice, 1981; Social-CognitiveMaps, Cairns, Perrin, & Cairns, 1985) provide mapsor sociograms of various groups within a school orclass, but these maps may not correspond to whatthe individuals consider to be their groups. Thisoccurs when various group members nominate someoverlapping members but not all overlappingmembers (e.g., if child A nominates children B, C,and D, and children C and D reciprocate thesenominations and also nominate children E, F, and G,then the group would likely include children A, B, C,D, E, F, and G, even though child A did not nominatechildren E, F, and G). The social map for the specificindividual then can become larger and more inclu-sive than the individual would consider to be his orher group. Moreover, an individual may nominate aperson in his or her group, but if most other groupmembers do not also nominate that person, then heor she will not be included in the group, even thoughthat person is very important for the target individ-ual. Using the above example, this would occur ifchild B was not nominated by most of the othermembers, but was considered to be a very importantindividual, and group member, to the target child.Therefore, although the approach used for creatinggroup-delinquency scores in this study has somelimitations, we believe that it is the best way tocapture the peer context of who the individual con-siders to be his or her group.

A third limitation concerns the measures of peeracceptance that were used. Although classroom peerpreference and nominations received as a groupmember do capture certain aspects of peer accept-ance, the full picture is certainly more complex thancan be captured with these measures. For example,LaFontana and Cillessen (2002) found that, althoughphysical aggression was associated with high levelsof being disliked and low levels of being liked, ag-gressive youth were perceived to be the most popularand the least unpopular. These findings underlinethe difference between being liked by peers (mea-sured by asking students who they like and dislike)and being perceived as popular by peers (askingstudents who they think is popular and unpopular).Other studies have also demonstrated clear differ-

ences in being liked by peers and being perceived aspopular, as well as how these two measures aredifferently related to aggressive behavior (Cillessen& Mayeux, 2004; Farmer, Estell, Bishop, O’Neal, &Cairns, 2003; Prinstein & Cillessen, 2003). Also, arecent study by Lease, Musgrove, and Axelrod (2002)integrated various aspects of peer relations and ac-ceptance, including likeability, perceived popularity,and social dominance, showing that integratingthese different aspects of peer relations will likely beimportant for better understanding relations be-tween child behavior, peer relations, and future de-velopment. As suggested by Lease et al., futureresearch should attempt to integrate these differentaspects and complexities of peer relations.

Finally, the sample used in this studywas ethnicallyhomogeneous, with 93% identifying themselves asethnically Italian. Although such a homogeneoussample simplifies hypothesis testing because it mini-mizes the variance attributable to ethnicity, it canalso be considered as a limitation because we werenot able to test for ethnic differences or interactionsincluding ethnicity. It is very possible that, within thesame school and community, adolescents of differentethnic groups will have different norms that willcontribute to the relations between behavior andacceptance. Future research will need to extend thepresent line of research by also considering ethnicdifferences within a similar paradigm.

These limitations considered, it should be notedthat this study is one of the very few studies to ex-tend peer-relations research outside of the classroomand school. To our knowledge, it is the only study tocompare directly how antisocial behavior and peeracceptance are related to each other across contextsand across age groups.

In summary, this study provides partial supportfor the proposal that, during early adolescence,antisocial behavior becomes associated with peeracceptance. However, the higher level of peer ac-ceptance of antisocial youth in the eighth grade ap-pears to result from increased networking amongantisocial youth, rather than acceptance by averageyouth. These results suggest that future research, aswell as future prevention and intervention efforts,should attend to these age-related and context-spe-cific differences. In doing so, it may be especiallyimportant to consider the apparent increases in net-working among antisocial youth.

References

Attili, G., Vermigli, P., & Schneider, B. H. (1997). Peer ac-ceptance and friendship patterns among Italian school-

Antisocial Behavior and Peer Acceptance 1291

children within a cross-cultural perspective. InternationalJournal of Behavioral Development, 21, 277 – 288.

Brendgen, M., Vitaro, F., Turgeon, L., & Poulin, F. (2002).Assessing aggressive and depressed children’s social rela-tions with classmates and friends: A matter of perspec-tive. Journal of Abnormal Child Psychology, 30, 609 – 624.

Bukowski, W. M., Sippola, L. K., & Newcomb, A. F. (2000).Variations in patterns of attraction to same- and other-sex peers during early adolescence. Developmental Psy-chology, 36, 147 – 154.

Cairns, R. B., Cairns, B. D., Neckerman, H. J., Gest, S. D., &Gariepy, J. (1988). Social networks and aggressive be-havior: Peer support or peer rejection? DevelopmentalPsychology, 24, 815 – 823.

Cairns, R. B., Perrin, J. E., & Cairns, B. D. (1985). Socialstructure and social cognition in early adolescence:Affiliative patterns. Journal of Early Adolescence, 5,339 – 355.

Capaldi, D. M. (1991). Co-occurrence of conduct problemsand depressive symptoms in early adolescent boys:I. Familial factors and general adjustment at grade 6.Development and Psychopathology, 3, 277 – 300.

Casiglia, A. C., Lo Coco, A., & Zappulla, C. (1998). Aspectsof social reputation and peer relationships in Italianchildren: A cross-cultural perspective. DevelopmentalPsychology, 34, 723 – 730.

Cillessen, A. H. N., & Mayeux, L. (2004). From censure toreinforcement: Developmental changes in the associa-tion between aggression and social status. Child Devel-opment, 75, 147 – 163.

Claes, M., Lacourse, E., Bouchard, C., & Luckow, D. (2001).Adolescents’ relationships with members of the ex-tended family and non-related adults in four countries:Canada, France, Belgium and Italy. International Journalof Adolescence and Youth, 9, 207 – 225.

Dishion, T. J., Andrews, D. W., & Crosby, L. (1995). Anti-social boys and their friends in early adolescence: Re-lationship characteristics, quality, and interactionalprocess. Child Development, 66, 139 – 151.

Dishion, T. J., French, D. C., & Patterson, G. R. (1995). Thedevelopment and ecology of antisocial behavior. InD. Cicchetti & D. J. Cohen (Eds.), Developmental psycho-pathology. Vol. 2: Risk, disorder, and adaptation (pp. 421 –471). New York: John Wiley & Sons.

Dishion, T. J., Patterson, G. R., & Griesler, P. C. (1994). Peeradaptation in the development of antisocial behavior: Aconfluence model. In L. R. Huesmann (Ed.), Aggressivebehavior: Current perspectives (pp. 61 – 95). New York:Plenum.

Durrant, N. C., & Henggeler, S. W. (1986). The stability ofpeer sociometric ratings across ecological settings. Jour-nal of Genetic Psychology, 147, 353 – 358.

Eslea, M., Menesini, E., Morita, Y., O’Moore, M., Mora-Merchan, J. A., Pereira, B., et al. (2004). Friendship andloneliness among bullies and victims: Data from sevencountries. Aggressive Behavior, 30, 71 – 83.

Espelage, D. L., Holt, M. K., & Henkel, R. R. (2003). Ex-amination of peer-group contextual effects on aggres-

sion during early adolescence. Child Development, 74,205 – 220.

Farmer, T. W., Estell, D. B., Bishop, J. L., O’Neal, K. K., &Cairns, B. D. (2003). Rejected bullies or popular leaders?The social relations of aggressive subtypes of ruralAfrican American early adolescents. DevelopmentalPsychology, 39, 992 – 1004.

Farmer, T. W., Leung, M. C., Pearl, R., Rodkin, P. C., Cad-wallader, T. W., & Van Acker, R. (2002). Deviant or di-verse peer groups? The peer affiliations of aggressiveelementary students. Journal of Educational Psychology, 94,611 – 620.

French, D. C., & Conrad, J. (2001). School dropout as pre-dicted by peer rejection and antisocial behavior. Journalof Research on Adolescence, 11, 225 – 244.

Haselager, G. J. T., Cillessen, A. H. N., Van Lieshout, C. F.M., Riksen-Walraven, J. M. A., & Hartup, W. W. (2002).Heterogeneity among peer-rejected boys across middlechildhood: Developmental pathways of social behavior.Developmental Psychology, 38, 446 – 456.

Joreskog, K. G. (1971). Simultaneous factor analysis inseveral populations. Psychometrika, 36, 409 – 426.

Joreskog, K. G., & Sorbom, D. (1996). LISREL 8: User’s ref-erence guide. Chicago, IL: Scientific Software.

Kiesner, J. (2002). Depressive symptoms in early adoles-cence: Their relations with classroom problem behaviorand peer status. Journal of Research on Adolescence, 12,463 – 478.

Kiesner, J., Cadinu, M., Poulin, F., & Bucci, M. (2002).Group identification in early adolescence: Its relationwith peer adjustment and its moderator effect on peerinfluence. Child Development, 73, 200 – 212.

Kiesner, J., Nicotra, E., & Notari, G. (2005). Target specifi-city of subjective relationship measures: Understandingthe determination of item variance. Social Development,14, 109 – 135.

Kiesner, J., Poulin, F., & Nicotra, E. (2003). Peer relationsacross contexts: Individual-network homophily andnetwork inclusion in and after school. Child Development,74, 1328 – 1343.

Ladd, G. W. (1983). Social Networks of popular, average,and rejected children in school settings. Merrill-PalmerQuarterly, 29, 283 – 307.

LaFontana, K. M., & Cillessen, A. H. N. (2002). Children’sperceptions of popular and unpopular peers: A multi-method assessment. Developmental Psychology, 38,635 – 647.

Lease, M.A, Musgrove, K. T., & Axelrod, J. L. (2002). Di-mensions of social status in preadolescent peer groups:Likability, perceived popularity, and social dominance.Social Development, 11, 508 – 533.

Maggs, J. L., Almeida, D. M., & Galambos, N. L. (1995).Risky business: The paradoxical meaning of problembehavior for young adolescents. Journal of Early Adoles-cence, 15, 344 – 362.

Moffitt, T. E. (1993). Adolescent-limited and life-course-persistent antisocial behavior: A developmental taxon-omy. Psychological Review, 100, 674 – 701.

1292 Kiesner and Pastore

Mulaik, S. A., James, L. R., Van Alstine, J., Bennett, N.,Lind, S., & Stilwell, D. C. (1989). Evaluation of goodness-of-fit indices for structural equation models. Psychologi-cal Bulletin, 105, 430 – 445.

Newcomb, A. F., Bukowski, W. M., & Pattee, L. (1993).Children’s peer relations: A meta-analytic review ofpopular, rejected, neglected, controversial, and averagesociometric status. Psychological Bulletin, 113, 99 – 128.

Pettit, G. S., Bates, J. E., Dodge, K. A., & Meece, D. W.(1999). The impact of after-school peer contact on earlyadolescent externalizing problems is moderated by pa-rental monitoring, perceived neighborhood safety, andprior adjustment. Child Development, 70, 768 – 778.

Poulin, F., Cillessen, A. H. N., Hubbard, J. A., Coie, J. D.,Dodge, K. A., & Schwartz, D. (1997). Children’s friendsand behavioral similarity in two social contexts. SocialDevelopment, 6, 224 – 236.

Prinstein, M. J., & Cillessen, A. H. N. (2003). Forms andfunctions of adolescent peer aggression associated with

high levels of peer status. Merrill-Palmer Quarterly, 49,310 – 342.

Ray, G. E., Cohen, R., & Secrist, M. E. (1995). Best friendnetworks of children across settings. Child Study Journal,25, 169 – 187.

Richards, W., & Rice, R. (1981). The NEGOPY networkanalysis program. Social Networks, 3, 215 – 223.

Schneider, B. H., Fonzi, A., Tani, F., & Tomada, G. (1997). Across-cultural exploration of the stability of children’sfriendships and the predictors of their continuation.Social Development, 6, 322 – 339.

Tomada, G., & Schneider, B. H. (1997). Relational aggres-sion, gender, and peer acceptance: Invariance acrossculture, stability over time, and concordance amonginformants. Developmental Psychology, 33, 601 – 609.

Urberg, K. A., Degirmencioglu, S. M., & Pilgrim, C. (1997).Close friend and group influence on adolescent cigarettesmoking and alcohol use. Developmental Psychology, 33,834 – 844.

Antisocial Behavior and Peer Acceptance 1293