Compared with friendships during fifth grade, adolescent friendships tend to be based more on

School-based studies have a long history in peer research, dating back more than two decades, and they have provided ample support for peer homophily. In an early study examining similarity in drug use among adolescent friends, Kandel (1978) studied reciprocated school-based best friendships and found correlations ranging from r=0.12 for the use of inhalants to r=0.46 for marijuana use. Since then, many studies have replicated the peer homophily findings on a variety of measures. For example, during adolescence, homophily has been found for sports activities, delinquency, and substance use (Urberg et al., 1998), prejudice towards stigmatized ethnic groups, but not non-stigmatized ethnic groups (Kiesner et al., 2003), general antisocial behaviour (Dishion, Andrews, & Crosby, 1995; Kiesner et al., 2002), aggressive behaviour (Cairns, Cairns, Neckerman, Gest, & Gariépy, 1988), and depressive symptoms (Hogue & Steinberg, 1995). Considering that peer homophily has been replicated many times, across samples and across behaviours, there is no doubt that adolescents select in-school friends who are behaviourally similar. However, because most of this research has focused on school-based friendships, little is known about peer homophily and the risks associated with it outside of the school context.

A growing literature suggests that although school-based studies capture the majority of adolescents’ peer relationships (Ennett & Bauman, 1994), the minority that is not captured might be important for understanding antisocial peer influence. For example, in Dishion's work on deviancy training (Dishion et al., 1996; Dishion & Owen, 2002), each individual was allowed to nominate (and bring to the lab for observation) a friend from any context. The results showed that antisocial friendship dyads were more likely to live in the same neighbourhoods and to have met in unstructured and unsupervised settings, as compared with non-antisocial friendship dyads (Dishion et al., 1995). In another line of research, several studies have shown that both boys and girls who start attending unstructured neighbourhood youth recreation centres show increases in antisocial behaviour, and that this cannot be explained by their backgrounds or prior behavioural characteristics (Mahoney & Stattin, 2000; Mahoney, Stattin, & Magnusson, 2001; Persson, Kerr, & Stattin, this volume; Stattin, Kerr, Mahoney, Persson, & Magnusson, in press). Furthermore, for girls, the more often a nominated friend attended a neighbourhood youth centre, the more antisocial acts the girl reported doing with that friend (Stattin et al., in press). Taken together, these results suggest that non-school peers might be more important than in-school peers when studying antisocial behaviour.

To our knowledge, only one study published to date has directly compared school and non-school peers (Kiesner, Poulin, & Nicotra, 2003). The study focused on peer groups, and found that both in- and after-school peer groups uniquely contributed to explaining variance in individual problem behaviour. Moreover, the in-school group showed a stronger unique relation with the individual's school-based problem behaviours, and the after-school group showed a stronger unique relation with the individual's out-of-school delinquency. Thus, homophily and peer selection may be based on different criteria across different contexts. This study suggests that there are good reasons to consider peer relations outside of the school and to include non-school friends in studies of homophily.

A second limitation in the literature on peer homophily is that most studies have focused only on friends and on one friend at a time—usually best friends or reciprocated best friends. There are two ways in which these studies might miss important antisocial influences. First, peers other than friends—older siblings, for instance, or romantic partners—might be important antisocial influences. In fact, recent studies on both romantic partners and siblings have implicated these relationships in the development of antisocial behaviour.

Although developing romantic relationships is a normal part of adolescent development, recent research suggests that such relationships can signal problems, under some conditions. Having many dating partners during adolescence has been linked to high levels of externalizing behaviours and poor emotional health and academic performance for both males and females, although the findings were stronger for females (Zimmer-Gembeck, Siebenbruner, & Collins, 2001). Similarly, heterosexual involvement has been linked to girls’ antisocial behaviour, but this was mainly so for girls who spent time in settings in which delinquent boys tended to congregate, which suggests that it was probably involvement with delinquent boys that was problematic (Persson et al., this volume; Stattin et al., in press). Girls have also reported doing more antisocial acts with boys than with girls (Stattin et al., in press). Although none of these studies actually demonstrates influence, they all suggest that, especially for girls, romantic partners might draw youths into social contexts or groups that encourage antisocial behaviour. Because romantic relationships do not show up in studies that ask youths to nominate friends, those studies miss what appears to be an important antisocial influence.

Recent work considering siblings has also underlined the importance of these peer relations when considering antisocial development. For example, in one study, sibling communication dealing with antisocial or rule-breaking behaviour predicted later problem behaviour, independent of deviant peer involvement (Bullock & Dishion, 2002). On the other hand, when antisocial talk with romantic partners, friends, and siblings were considered together as predictors of young adult delinquent peer involvement, antisocial talk with siblings did not have a unique effect (Shortt, Capaldi, Dishion, Bank, & Owen, 2003). Thus, the relative importance of siblings remains unclear.

Although the above research provides clear links between antisocial behaviour and sibling and romantic relationships, little is know about how these different types of relationships fit within individuals’ networks of peer relations. For example, if we simply ask individuals to nominate the peers who are most important to them, whom will they nominate, and how will the different types of relationships that are nominated be related to antisocial behaviour?

Another way in which studies of single friendships might miss important antisocial influences is by not considering additive homophily, or the idea that additional important peers might explain additional unique portions of the variance in an individual's antisocial behaviour. Although several studies have gone beyond single friendships and examined groups, these studies have typically used group scores based on an average or composite score collapsing across group members (Kiesner et al., 2002), averaging across multiple dyadic friendships (Hogue & Steinberg, 1995), conducting intraclass correlations to determine within-cluster similarity (Cairns et al., 1988), or treating peer groups as a nesting variable in Hierarchical Linear Modeling (Espelage, Holt, & Henkel, 2003). Thus, little is known about whether multiple non-aggregated peer relationships provide unique experiences to the individual, or whether these peers’ behaviours are so highly correlated that no information can be gained by studying multiple relationships.

Two studies that have addressed the issue of homophily in multiple relationships have looked at additional groups rather than additional dyadic relationships. In the first study, Urberg et al. (1997) examined homophily and additive peer influence from dyads and peer groups. The results showed that homophily was stronger for friendships than for groups, but that individuals were exposed to higher levels of substance use in the group context. The finding that homophily was stronger for friendships than for groups indicates more behavioural diversity in the group context—possibly because the Social Network Analysis used to create groups does not require that each individual identify or nominate all other group or clique members as friends or group members. Although these authors did not test for additive homophily, they did show that the best friend and the group both had unique effects on individual change in substance use. The important implication of this study is that it demonstrates additive peer influence. That is to say, considering multiple peer relations provides additional information for explaining changes in behaviour.

In the second relevant study, Kiesner et al. (2003) demonstrated additive homophily for in-school and after-school groups. As noted earlier, the in-school group tended to demonstrate a stronger unique relation with the individual's school-based problem behaviours, and the after-school group tended to show a stronger unique relation with the individual's out-of-school delinquency. Thus, these authors provided evidence for additive homophily as well as context specific homophily.

Finally, the studies discussed earlier regarding sibling relations are also relevant to considering multiple peer relations and additive peer influence. One found evidence for additive influence when considering a sibling and a friend (Bullock & Dishion, 2002), and the other provided evidence for additive influence when considering a romantic partner and a friend (Shortt et al., 2003). In sum, there are clear hints in the literature that multiple peer relationships are important for understanding peer homophily and influence.

In the present study, we overcome both of these limitations in the peer homophily literature by examining, in addition to friends, other important peers, by examining multiple peer relationships, and by not restricting those relationships to the school context. Specifically, we look beyond single friendships to three “Very Important Persons,” which we define as friends, siblings, or boy- or girlfriends—not limited to the school context. We describe the types of relationships adolescents have with their VIPs, where they reported meeting those peers, and where they spend time with them. Then, we test whether levels of antisocial behaviour, for both the target individual and their first VIP, differ across relationship type, meeting place, and place where they spend time together. Finally, we test for additive homophily by looking at whether the second and third VIPs’ antisocial behaviour scores increase the variance explained in the target youth's antisocial behaviour, over and above what the first VIP's antisocial score explains.