Journal of Abnormal Child Psychology

, Volume 35, Issue 4, pp 665–680

Do Friends’ Characteristics Moderate the Prospective Links between Peer Victimization and Reactive and Proactive Aggression?

Authors

    • Department of PsychologyUniversité du Québec à Montréal
  • Mara Brendgen
    • Department of PsychologyUniversité du Québec à Montréal
  • Michel Boivin
    • Department of PsychologyLaval University
  • Frank Vitaro
    • Research Unit on Children’s Psychological MaladjustmentUniversity of Montreal
  • Ginette Dionne
    • Department of PsychologyLaval University
  • Daniel Pérusse
    • Research Unit on Children’s Psychological MaladjustmentUniversity of Montreal
Article

DOI: 10.1007/s10802-007-9122-7

Cite this article as:
Lamarche, V., Brendgen, M., Boivin, M. et al. J Abnorm Child Psychol (2007) 35: 665. doi:10.1007/s10802-007-9122-7

Abstract

This study examined (a) the predictive link between peer victimization and children’s reactive and proactive aggression, and (b) the potential moderating effect of reciprocal friends’ reactive and proactive aggression in this context. The study also examined whether these potential moderating effects of friends’ characteristics were stronger with respect to more recent friends compared to previous friends. Based on a convenience sample of 658 twin children (326 boys and 332 girls) assessed in kindergarten and first grade, the results showed that peer victimization uniquely predicted an increase in children’s teacher-rated reactive aggression, but not teacher-rated proactive aggression. The relation of peer victimization to increased reactive aggression was, however, moderated by recent ¯ not previous ¯ reciprocal friends’ similarly aggressive characteristics. These findings, however, tended to be mostly true for boys, but not for girls. The findings are discussed in terms of their theoretical and practical implications for victimized children’s risk of displaying reactive and proactive aggressive behaviors.

Keywords

VictimizationFriendshipsReactive aggressionProactive aggression

Introduction

Empirical evidence suggests that approximately 10%–15% of school-aged children are identified as frequent and systematic victims of peers’ aggressive acts (Kochenderfer and Ladd 1996a; Olweus 1991). Children who are victimized by their peers are at greater risk for a wide range of social, emotional, behavioral, and academic adjustment problems, both concurrently and prospectively (e.g., Hanish and Guerra 2002; Hawker and Boulton 2000; Kochenderfer-Ladd and Ladd 2001; Schwartz et al. 1998). Although a number of studies investigating the negative effects of peer victimization on children’s adjustment have focused on different types of internalizing outcomes (e.g., Hawker and Boulton 2000), aggressive behavior as an outcome of peer victimization and potential moderating factors in this context have been largely overlooked. The present study addresses these issues by examining the predictive link between peer victimization and children’s reactive and proactive aggression, as well as the potential moderating effect of reciprocal friends’ reactive and proactive aggression in regard to this link.

To date, only few studies have tested the associations between peer victimization and children’s display of aggression as well as related externalizing problems such as disruptive and antisocial behavior. For example, a study by Schwartz et al. (Schwartz et al. 1998) showed that peer victimization was concurrently and prospectively associated with aggressive and antisocial behavior as rated by mothers and teachers in a sample of third and fourth grade children. Likewise, Hanish and Guerra’s (2002) investigation with a large community sample of boys and girls attending first through fourth grade showed that peer victimization predicted concurrent and subsequent teacher-rated aggression two years later. These authors also examined whether their findings varied as a function of sex and age and found no difference between boys and girls or between younger and older children with respect to the observed links between peer victimization and aggression. More recently, Snyder et al. (Snyder et al. 2003) investigated whether individual differences in playground observational measures of peer victimization assessed on multiple occasions throughout kindergarten and first grade were related to differences in children’s aggressive and antisocial behaviors as rated by teachers and parents over this 2-year period. Their results showed that, for boys, an increase in peer victimization across kindergarten and first grade was related to an increase in aggressive and antisocial behaviors throughout the assessed period. For girls, initial (i.e., kindergarten) peer victimization predicted an increase in aggressive and antisocial behavior over the 2-year period. Together, these findings suggest that being victimized by peers may result in children exhibiting high levels of aggression.

In the last decade, researchers have highlighted the importance of distinguishing among different subtypes of aggressive behaviors on the basis of their underlying function, namely reactive and proactive aggression (e.g., Camodeca et al. 2002; Pellegrini et al. 1999; Poulin and Boivin 2000a; Salmivalli and Nieminen 2002). Reactive aggression has been defined as affective, defensive, impulsive, and involving angry outbursts in response to an actual or perceived threat or provocation. In contrast, proactive aggression has been described as an instrumental, offensive, non-provoked, and aversive act aimed at influencing or dominating others. These two aggressive dimensions have been found to be moderately to highly correlated in previous studies, but exploratory and confirmatory factor analyses have demonstrated that reactive and proactive aggression are factorially distinct (e.g., Brown et al. 1996; Crick and Dodge 1996; Day et al. 1992; Dodge and Coie 1987; Poulin and Boivin 2000a; Salmivalli and Nieminen 2002; Vitaro and Brendgen 2005). Moreover, behavioral genetic analyses showed that the most important contribution to both reactive and proactive aggression in young children seems to come from environmental effects that are ¯ for the most part ¯ specific to each of the two types of aggression (Brendgen et al. 2006).

One important environmental influence that may distinguish between reactive and proactive aggression refers to children’s socialization experiences with peers, notably peer victimization. For example, in an investigation with preadolescent boys, Poulin and Boivin (2000a) showed that the display of reactively aggressive behavior was concurrently associated with negative social status and peer victimization. In contrast, the display of proactive aggressive behavior was not linked to either negative social status or peer victimization. In a recent study with fourth, fifth, and sixth graders, Salmivalli and Nieminen (2002) compared children in different bullying roles with respect to reactive and proactive aggression. These authors found that being a victim of peer aggression was concurrently linked to high levels of reactive, but not proactive, aggression. However, being both a bully and a victim was related to high levels of both reactive and proactive aggression. These links were true for boys and for girls. Similar findings were reported by Camodeca et al. (Camodeca et al. 2002) with a sample of younger children.

Overall, the findings reviewed above support Dodge’s (1991) suggestion that reactive aggression results from exposure to a harsh and threatening environment, such as peer victimization. In line with this reasoning, Pellegrini (1998) argued that many victimized children come to use aggression reactively (i.e., in a defensive and impulse manner) as a means to retaliate against hostile peer attacks. Pellegrini (1998) further argued, however, that some victimized children, perhaps as a result of social learning mechanisms such as modeling and reinforcement, may also become inclined to initiate proactive aggressive acts towards peers (e.g., in order to dominate or acquire resources from other ¯ and presumably weaker ¯ children). The findings in regard to proactive aggression by Camodeca et al. (2002) and Salmivalli and Nieminen (2002) indeed seem to support this latter proposition. Nevertheless, the concurrent nature of the existing data precludes any conclusions regarding the potential role that victimization may play in predicting increases in children’s reactive and proactive aggression. As such, a first question addressed by the present study was the extent to which peer victimization predicts an increase in children’s display of reactive or proactive aggression or both.

Friends’ Aggressive Characteristics as a Moderator of the Prospective Links between Peer Victimization and Children’s Reactive and Proactive Aggression

Related to the question of whether peer victimization contributes to an increase in children’s display of reactive and proactive aggression is the question of whether there are factors that may mitigate these links. Empirical evidence suggests that friendships may play an important moderating role in the link between peer victimization and social-psychological adjustment, notably aggression. For example, a study by Hodges et al. (Hodges et al. 1999) revealed that peer victimization predicted significant increases in aggressive behavior 1 year later only for children without a very best friend but not for friended children. In a related vein, Prinstein et al. (2001) found that peer victimization was associated with concurrent aggression only for adolescents who reported low social support from a close friend but not for those with high social support from a close friend. No interactive effect involving gender was found in either of these studies, suggesting that the moderating role of friendships in victimized children’s aggression is true for both boys and girls.

The presence of friendships, particularly high quality friendships, seems to have a protective effect against later aggressive behavior in victimized children. Research suggests, however, that ¯ depending on friends’ characteristics ¯ friends can also lead to increased aggressive behavior. Thus, in a study investigating single and cumulative risk factors in predicting third through seventh graders’ aggressive behavior, Kupersmidt et al. (1995) found that having aggressive friends predicted children’s subsequent aggression. Moreover, their results showed that children who are both rejected by peers and have aggressive friends are at even greater risk of becoming aggressive over time. In contrast, when children’s level of rejection and their best friends’ level of aggression decreased over time, children’s risk of exhibiting subsequent aggression also decreased. These results were consistent across boys and girls. Friends’ aggressive characteristics thus seem to compensate or exacerbate the effects of problematic social experiences such as peer rejection. However, no study so far has examined specifically whether friends’ aggressive characteristics might moderate victimized children’s risk of displaying aggressive behavior themselves. Nevertheless, in light of the findings discussed here, aggressive characteristics of a child’s friends may play an important moderating role in the link between peer victimization and subsequent reactive and proactive aggression. Moreover, distinguishing between the potential moderating effects of friends’ reactive versus proactive aggression may be important in this context, given evidence that these two aggressive subtypes are largely influenced by specific (i.e., non-overlapping) environmental influences (Brendgen et al. 2006). As such, friends’ reactive aggression may specifically moderate victimized children’s risk of subsequent reactive aggression, whereas friends’ proactive aggression may specifically moderate victimized children’s risk of subsequent proactive aggression.

Through processes such as observational learning and reinforcement of social attitudes and behaviors (Bagwell and Coie 2004; Bandura 1986; Dishion et al. 1996), victimized children whose friends’ are reactively or proactively aggressive may learn and develop similar aggressive strategies as a means to retaliate against peers or to bully other children. In line with this view, Poulin and Boivin’s (2000b) investigation with fourth through sixth grade boys showed that proactively aggressive boys were likely to affiliate with other similarly aggressive friends. The authors suggested that friendships between proactively aggressive children may create an environment that promotes and reinforces the use of proactive aggression. Since no such pattern of results was found for reactive aggression, the authors suggested that social reinforcement of aggressive behaviors by peers may not pertain to reactive aggression. Other researchers have suggested otherwise, however. For example, Prinstein and Cillessen (2003) have argued that the display of reactive aggression may be socially reinforced by peers through attitudes and behaviors such as increased attention. Relatedly, Pellegrini et al. (Pellegrini et al. 1999) have suggested that the display of reactive aggression by victimized children may be viewed by others as justified retaliation or as a legitimate means to defend against hostile peers. It thus seems plausible that having reactively or proactively aggressive friends’ exacerbates victimized children’s risk of displaying similarly aggressive behaviors. In contrast, having friends who do not display such aggressive characteristics may attenuate victimized children’s risk of displaying reactive or proactive aggression because affiliations with these friends may provide children with more skillful and adaptive strategies (e.g., problem-solving and emotion-regulation techniques) to cope with the negative consequences of being a target of peer aggression (Wentzel et al. 2004). Accordingly, a second question addressed by the present study was the extent to which friends’ reactive or proactive aggression, respectively, moderates the link between peer victimization and children’s increased display of reactive or proactive aggression.

When examining the potential moderating effect of friends’ aggressive characteristics on the links between victimization and later reactive and proactive aggression, it may also be important to assess whether this effect is stronger when more recent friends’ characteristics are considered compared to the characteristics of previous friends. Findings from a study by Brendgen et al. (2000) showed that recent friends’ antisocial behavior (including aggression) had a much stronger predictive effect on children’s own antisocial behavior than previous friends’ antisocial behavior, suggesting that social learning among children’s friendships may be particularly enhanced when socialization agents have a strong affective valence, such as may be the case with children’s most recent friends. Therefore, a third question addressed in the present study was whether the potential moderating effects of friends aggressive characteristics on the link between peer victimization and children’s later reactive or proactive aggression is stronger for recent friends compared to previous friends.

Objectives of the Present Study

To summarize, the goals of the present study were to examine (a) whether peer victimization predicts an increase in children’s display of reactive or proactive aggression or both, (b) whether friends’ reactively and proactively aggressive characteristics, respectively, moderate the potential links between peer victimization and children’s similarly aggressive behaviors, and (c) whether the potential moderating effects of friends’ aggressive characteristics in this context are stronger with respect to more recent friends compared to previous friends. It was hypothesized that peer victimization would predict an increase in children’s display of both reactive and proactive aggression. However, it was expected that these links would be moderated by friends’ similarly aggressive characteristics. Specifically, the link between peer victimization and increased reactive aggression should be strong at a high level of friends’ reactive aggression, but weak at a low level of friends’ reactive aggression. Similarly, the link between peer victimization and increased proactive aggression should be strong at a high level of friends’ proactive aggression, but weak at a low level of friends’ proactive aggression. Lastly, it was hypothesized that the moderating effects of friends’ reactive and proactive aggression on the link between children’s peer victimization and increased reactive and proactive aggression should be stronger with respect to more recent friends compared to previous friends. No gender effect was expected in these interactive links, given that previous studies have found no difference between boys and girls regarding the moderating effect of friendships on victimized children’s aggressive behavior (e.g., Hodges et al. 1999; Prinstein et al. 2001). Nevertheless, potential moderating effects involving gender were tested. Also consistent with other studies examining the moderating effect of friends (e.g., Hodges et al. 1999), and because reciprocity in children’s friendships is often considered a good indicator that the friendship really exists (Bukowski and Hoza 1989), only reciprocated friendships were considered in the present study.

The questions addressed by the present study were investigated using a convenience sample of twins. Twin sample have been used in previous studies on the effects of peer victimization on child adjustment even when genetic effects were not the focus of the research question (Arseneault et al. 2006). Importantly, empirical evidence suggests that the nature of twins’ peer relations (e.g., the number of friends and friendship quality features) does not differ from that of non-twin children (Koch 1966; Thorpe 2003). Moreover, twin samples and singleton samples do not differ with respect to social-psychological adjustment, including aggressive behavior (e.g., Pulkkinen et al. 2003). The twin sample was assessed when children were six and seven years of age (i.e., when they were in kindergarten and first grade). To date, the vast majority of studies investigating the factors that may mitigate victimized children’s risk of experiencing negative outcomes have been conducted with pre-adolescents or middle adolescents (e.g., Hodges et al. 1999; Prinstein et al. 2001). However, recent evidence suggests that the functional significance of friends for young children’s social-behavioral adjustment may be similar to that for older school-aged children (Sebanc 2003). Furthermore, in light of empirical evidence that already a substantial amount of kindergarteners are exposed to peer victimization (e.g., Kochenderfer and Ladd 1996b; Crick et al. 1999), it is imperative to identify the factors that might moderate the potential impact of peer victimization on young children’s risk of maladjustment before problems start to crystallize.

Materials and Methods

Sample

Participants for the present study were a twin sample from the greater Montreal area who were recruited at birth between November 1995 and July 1998 (N = 648 twin pairs). For the same-sex twin pairs, zygosity was assessed at 18 months based on physical resemblance via the Zygosity Questionnaire for Young Twins (Goldsmith 1991) and via DNA tests for 30% of the population for whom the zygosity questionnaire was inconclusive. Eighty-four percent of the families were of European descent, 3% were of African descent, 2% were of Asian descent, and 2% were Native North Americans. The remaining families (9%) did not provide ethnicity information. The average yearly household income (54,000 $ CAN) in the twin sample was slightly above the national average for couples with children. However, a comparison of family characteristics of this sample at 5 months of age with an epidemiological sample of singletons from the Montreal and Quebec City area (SantéQuébec et al. 1998) indicated that the samples were very similar in terms of parental education, yearly income, age of parents at birth of children, and marital status.

The sample was followed longitudinally each year with the most recent data collections completed at 6 and 7 years of age (kindergarten and grade 1). The present paper describes findings from these two latest waves of data collection. The average age of assessment at T1 and T2 were 6.04 years (0.28 SD) and 7.08 years (0.27 SD), respectively. Attrition in the sample averaged at approximately 5% per year, resulting in a total of 329 twin pairs who participated in the data collection at age 6 and 7 years, 134 monozygotic twins and 195 dizygotic twins (101 same-sex dizygotic pairs, 94 mixed-sex dizygotic pairs). Overall, there were 326 boys and 332 girls in the study sample. Participants remaining in the study at 7 years of age did not differ from those lost in regard to zygosity status, family status, parent-rated temperament, and mother’s level of education at 5 months of age. However, fathers in the remaining study sample had a slightly higher level of education than fathers of the participants who were lost from the study. In kindergarten, 30% of the two twins in a pair attended the same classroom, whereas 70% attended different classrooms (albeit in the same school). In grade one, 23% of the two twins in a pair attended the same classroom, whereas 77% attended different classrooms.

Measures and Procedure

All instruments were administered in either English or French, depending on the language spoken by the children and the teachers (see descriptions of measures below). Peer victimization was measured at T1 using group-administered peer nominations, reciprocal friendships were recorded by children and their classmates assessed at both T1 and T2 using group-administered peer nominations, and children’s and reciprocal friends’ reactive and proactive aggression were measured at both T1 and T2 using teacher reports. Following a procedure suggested by Vallerand (1989), instruments that where administered in French but were originally written in English were first translated into French and then translated back into English. Bilingual judges verified the semantic similarity between the back-translated items and the original items in the questionnaire. The research questions and instruments were approved by the IRB and by the school board administrators. Prior to data collection, active written consent from parents was obtained. Data collection took place in the spring of the kindergarten and grade one school year, respectively, to ensure that the children and teachers had gotten to know each other. The average percentages of class attendance across all schools at the time of the peer nomination procedure were 78 and 76% at T1 and T2, respectively. The peer nomination procedure took approximately 45 min per class. Children were encouraged not to share their responses with each other. In the same week, teachers completed the questionnaire for the target child and his or her nominated friends and returned them by mail.

Peer Victimization

Two items were used to assess the extent to which a child was perceived by his or her classmates as being victimized by peers at T1. These items were drawn from the seven-item victimization subscale of the Modified Peer Nomination Inventory (Perry et al. 1988). One item assessed physical victimization (“He/she gets hit and pushed by other kids”) and one item assessed verbal victimization (“He/she gets called names by other kids”). The two items were embedded within the peer nomination procedure along with other items (e.g., generalized aggression, hyperactivity, shyness) that do not directly bear on this paper. The Victimization subscale of the modified Peer Nomination Inventory has been shown to have good predictive validity and test–retest reliability (Hodges and Perry 1999; Perry et al. 1988). The two victimization items were selected because, while reflecting different forms of peer victimization, they represented a broad assessment of peer victimization. Both items showed high item-total correlations (i.e., r = 0.81, p = 0.00 for the physical victimization item and r = 0.83, p = 0.00, for the verbal victimization item). Although only two items were used, even single-item peer nomination assessments tend to demonstrate high reliability and validity in measuring behavioral constructs because the scoring of each peer nomination item is generated on the basis of multiple respondents (e.g., Coie et al. 1990; Perry et al. 1988). These measurement qualities tend to hold true for younger (i.e., preschool) as well as older (i.e., adolescent) age groups (Coie et al. 1990). In addition, these two items have been used to represent a global measure of peer victimization in previous studies with young children (i.e., first graders) (e.g., Hanish and Guerra 2000, 2002).

Booklets of photographs of all children in a given class were handed out to each child in a class. Two research assistants ensured that all children recognized the photos of all their classmates by presenting them individually. The children were then asked to circle the faces of up to three children who best fit each of the two behavioral descriptors. One booklet per behavioral descriptor was provided. The research assistants read questions aloud to the children. For each behavioral descriptor, the total number of received nominations was calculated for each child and z-standardized within each classroom to account for differences in classroom size. Following the procedure used in previous studies (e.g., Hanish and Guerra 2000, 2002), the two victimization item scores (r = 0.36, p = 0.00) were then summed up to yield a total victimization score.

Reactive and Proactive Aggression

Kindergarten teachers rated the children’s level of reactive and proactive aggression at T1 and Grade 1 teachers rated the children’s level of reactive and proactive aggression at T2 using the six items reactive–proactive measure developed by Dodge and Coie (1987). This measure has been shown to have good concurrent discriminant validity of the two types of aggression in previous studies (e.g., Dodge and Coie 1987; Poulin and Boivin 2000a). In regard to reactive aggression, the teachers indicated to what extent the child “reacts in an aggressive manner when teased,” “when somebody accidentally hurt him/her (such as by bumping into him/her), he/she reacts with anger and fighting,” and “reacts in an aggressive manner when something was taken away from him/her.” In regard to proactive aggression, the teachers indicated to what extent the child “tries to dominate the other children,” “scares other children to get what he/she wanted,” and “encourages other children to pick on a particular child.” Responses were given on a 3-point scale (0 = never, 1 = sometimes, 2 = often). For each type of aggression, the respective scores were averaged to yield a total reactive aggression score and a total proactive aggression score. Internal consistency of the total scales was acceptable in the present sample with Cronbach’s alphas for T1 and T2 reactive aggression = 0.84 and 0.85, respectively, and Cronbach’s alphas for T1 and T2 proactive aggression = 0.70 and 0.72, respectively.

Reciprocal Friendship

At T1 and T2, children and their classmates were asked to nominate up to three friends in the classroom. For twins who shared the same class, it was specified that a sibling could not be nominated as a friend. Children were considered to have a best friend if a nominated friend also nominated them among his or her three friends. In this sample, 470 children (71.4%) had at least one reciprocated friend at T1 and 479 children (72.8%) had at least one reciprocated friend at T2. Children with reciprocal friends at T1 and children with reciprocal friends at T2 did not significantly differ from those without reciprocal friends with respect to child sex, zygosity status, sex composition of the twin dyad, victimization, or mean levels of proactive aggression at T1 and T2. Children without reciprocal friends at T1 did, however, display a greater mean level of reactive aggression at T1 compared to children with reciprocal friends at T1, F (1, 327) = 3.98, p = 0.05. Similarly, there was a trend for children without reciprocal friends at T2 to display a greater mean level of reactive aggression at T2, F (1, 327) = 3.40, p = 0.06. Of the 470 children with reciprocal friends at T1, 400 (85.1%) had only same-sex friends whereas 70 (14.9%) had at least one opposite-sex friend. Of the 479 children with reciprocal friends at T2, 403 (84.1%) had only same-sex friends whereas 76 (15.9%) had at least one opposite-sex friend.

Reciprocal Friends’ Reactive and Proactive Aggression

Separately for children with at least one reciprocal friendship at T1 and for children with at least one reciprocal friend at T2, information about friends’ aggression was obtained from teachers using the same reactive and proactive aggression scales as used for the target children. Specifically, together with the questionnaire assessing the target children’s (i.e., twins’) reactive and proactive aggression, teachers were provided with the names of the target children’s nominated friends and were asked to also evaluate these friends’ reactive and proactive aggression. With respect to reactive aggression, teacher ratings for the reciprocal friends were based on an abbreviated scale, which included two items (i.e., “reacts in an aggressive manner when teased” and “when somebody accidentally hurt him/her, he/she reacts with anger and fighting”). The reduction in the number of items was necessary to reduce teacher’s workload and avoid the risk of nonparticipation. With respect to proactive aggression, teacher ratings were based on the same three items as were used for the target children (i.e., “tries to dominate the other children,” “scares other children to get what he/she wanted,” and “encourages other children to pick on a particular child”). Total friends’ reactive and proactive aggression scores were determined by calculating the mean across the two reactive aggression items and the mean across the three proactive aggression items, respectively, and then averaging across a child’s friends’ scores. Internal consistency of the friends’ total aggression scales was acceptable with Cronbach’s alphas for T1 and T2 reactive aggression = 0.83 and 0.85, respectively, and Cronbach’s alphas for T1 and T2 proactive aggression = 0.78 and 0.80, respectively.

Results

Preliminary Analyses

Preliminary analyses had shown no significant link between zygosity status (i.e., monozygotic twins versus dizygotic twins) and any of the study variables. There was also no significant effect of the two twins’ being in the same classroom or not at T1 or T2 on children’s reactive and proactive aggression at T1 or T2, respectively. As a consequence, zygosity status and the fact of being in the same classroom or not were not included in subsequent analyses.

Table 1 presents the bivariate intraclass correlations as well as the means and standard deviations for all measures for the whole sample. As can be seen, child sex was significantly related to peer victimization at T1, indicating that boys were more victimized than girls. Child sex was also related to T1 reactive aggression, but not to T1 proactive aggression. Similarly, child sex was related to T2 reactive aggression, but not to T2 proactive aggression, indicating that boys were more reactively aggressive than girls, whereas girls were just as likely to be proactively aggressive as boys at both T1 and T2. Peer victimization at T1 was positively related to child’s reactive and proactive aggression at T1 and T2, respectively. Reactive and proactive aggression were moderately correlated at both T1 and T2, thus replicating findings from previous studies (e.g., Dodge and Coie 1987; Poulin and Boivin 2000a). T1 reactive aggression was positively related to both reactive and proactive aggression at T2. Similarly, T1 proactive aggression was related to both types of aggression at T2.
Table 1

Bivariate intraclass correlations among measures for the whole sample (n = 329)

Measures

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

1. Child Sexa

         

2. T1 Child Peer Victimization

−0.22****

        

3. T1 Child Reactive Aggression

−0.16***

0.38****

       

4. T1 Child Proactive Aggression

−0.03

0.26****

0.60****

      

5. T2 Child Reactive Aggression

−0.22****

0.29****

0.42****

0.26****

     

6. T2 Child Proactive Aggression

−0.07

0.22****

0.33****

0.32****

0.64****

    

7. T1 Friends Reactive Aggressionb

−.13*

−0.05

0.22***

0.19***

0.13*

0.12

   

8. T1 Friends Proactive Aggressionb

−0.06

−0.02

0.13*

0.25****

0.04

0.06

0.71****

  

9. T2 Friends Reactive Aggressionc

−0.25****

0.00

−0.00

−0.07

0.22***

0.08

0.10

0.09

 

10. T2 Friends Proactive Aggressionc

−0.10

−0.01

0.00

−0.06

0.19***

0.16**

0.13

0.14

0.69****

M

50%

−0.09

0.36

0.22

0.34

0.21

0.41

0.33

0.39

0.32

SD

 

0.87

0.50

0.36

0.51

0.37

0.49

0.39

0.48

0.40

aSex is coded so that a higher value (1) represents girls. The relative percentage of boys in the sample is given instead of a mean parameter.

bn = 235 for correlations involving friends’ aggressive characteristics at T1.

cn = 240 for correlations involving friends’ aggressive characteristics at T2.

*p < 0.10, **p < 0.05, ***p < 0.01, ****p < 0.001.

Also, child’s reactive aggression at T1 and friends’ reactive aggression at T1 were positively related. Similarly, there was a significant positive correlation between child’s and friends’ reactive aggression at T2. Child’s proactive aggression at T1 and friends’ proactive aggression at T1 were positively related. Similarly, there was a significant positive correlation between child’s and friends’ proactive aggression at T2. Lastly, friends’ reactive and proactive aggression at T1 were positively correlated as were friends’ reactive and proactive aggression at T2, respectively. Bivariate intraclass correlations were also computed separately for boys and for girls. These correlations were then transformed into Fisher-z scores and corresponding correlations were compared across sex via z-tests. None of these correlations were significantly different for boys than for girls.

Analytical Rationale

Multilevel modeling with the PRELIS 8.52 statistical package (Jöreskog et al. 2000) was used for the analysis of our hierarchically structured data. In a two-level model, a hierarchy consists of lower-level observations (i.e., level 1 unit of analysis) nested within higher-level observations (i.e., level 2 unit of analysis). In the context of the present study, each individual child is nested within a sibling pair. It is assumed that observations across pairs are independent from one another. However, because siblings within a given pair share genetic and environmental factors, observa-tions within a given pair are interdependent, thus violating the assumption of independent observations in traditional linear models. Multilevel models allow for the estimation of within-pair and between-pair effects while simultaneously adjusting for the amount of data interdependency. In the present study, the level 1 unit of analysis represents each individual child, whereas the level 2 unit of analysis represents each individual sibling pair. The level 1 variance estimates describe the degree to which siblings within a pair differ from each other (i.e., within-pair variance), whereas the level 2 variance estimates indicate the degree to which sibling pairs differ from one another (i.e., between-pair variance) with respect to the dependent variable. Child-specific predictors (i.e., independent variables) are included in multi-level analyses as fixed effects. The fixed effect estimates provide information about the unique link between each child-specific predictor and the dependent variable and can be interpreted in a similar way as regression coefficients in a multiple regression.

Using multilevel modeling, a series of consecutive models of increasing complexity were fitted to the data to examine the predictive effect of T1 peer victimization on T2 reactive and proactive aggression, above and beyond T1 levels of reactive and proactive aggression, and the moderating effect of reciprocal friends’ characteristics in this context (i.e., reactive and proactive aggression). Each model was compared to the preceding one to evaluate whether the inclusion of additional predictors provided a better fit to the data. Goodness of fit for each model was evaluated based on the −2log likelihood estimate and a likelihood ratio test was used to evaluate the difference in fit between subsequent models. The first series of models examined the moderating effect of T1 (i.e., previous) friends’ characteristics and the second series of models examined the moderating effect of T2 (i.e., recent) friends’ characteristics. Notably, although level 2 variance estimates (i.e., relating to between-pair differences) are provided in the tables, they are not of theoretical interest in the present study and are therefore not described in the text.

Moderating Effects of T1 (Previous) Friends’ Characteristics on the Link between T1 Peer Victimization and T2 Reactive and Proactive Aggression

Predictions to T2 Reactive Aggression

Table 2 presents the results from the multilevel analyses predicting from T1 peer victimization to T2 reactive aggression. For each model, the fixed effects of the predictor variables, the level 1 and level 2 variance parameters, the model fit (i.e., −2log likelihood), and the likelihood ratio are provided. The first model tested was an unconditional model, without including any predictors. This model was used as a baseline for comparing subsequent (i.e., more complex) models. In a second model, child-specific predictors were added to the equation as fixed effects. These predictors included Child Sex, T1 Child Reactive Aggression, T1 Child Proactive Aggression, T1 Child Peer Victimization, T1 Reciprocal Friends Reactive Aggression, and T1 Reciprocal Friends Proactive Aggression. Controlling for child’s T1 proactive aggression, in addition to controlling for child’s T1 reactive aggression, was important given the significant correlation between the two aggressive subtypes. To facilitate the interpretation of the results, the dependent variable and all continuous independent variables were z-standardized and the z-standardized variables were used in the analysis. As indicated by the likelihood ratio test shown in Table 2, the addition of the six predictors in model 2 significantly improved model fit compared to the unconditional model, p = 0.00. The results showed a main effect of sex on child’s reactive aggression at T2, b = −0.23, p = 0.01, indicating that boys displayed higher levels of reactive aggression at T2 compared to girls. In addition, high levels of reactive aggression at T1 predicted high levels of reactive aggression at T2, b = 0.35, p = 0.00, whereas there was no link between proactive aggression at T1 and reactive aggression at T2, b = 0.02, p = 0.63. There was a statistical trend for an effect of peer victimization at T1 on child’s reactive aggression at T2, b = 0.07, p = 0.09, over and above the effects of child’s reactive and proactive aggression at T1. There was no effect of reciprocal friends’ reactive aggression at T1, b = 0.09, p = 0.12, and no effect of reciprocal friends’ proactive aggression at T1 on child’s reactive aggression at T2, b = −0.06, p = 0.28.
Table 2

Multilevel analyses with victimization and friends’ reactive aggression at T1 predicting child reactive aggression at T2 (n = 470)

Step

Predictor

Fixed effect (se)

Level 1 variance (se)

Level 2 variance (se)

−2log likelihood (df)

ΔLikelihood ratio (df)

1

  

0.60 (0.06)

0.40 (0.08)

1,304.87 (3)

 

2

  

0.59 (0.06)

0.19 (0.06)

1,206.70 (9)

98.17**** (6)

Child Sexa

−0.23*** (0.09)

    

T1 Child Reactive Aggression

0.35**** (0.05)

    

T1 Child Proactive Aggression

0.02 (0.05)

    

T1 Child Peer Victimization

0.07* (0.04)

    

T1 Friends Reactive Aggression

0.09 (0.06)

    

T1 Friends Proactive Aggression

−0.06 (0.06)

    

3

  

0.59 (0.06)

0.19 (0.06)

1,205.73 (10)

0.97 (1)

T1 Child Peer Victimization X

     

T1 Friends Reactive Aggression

−0.04 (0.05)

    

4

  

0.59 (0.06)

0.17 (0.06)

1,198.80 (13)

6.93* (3)

T1 Child Peer Victimization X Sex

−0.20** (0.09)

    

T1 Friends Reactive Aggression X Sex

−0.07 (0.09)

    

T1 Child Peer Victimization X

0.07 (0.11)

    

T1 Friends Reactive Aggression X Sex

     

aSex is coded so that a higher value (1) represents girls.

Δ = Difference of model fit between consecutive models. Each model is tested against the respective preceding model.

*p < 0.10, **p ≤ 0.05, ***p < 0.01, ****p < 0.001.

In a third step, it was examined whether friends’ reactive aggression at T1 moderated the effect of T1 peer victimization on child’s reactive aggression at T2. For this purpose, a two-way interaction term “T1 Child Peer Victimization X T1 Reciprocal Friends Reactive Aggression” was added to the preceding model. This two-way interaction term did not reach statistical significance, b = −0.04, p = 0.32, suggesting that peer victimization marginally predicted child’s reactive aggression at T2 regardless of T1 friends’ reactive aggression.

Test of Moderating Effect of Sex

In a final step, it was examined whether the main effect of T1 peer victimization on child’s reactive aggression at T2, as well as the moderating effect of T1 reciprocal friends’ reactive aggression in this context, differed as a function of child’s sex. For this purpose, two additional two-way interaction terms “T1 Child Peer Victimization X Child Sex” and “T1 Reciprocal Friends Reactive Aggression X Child Sex” and a three-way interaction term “T1 Child Peer Victimization X T1 Reciprocal Friends Reactive Aggression X Child Sex” were added to the preceding model. The addition of these three interaction terms marginally improved model fit compared to the preceding model, p = 0.07. No significant effect of the three-way interaction “T1 Child Peer Victimization X T1 Reciprocal Friends Reactive Aggression X Child Sex” was found, b = 0.07, p = 0.54. There was, however, a significant interaction between T1 victimization and sex, b = −0.20, p = 0.03. Examination of this two-way interaction (Jaccard et al. 1990) revealed that T1 victimization predicted T2 reactive aggression, over and above the effects of T1 reactive and proactive aggression in boys, b = 0.15, p = 0.01, but not in girls, b = −0.05, p = 0.50.

Predictions to T2 Proactive Aggression

Table 3 presents the results from the multilevel analyses predicting from T1 peer victimization to T2 proactive aggression, over and above the effects of T1 reactive and proactive aggression. The first model tested was an unconditional model, which was used as a baseline for comparing subsequent models. In a second model, child-specific predictors (i.e., Child Sex, T1 Child Reactive Aggression, T1 Child Proactive Aggression, T1 Child Victimization, T1 Reciprocal Friends Reactive Aggression, and T1 Reciprocal Friends Proactive Aggression), were added to the equation as fixed effects. As shown, the addition of the six predictors significantly improved model fit compared to the unconditional model, p = 0.00. The results indicated no difference between boys and girls in proactive aggression at T2, b = −0.05, p < 0.61. Also, child’s victimization at T1 did not predict child’s proactive aggression at T2, b = 0.07, p < 0.14, once the effects of child’s reactive and proactive aggression at T1 were controlled. There were, however, main effects of both T1 child’s reactive aggression, b = 0.19, p = 0.00 and T1 child’s proactive aggression on child’s proactive aggression at T2, b = 0.18, p = 0.00. In addition, the results showed a statistical trend for T1 reciprocal friends’ reactive aggression on child’s proactive aggression at T2, b = 0.12, p = 0.06, although no effect of T1 reciprocal friends’ proactive aggression on child’s proactive aggression at T2 was revealed, b = −0.08, p = 0.19.
Table 3

Multilevel analyses with victimization and friends’ proactive aggression at T1 predicting child proactive aggression at T2 (n = 470)

Step

Predictor

Fixed effect (se)

Level 1 variance (se)

Level 2 variance (se)

−2log likelihood (df)

ΔLikelihood ratio (df)

1

  

0.69 (0.07)

0.32 (0.08)

1,318.81 (3)

 

2

  

0.67 (0.07)

0.18 (0.06)

1,248.33 (9)

70.48**** (6)

Child Sexa

−0.05 (0.09)

    

T1 Child Reactive Aggression

0.19*** (0.06)

    

T1 Child Proactive Aggression

0.18**** (0.05)

    

T1 Child Peer Victimization

0.07 (0.05)

    

T1 Friends Reactive Aggression

0.12* (0.06)

    

T1 Friends Proactive Aggression

−0.08 (0.06)

    

3

  

0.67 (0.07)

0.18 (0.06)

1,248.15 (10)

0.18 (1)

T1 Child Peer Victimization X

0.02 (0.04)

    

T1 Friends Proactive Aggression

     

4

  

0.66 (0.07)

0.18 (0.06)

1,246.44 (13)

1.71 (3)

T1 Child Peer Victimization X Sex

0.02 (0.09)

    

T1 Friends Proactive Aggression X Sex

−0.11 (0.09)

    

T1 Child Peer Victimization X T1 Friends Proactive Aggression X Sex

0.02 (0.10)

    

aSex is coded so that a higher value (1) represents girls. Δ = Difference of model fit between consecutive models. Each model is tested against the respective preceding model.

*p < 0.10, **p ≤ 0.05, ***p < 0.01, ****p < 0.001.

In a third step, it was examined whether friends’ proactive aggression at T1 moderated the effect of T1 peer victimization on child’s proactive aggression at T2. For this purpose, a two-way interaction term “T1 Child Peer Victimization X T1 Reciprocal Friends Proactive Aggression” was added to the preceding model. This interaction term was not significant, b = 0.02, p = 0.67.

Test of Moderating Effects of Sex

In a final step, it was examined whether the main effect of T1 peer victimization on child’s proactive aggression at T2 or the moderating effect of T1 friends’ proactive aggression in this context differed as a function of child’s sex. No moderating effects of child sex were found.

Moderating Effects of T2 (Recent) Friends’ Characteristics on the Link between T1 Peer Victimization and T2 Reactive and Proactive Aggression

Predictions to T2 Reactive Aggression

Table 4 presents the results from the multilevel analyses with T1 peer victimization predicting to T2 reactive aggression, over and above the effects of T1 reactive and proactive aggression. The first model tested was an unconditional model. In a second model, child-specific predictors (i.e., Child Sex, T1 Child Reactive Aggression, T1 Child Proactive Aggression, T1 Child Victimization, T2 Reciprocal Friends Reactive Aggression, and T2 Reciprocal Friends Proactive Aggression), were added to the equation as fixed effects. As can be seen in Table 4, the addition of the six predictors significantly improved model fit compared to the unconditional model, p = 0.00. Similar results as those described earlier with respect to the predictions of T2 reactive aggression were found. Specifically, there was a main effect of sex on child’s reactive aggression at T2, b = −0.29, p = 0.00, indicating that boys displayed higher levels of reactive aggression at T2 compared to girls. Also, high levels of reactive aggression at T1 predicted high levels of reactive aggression at T2, b = 0.39, p = 0.00, whereas there was no link between proactive aggression at T1 and reactive aggression at T2, b = −0.05, p = 0.36. In addition, there was a main effect of T1 peer victimization on child’s reactive aggression at T2, over and above the effects child’s reactive and proactive aggression at T1, b = 0.12, p = 0.00. Lastly, the results showed a main effect of reciprocal friends’ reactive aggression at T2 on child’s reactive aggression at T2, b = 0.12, p = 0.02. In contrast, no effect of reciprocal friends’ proactive aggression at T2 on child’s reactive aggression at T2 was found, b = 0.08, p = 0.12.
Table 4

Multilevel analyses with victimization and friends’ reactive aggression at T2 predicting child reactive aggression at T2 (n = 479)

Step

Predictor

Fixed effect (se)

Level 1 variance (se)

Level 2 variance (se)

−2log likelihood (df)

ΔLikelihood ratio (df)

1

  

0.63 (0.06)

0.36 (0.07)

1,328.74 (3)

 

2

  

0.60 (0.06)

0.11 (0.05)

1,193.18 (9)

135.56**** (6)

Child Sexa

−0.29**** (0.08)

    

T1 Child Reactive Aggression

0.39**** (0.05)

    

T1 Child Proactive Aggression

−0.05 (.05)

    

T1 Child Peer Victimization

0.12*** (0.04)

    

T2 Friends Reactive Aggression

0.12** (0.05)

    

T2 Friends Proactive Aggression

0.08 (0.05)

    

3

  

0.59 (0.06)

0.11 (0.05)

1,187.32 (10)

5.86** (1)

T1 Child Peer Victimization X

0.09** (0.04)

    

T2 Friends Reactive Aggression

     

4

  

0.58 (.06)

0.11 (0.05)

1,180.13 (13)

7.19*(3)

T1 Child Peer Victimization X Sex

−0.17** (0.08)

    

T2 Friends Reactive Aggression X Sex

−0.09 (0.08)

    

T1 Child Peer Victimization X

−0.15*(0.08)

    

T2 Friends Reactive Aggression X Sex

     

aSex is coded so that a higher value (1) represents girls. Δ = Difference of model fit between consecutive models. Each model is tested against the respective preceding model.

*p < 0.10. **p ≤ 0.05. ***p < 0.01. ****p < 0.001.

In a third step, it was examined whether friends’ reactive aggression at T2 moderated the effect of T1 peer victimization on child’s reactive aggression at T2. For this purpose, a two-way interaction term “T1 Child Peer Victimization X T2 Reciprocal Friends Reactive Aggression” was added to the preceding model. The inclusion of this interaction term significantly improved model fit compared to the previous model, p = 0.02. Moreover, the results revealed a significant interaction effect, b = 0.09, p = 0.01, suggesting that the link between T1 peer victimization and child’s reactive aggression at T2 varied as a function of reciprocal friends’ reactive aggression at T2.

Test of Moderating Effects of Sex

In a final step, it was examined whether the main effect of T1 victimization on child’s reactive aggression at T2, and the moderating effect of T2 friends’ reactive aggression in this context differed as a function of child’s sex. The results indicated a significant interaction effect between T1 child peer victimization and child sex, b = −0.17, p = 0.04. There was also a statistical trend for a three-way interaction between T1 child peer victimization, T2 friends’ reactive aggression, and child sex, b = −0.15, p = 0.08.

The two-way interaction between T1 victimization and child sex was examined first. The results revealed that for boys, T1 peer victimization predicted T2 reactive aggression, over and above the effects of T1 reactive and proactive aggression, b = 0.16, p = 0.00. In contrast, for girls, there was no significant link between T1 peer victimization and T2 reactive aggression once the effects of T1 reactive and proactive aggression were controlled, b = −0.01, p = 0.91.

Second, the three-way interaction between T1 child peer victimization, T2 friends reactive aggression, and child sex was broken down. The results showed a significant interactive effect between T1 peer victimization and T2 friends reactive aggression for boys, b = 0.11, p = 0.01, but not for girls, b = −0.04, p = 0.61. To break down the significant two-way interaction in boys, the relation of T1 child’s peer victimization to T2 child’s reactive aggression was examined at three levels of T2 friends’ reactive aggression: high (= 1 SD above the mean), medium (= at the mean), and low (= 1 SD below the mean). The results revealed that T1 peer victimization significantly predicted boys’ reactive aggression at T2 when T2 friends’ reactive aggression was moderate, b = 0.16, p = 0.00. When T2 friends’ reactive aggression increased by one standard deviation (i.e., when T2 friends’ reactive aggression was high) the relation of T1 peer victimization to boys’ reactive aggression at T2 was stronger, b = 0.27, p = 0.00. However, when T2 friends’ reactive aggression decreased by one standard deviation (i.e., when T2 friends’ reactive aggression was low) the relation of T1 peer victimization to boys’ reactive aggression at T2 was close to zero and no-longer statistically significant, b = 0.05, p = 0.48.

Predictions to T2 Proactive Aggression

Table 5 presents the results from the multilevel analyses predicting from T1 peer victimization to T2 proactive aggression, over and above the effects of T1 reactive and proactive aggression. The first model tested was an unconditional model. In a second model, child-specific predictors (i.e., Child Sex, T1 Child Reactive Aggression, T1 Child Proactive Aggression, T1 Child Victimization, T2 Reciprocal Friends Reactive Aggression, and T2 Reciprocal Friends Proactive Aggression), were added to the equation as fixed effects. As shown, the addition of the six predictors significantly improved model fit compared to the unconditional model, p = 0.00. The results revealed no difference between boys and girls with respect to proactive aggression at T2, b = −0.07, p = 0.42. Also, there was no effect of T1 peer victimization on child’s proactive aggression at T2, once the effects of child’s reactive and proactive aggression at T1 were controlled, b = 0.06, p = 0.15. There were, however, main effects of both T1 child’s reactive aggression, b = 0.24, p = 0.00 and T1 child’s proactive aggression on child’s proactive aggression at T2, b = 0.15, p = 0.01. In addition, there was a main effect of T2 reciprocal friends’ proactive aggression on child’s proactive aggression at T2, b = 0.19, p = 0.00, whereas no effect of T2 reciprocal friends’ reactive aggression on child’s proactive aggression at T2 was revealed, b = −0.05, p = 0.40.
Table 5

Multilevel analyses with victimization and friends’ proactive aggression at T2 predicting child proactive aggression at T2 (n = 479)

Step

Predictor

Fixed effect (se)

Level 1 variance (se)

Level 2 variance (se)

−2log likelihood (df)

ΔLikelihood ratio (df)

1

  

0.70 (0.07)

0.30 (0.07)

1,341.10 (3)

 

2

  

0.66 (0.07)

0.15 (0.06)

1,251.70 (9)

89.40**** (6)

Child Sexa

−0.07 (0.09)

    

T1 Child Reactive Aggression

0.24**** (0.06)

    

T1 Child Proactive Aggression

0.15*** (0.05)

    

T1 Child Peer Victimization

0.06 (0.04)

    

T2 Friends Reactive Aggression

−0.05 (0.06)

    

T2 Friends Proactive Aggression

0.19**** (0.06)

    

3

  

0.66 (0.07)

0.15 (0.06)

1,251.65 (10)

0.05 (1)

T1 Child Peer Victimization X

0.01 (0.04)

    

T2 Friends Proactive Aggression

     

4

  

0.66 (0.07)

0.13 (0.06)

1,244.74 (13)

6.91* (3)

T1 Child Peer Victimization X Sex

0.05 (0.09)

    

T2 Friends Proactive Aggression X Sex

−0.14 (0.09)

    

T1 Child Peer Victimization X

0.16* (0.09)

    

T2 Friends Proactive Aggression X Sex

     

aSex is coded so that a higher value (1) represents girls. Δ = Difference of model fit between consecutive models. Each model is tested against the respective preceding model.

*p < 0.10. **p ≤ 0.05. ***p < 0.01. ****p < 0.001.

In a third step, it was examined whether reciprocal friends’ proactively aggressive characteristics at T2 moderated the effect of T1 peer victimization on child’s proactive aggression at T2. For this purpose, a two-way interaction term “T1 Child Peer Victimization X T2 Reciprocal Friends Proactive Aggression” was added to the preceding model. This interaction term was not significant, b = 0.01, p = 0.84.

Test of Moderating Effects of Sex

In a final step, potential sex differences in the main effect of T1 peer victimization on child’s proactive aggression at T2, and the moderating effect of T2 friends’ proactive aggression in this context were examined. No significant two-way interaction was found, although the results did reveal a significant trend for the three-way interaction between T1 child peer victimization, T2 friends proactive aggression, and child sex, b = −0.16, p = 0.07. However, the break down of this interaction did not indicate any distinct pattern between boys and girls.

Discussion

The present study examined whether peer victimization is related to an increase in children’s reactive and proactive aggression, respectively, and whether reciprocal friends’ similarly aggressive characteristics moderate these links. It was also examined whether the potential moderating effects of friends’ aggressive characteristics in this context are stronger for more recent friends compared to previous friends.

Effect of Peer Victimization on Reactive and Proactive Aggression and Friends’ Moderating Effect in this Context

As expected, the results of the present study indicated that peer victimization predicted children’s increased display of reactive aggression, above and beyond their previous levels of reactive and proactive aggression. This result supports Dodge’s proposition (1991) and extends previous findings on the concurrent link between peer victimization and reactive aggression (e.g., Camodeca et al. 2002; Poulin and Boivin 2000a; Salmivalli and Nieminen 2002) as it suggests that peer victimization may play a unique role in the development of reactive aggression in children. Being victimized by peers may lead to children exhibiting increased levels of reactive aggression as a means to retaliate or defend themselves against hostile peer attacks. The link between peer victimization and increased reactive aggression was, however, moderated by reciprocal friends’ similarly aggressive characteristics. Specifically and in line with expectations, when children had friends who showed moderate to high levels of reactive aggression, peer victimization predicted an increase in children’s reactive aggression. However, when children had friends who displayed a low level of reactive aggression, peer victimization no longer predicted an increase in reactive aggression over time. This finding lends support to the notion that the use of reactively aggressive behaviors in response to peer victimization may be socially reinforced by other children, thus increasing the display of such behaviors over time (e.g., Pellegrini et al. 1999; Prinstein and Cillessen 2003). This may be especially the case when children have friends’ who display similarly reactive aggressive behaviors because such friends may be particularly likely to reinforce or model such aggressive behaviors. Conversely, victimized children’s inclination to use reactive aggression as a response to peer provocations may be reduced when children have friends who do not display such reactively aggressive characteristics because these friends may provide more skillful and adaptive strategies (e.g., problem-solving and emotion-regulation techniques) to cope with the negative consequences of being a target of peer victimization. Some, albeit indirect support for this notion are provided by findings that children who affiliate with highly prosocial friends are likely to display higher level of prosocial behaviors themselves over time (Wentzel et al. 2004). In contrast, children who affiliate with lowly prosocial friends tend to display lower levels of prosocial behaviors themselves over time.

Importantly, the moderating effect of friends’ reactively aggressive characteristics on the link between peer victimization and increased reactive aggression was only found with respect to most recent friends. This finding is in line with results reported by Brendgen et al. (Brendgen et al. 2000), who found that the predictive effect of delinquent friends on adolescents’ own aggressive behavior was more pronounced for most recent friends than for previous friends. As argued by Brendgen et al. (Brendgen et al. 2000), social learning among children’s friendships may be particularly enhanced in a child’s most recent friendships, which may have a specifically strong affective valence for children.

The findings described above were true only for boys, however. For girls, peer victimization was not significantly linked to increased reactive aggression, after controlling for their previous levels of reactive and proactive aggression. Moreover, there was a statistical trend indicating that the interactive effect between peer victimization and recent friends’ reactive aggression was true for boys but not for girls. One explanation for these findings may lie in the fact that the reactive aggression items used in the present study captured more physical than relational forms of aggression, with the former being more prevalent in boys than in girls. Yet another explanation may be related to the exclusive focus on overt forms of peer victimization (i.e., verbal and physical) examined in the present study. Some evidence suggests that covert (i.e, relational) forms of peer victimization may be more prevalent among girls and that these forms of peer victimization may have unique detrimental adjustment consequences, especially for girls (Crick et al. 1999, 2001). Thus, it will be important for future studies to investigate potential links between covert forms of peer victimization and children’s subsequent display of reactive aggression as well as potential moderating factors in this context.

In contrast to the findings for reactive aggression, peer victimization did not predict an increase in children’s proactive aggression, once their previous levels of reactive and proactive aggression were accounted for. Moreover, no moderating effect of recent or previous friends’ proactively aggressive characteristics on victimized children’s increase in proactive aggression was found. Although previous evidence has shown that, in some cases, peer victimization may be concurrently linked to proactive aggression (e.g., Camodeca et al. 2002; Salmivalli and Nieminen 2002), none of these studies has examined this link prospectively. The results of this study are the first to show that the relation of peer victimization to children’s later aggression may pertain only to reactive ¯ but not to proactive ¯ aggression. Hence, in accord with Dodge’s theoretical model (1991), it seems that peer victimization does not necessarily incite children to adopt proactively aggressive behaviors (e.g., in order to bully or dominate other ¯ and presumably weaker ¯ children themselves). This result could also be due to the relatively young age of the children in the present sample, however, as children in first grade may not easily find many younger (and thus weaker) victims to bully. More research comparing different age groups is thus necessary before any firm conclusion can be drawn in regard to the potential link between peer victimization and increase in children’s proactive aggression.

Additional Main Effect of Friends’ Aggression on Children’s Reactive and Proactive Aggression

Although the main research question pertained to the moderating effect of friends’ reactively and proactively aggressive characteristics, it is worth mentioning that recent friends’ proactively or reactively aggressive characteristics also had a main effect on children’s similarly aggressive behavior, apart from any interactive effect with victimization. Specifically, having recent friends who display high levels of reactive aggression predicted an increase ¯ albeit weakly ¯ in children’s reactive aggression. Similarly, having recent friends’ who displayed proactively aggressive characteristics predicted an increase in children’s proactive aggression. This context-specific effect of friends’ proactively and reactively aggressive characteristics is in line with the finding from behavioral genetic research that the two types of aggression are subject to largely different environmental influences (Brendgen et al. 2006). By the same token, the context-specific effect of friends’ aggressive characteristics is discordant with Poulin and Boivin’s argument (2000b) that the modeling and reinforcement of aggressive behaviors in children’s friendships pertains only to proactive aggression, but not to reactive aggression. It is important to note, however, that Poulin and Boivin’s findings were based on data from a sample of older school-aged boys only, which may not necessarily generalize to samples of younger boys and girls. Hence, based on the present data, it appears that friends’ aggressive characteristics may play a role in the development of both children’s proactive and reactive aggression. Modeling of proactive aggression as a successful instrumental means to dominate others or to obtain desired resources may explain the main effect of friends’ proactive aggression on children’s own proactive aggression. As argued previously, however, the use of reactively aggressive behavior ¯ such as in response to peer maltreatment ¯ may also be socially reinforced or modeled by friends with similar characteristics, who may perceive such behavior as a justified defense, thereby increasing the display of such behaviors over time.

Strengths, Limitations, and Conclusions

This study is the first to examine the links between peer victimization and children’s reactive and proactive aggression using a longitudinal perspective. In addition, this study is the first to investigate the unique and interactive effects of friends’ reactive and proactive aggression on child’s own aggression. Other strengths of this study include the ability to look at both recent and past children’s friendships as well as the use of multiple informants (i.e., peer-reports and teacher-reports), which minimizes the possibility that the findings were due to shared method variance.

Despite these strengths, several limitations of the study deserve consideration, which may influence the interpretation of the present results. A first limitation is the possibility that the findings derived from the present convenience sample of twins may not necessarily generalize to the general population including singletons. Given that children’s adjustment problems were rated by teachers, who only evaluated the target children and their friends, it was not possible to examine potential adjustment differences between the twins and other classmates. Nevertheless, other studies have shown no difference between twin samples and singleton samples with respect to social-psychological adjustment (e.g., Pulkkinen et al. 2003). The average level of peer-rated victimization for the children in this sample, however, was somewhat lower than their respective classroom mean. This finding may be explained by the fact that the presence of a twin sibling in the same school, similar to the presence of a friend (e.g., Hodges et al. 1999), can in itself decrease the likelihood of being a target of peer aggression. This particularity of the twin sample may also explain why friendless children did not differ from friended children on the level of victimization in the present study. It should be kept in mind, however, that most of the patterns found in this study are comparable to those observed in other studies based on singleton samples. Importantly, empirical evidence also suggests that the nature of twins’ peer relations (e.g., number of friends) does not differ from that of non-twin children (Koch 1966; Thorpe 2003). The developmental significance of twins’ peer relationships has not been systematically investigated, however. Thus, the extent to which friendship relations have a similar impact on twin children’s social-psychological adjustment relative to their non-twin counterparts remains unknown.

Second, the limited number of children who had reciprocal friends of the opposite sex in the present study prevented the exploration of the possibility that the sex composition of the friendship dyad (i.e., same-sex versus mixed-sex friendship dyads) might influence the expected moderating effects of friends’ characteristics on the link between peer victimization and aggression. Given that research on the dyadic sex composition in children’s friendships has consistently shown that cross-sex affiliations become increasingly less frequent over the course of middle childhood (e.g., Maccoby 1988), it is possible that the pattern of results found in the present study does not generalize to older mixed-sex friendship dyads. Finally, replication is needed in order to test the generalizability of the present findings to children in other age groups.

Despite these limitations, the present study offers an important new perspective on the prospective links between peer victimization and children’s display of reactive and proactive aggression. The finding that peer victimization predicted children’s later display of reactive ¯ but not proactive ¯ aggression supports Dodge’s theoretical model (1991) and further highlights the importance of distinguishing between these two aggressive subtypes. Moreover, the present results provide support for the notion that friends’ aggressive characteristics (or lack thereof) may play an important role in fostering or preventing aggressive behavior problems related to being victimized. This finding may also have practical implications, as it suggests that preventive intervention with victimized children should discourage affiliation with aggressive friends and rather encourage the development and maintenance of friendships with peers who possess characteristics that are most likely to convey protection from maladjustment.

Acknowledgement

We wish to thank the participating families, and the authorities and directors as well as the teachers of the participating schools.

Copyright information

© Springer Science+Business Media, LLC 2007