Developmental Continuity and Change in Physical, Verbal and Relational Aggression and Peer Victimization from Childhood to Adolescence

Abstract

To investigate the developmental course of aggression and peer victimization in childhood and adolescence, distinct subgroups of children were identified based on similarities and differences in their physical, verbal, and relational aggression and victimization. Developmental continuity and change were assessed by examining transitions within and between subgroups from grades 1 to 11. This longitudinal study consisted of 482 children (50% females) and was based on peer report data on multiple forms of aggression and peer victimization. Using person-centered methods including latent profile and latent transition analyses, most of the identified subgroups were distinguishable by their frequencies (i.e., levels) of aggression and victimization, rather than forms (physical, verbal and relational), with the exception of one group that appeared to be more form-specific. Across subgroups, multiple developmental patterns emerged characterized as early- and late-onset, social interactional continuity, desistance and heterotypic pathways. Collectively, these pathways support the perspective that the development of aggression and peer victimization in childhood and adolescence is characterized by heterogeneity.

It is widely recognized that peer victimization and aggression are pervasive social problems in childhood and adolescence. Indeed, one inference that can be made from a growing body of research is that the perpetration of peer directed aggression and experiences of peer victimization impact children and adolescents of different ages, genders, social classes, ethnic groups and nationalities (Juvonen & Graham, 2001; Lansford et al., 2012). When examined conjointly, some children are both perpetrators and victims of peer-directed aggression—aggressive-victims—and appear to have individual characteristics and social experiences that are unique from those of children who are primarily aggressors or victims (Kochenderfer-Ladd, 2003; Pellegrini, Bartini, Brooks, 1999; Schwartz, 2000; Schwartz, Dodge, Pettit, & Bates, 1997). Although distinguishing aggressive-victims from aggressors and (non-aggressive) victims appears to be an important conceptual distinction with significant implications for children’s adjustment more broadly, most of the investigations on the development of peer aggression and victimization have focused on these factors independently of one another. Thus, additional longitudinal research is warranted to provide further insights into the distinct developmental course of aggressive-victims from childhood to adolescence. Accordingly, this study had two specific aims: 1) to identify the nature (frequencies and forms) of aggressive-victims, aggressors, and victims in childhood and adolescence, and 2) to investigate the continuity and changes between and within these distinct subgroups during this developmental period.

Theoretical Perspectives on the Development of Childhood Aggression

Early- and late-onset

Several developmental theories have been proposed to explicate the development of antisocial behaviors, and aggression in particular, during childhood and adolescence. Moffitt’s (1993) developmental taxonomy theorized two primary types of antisocial children. Those with life-course persistent antisocial behavior are characterized by childhood onset aggression that remains stable throughout adolescence. In contrast, those with adolescence-limited antisocial behavior are characterized by adolescent-onset antisocial behavior which is more transient in nature such that it occurs primarily during adolescence. Alternatively, Patterson and colleagues (1989) proposed an early- and late-starter model of antisocial behavior which differentiated antisocial children based on their age of onset. These two perspectives were similar in that they both argued that some children engage in early-onset aggression and that among these children, aggression demonstrates considerable intra-individual stability into adolescence.

Consistent with these early-onset perspectives of childhood aggression, within our conceptual models we hypothesized that children who initially engaged in aggression (i.e., aggressors) would exhibit continuity in aggression during childhood and adolescence (see Figure 1a). Consistent with a late- or adolescent-onset pathway, it was hypothesized that some children who were initially uninvolved (i.e., non-aggressive) would exhibit a pathway in which they became more aggressive over time (see Figure 1b).

Figure 1.

Conceptual models depicting the development of aggression and peer victimization from childhood through adolescence.

Undoubtedly, early- and late-onset perspectives have profoundly influenced research on child and adolescent aggression; however, there are several notable limitations. These perspectives: 1) did not explicitly account for the influence of peer victimization and the potentially distinct development of aggressive-victims from non-victimized aggressors, 2) patterns of desistance or discontinuity that have been found among early-onset aggressors, or 3) distinct subtypes of aggression. The implications of these limitations for investigating the development of childhood aggression and peer victimization are discussed.

Development of aggressive-victims and social interactional continuity

Although developmental taxonomic perspectives did not explicitly differentiate the development of aggressive-victims from (non-victimized) aggressors, this may be an important conceptual distinction given the centrality of peer socialization processes in promoting the continuity of aggression in childhood and adolescence. According to the social interactional continuity model (Caspi et al., 1987), behavioral continuity is best understood in the context of person by environment interactions. Applying this model to the conjoint and dynamic development of aggression and peer victimization, it is postulated that children with aggressive dispositions engender a hostile and coercive social climate which leads to the formation of adverse peer relationships and subsequent experiences of peer victimization. In turn, experiencing victimization promotes hostile attributions and the modeling of aggressive behavioral responses which further exacerbates and sustains children’s aggression (Ladd, 2003). Extant evidence indicates bidirectional and transactional associations between aggression and similar forms of peer victimization concurrently and longitudinally (Giesbrecht, Leadbeater, & Macdonald, 2011; Ostrov, 2010; Reijntjes et al., 2011).

Consistent with the social interactional continuity model, it is hypothesized that the co-occurrence of aggression and peer victimization promote continuity among aggressive-victims, making it difficult for them to desist from this group (Figure 1c). Stated differently, similar to aggressors, aggressive-victims would exhibit an early-onset pathway, and given the repeated transactions between aggression and victimization over time, high rates of continuity were expected into adolescence. Within the conceptual model, interactional continuity was conceived in several ways. First, among children who are early aggressive-victims, interactional continuity implies that they remain aggressive-victims over time. Second, among children who are initially aggressors, interactional continuity would imply an increased risk for becoming aggressive-victims over time. Thus, it accounts for the concurrent associations between aggression and peer victimization as well as longitudinal associations such that aggression increases subsequent victimization (and thus membership into the aggressive-victim group over time). The social interactional continuity model also raises questions about whether children who are initially victims are at greater risk for becoming aggressive-victims over time. However, because (non-aggressive) victims tend to exhibit a passive-submissive and withdrawn behavioral style (Schwartz, 2000), this direction of effect was not hypothesized. Rather, it was expected that experiences of victimization would reinforce their submissive and withdrawn behaviors, thus promoting continuity in (non-aggressive) victimization over time.

Desistance pathways

A second limitation pertaining to the early-onset perspective is that it does not adequately recognize patterns of desistance or discontinuity that have been found among early-onset aggressors (Burt, 2012). Although many early-onset aggressors exhibit continuity, there are many that decline or desist from being aggressive by adolescence (e.g., childhood limited aggression; Barker & Maughan, 2009).

Accordingly, it was hypothesized that some aggressive children would be able to desist from being aggressive over time (i.e., transition to being uninvolved in aggression; see Figure 1d). Among aggressive-victims in particular, extending the interactional continuity perspective, it is hypothesized that declines in aggression would co-occur with declines in victimization, such that children who exhibited fewer aggressive behaviors would decrease their risks for victimization as well, and therefore transition into the uninvolved group.

Form-specific heterotypic continuity

A third limitation pertaining to the early- and late-onset perspectives is that they do not sufficiently distinguish the heterogeneity among distinct types of aggression or other antisocial behaviors (Burt, 2012). Towards this end, we propose that investigations on the developmental course of aggression require greater scrutiny of its different forms (or subtypes). Indeed, the majority of investigations on the development of childhood aggression have tended to focus on physical aggression or aggregated measures of aggression and rarely have they assessed the parallel development of multiple forms of aggression from childhood through adolescence (Broidy et al., 2003; Ettekal & Ladd, 2009).

According to the heterotypic continuity perspective, there are age-dependent behavioral manifestations of aggression, such that its form changes as children get older (Björkqvist, 1994; Björkqvist, Lagerspetz, & Kaukianen, 1992; Underwood et al., 2009). More specifically, it is postulated that among aggressive children, there is a specific age-dependent developmental pattern such that they are more physically aggressive in early and middle childhood, become more verbally aggressive by late childhood and, in turn, relationally aggressive by adolescence. The logic underlying this postulation is twofold. First, with normative gains in perspective-taking skills, verbal ability and emotional intelligence by late childhood and early adolescence, children are able to more effectively harm their peers using non-physical means. Second, by adolescence, physical aggression becomes less normative and socially acceptable, and sanctions (from adults and other peers) make it a less desirable behavior among those with aggressive behavioral styles. Therefore the heterotypic continuity perspective stipulates that aggressive adolescents use more sophisticated forms of relational aggression (e.g., social exclusion or friendship manipulation) as a substitute for physical aggression as it can be as damaging with fewer risks for retribution.

Applying the heterotypic perspective to the classification of children into distinct groups of aggressors and aggressive-victims, it is assumed that the nature of these groups would qualitatively change depending on the developmental periods assessed. Thus, rather than investigating the continuity and changes as proposed in the previous conceptual models (i.e., Figures 1a–1d) which do not account for form-specific development, heterotypic continuity would require further distinguishing these groups based on specific forms of aggression and victimization. Accordingly, it is hypothesized that children follow an age-dependent transitional pattern from primarily physical forms in middle childhood to verbal forms in late childhood, and in turn, relational forms in adolescence (see Figure 2). To investigate form-specific continuity and changes among aggressors and aggressive-victims, several researchers have used person centered methods such as latent transition analysis (LTA) to examine the conjoint development of multiple forms of aggression and victimization. Among these studies, which were conducted during late childhood and early adolescence, investigators found that the identified subgroups were more clearly differentiated by their frequencies (levels), rather than forms (Bettencourt, Farrell, Liu, & Sullivan, 2013; Giang & Graham, 2008; Williford, Brisson, Bender, Jenson, & Forrest-Bank, 2011). In other words, there was not much support for the premise that aggressive-victims or aggressors were specialized in particular forms of physical, verbal or relational aggression or victimization. However, because these investigations were conducted during late childhood to early adolescence (i.e., grades 4 to 7) and assessed short-term changes (e.g., across one-year intervals), it is plausible that they were limited in their ability to capture patterns of heterotypic continuity which conceivably manifest over longer developmental periods—that is, from early/middle childhood through adolescence. Thus, although findings from these studies do not conclusively support heterotypic continuity, additional research is warranted that examines these developmental processes over a longer developmental period that spans the majority of the childhood and adolescent years.

Figure 2.

Conceptual model depicting the heterotypic development of physical, verbal, and relational aggressors and aggressive-victims from childhood through adolescence.

Attempting to integrate the heterotypic perspective with early- and late-onset, desistance, and interactional continuity perspectives raises several important considerations. With respect to the early-onset perspective, the heterotypic perspective is not necessarily at odds, and may even provide greater specification. Both perspectives recognize the development and continuity of early aggressors, and from an integrative perspective, it would be postulated that early aggressors are primarily physically aggressive, and that their continuity is manifested in distinct age-dependent forms. In a similar vein, heterotypic continuity appears to be incompatible with developmental pathways reflecting late-onset or desistance. Moreover, attempting to reconcile the heterotypic and social interactional continuity perspectives also raises questions of theoretical incompatibility. If the interactional model can be extended to stipulate behavioral continuity in specific forms of aggression as a result of co-occurring victimization experiences, it is conceivable that children who are primarily early physical aggressive-victims would have a difficult time transitioning out of this group as their experiences of physical victimization would reinforce their physical aggression (Ostrov, 2010), thus remaining physical aggressive-victims over time. Because original formulations of heterotypic continuity (Björkqvist, 1994; Björkqvist et al., 1992) were intended to describe the development of aggression as an independent phenomenon from peer victimization it remains unclear whether this perspective could be applied to the developmental course of aggressive-victims, and to our knowledge, this study is the first to test these alternative hypotheses across the middle childhood to late adolescent years.

Gender Differences

Existing evidence pertaining to the role of gender in children’s aggression and victimization, and the extent to which it may contribute to the stability of distinct subgroups, is mixed. Many studies have found that boys tend to be more physically aggressive than girls, and although girls are sometimes more relationally aggressive, these differences appear to be trivial (Card et al., 2008). Other investigations have not reported form-specific differences, but rather that boys are more likely to be aggressors, victims and aggressive-victims (Hanish & Guerra, 2004; Nylund, et al., 2007; Schwartz, 2000). Even if gender differences exist among these subgroups, it does not necessarily imply greater stability among boys. On the contrary, Wolke et al. (2009) found that direct (e.g., physical) forms of victimization were more stable among girls. However, Hanish and Guerra (2004) reported higher rates of stability for aggressive-victim boys.

Study Aims and Hypotheses

This study aims to build on existing research by addressing two primary objectives. Aim 1 of this study was to explore the nature (i.e., forms and frequencies) of co-occurring peer aggression and victimization from middle childhood to late adolescence. Towards this end, at four different ages (i.e., grades 1, 5, 8 and 11), children and adolescents were classified into distinct subgroups based on similarities and differences in their usage and experiences (i.e., frequencies) of physical, verbal and relational peer aggression and victimization. Each of these ages (grade levels) was selected to reflect distinct developmental periods. More specifically, grade 1 was intended to reflect middle childhood, grade 5 late childhood, grade 8 early adolescence, and grade 11 late adolescence. To address this aim, this study utilized latent profile analysis (LPA), a person-oriented methodology.

Consistent with the empirical evidence, it was expected that of the subgroups identified, several would be distinguishable primarily by their frequencies, rather than forms, of aggression and victimization. Consequently, at least four subgroups were hypothesized at each grade level: 1) aggressors (i.e., children with high levels of physical, verbal and relational aggression), 2) aggressive-victims (high levels of the three forms of aggression and victimization), 3) victims (high levels of the three forms of victimization), and 4) uninvolved (low levels of all forms of aggression and victimization). In addition to these four subgroups, more ‘specialized’ (form-specific) subgroups of physical, verbal or relational aggressors and aggressive-victims were theoretically warranted, consistent with heterotypic perspectives.

Aim 2 of this study was to explore the developmental continuity and change of the identified subgroups (from Aim 1). Toward this end, transition probabilities were estimated using LTA. Prior investigations have used similar methodological approaches to examine the (in)stability of aggression-victimization subgroups; however, they have focused on shorter developmental periods (e.g., over one year). This study was the first of this kind to explore these associations across multiple developmental periods spanning from middle childhood to late adolescence which would allow for more nuanced hypotheses about form-specific continuity and change. Transition probabilities were estimated three times: first, to assess developmental trends during childhood (i.e., middle to late childhood: grades 1 to 5); second, the transition to adolescence (i.e., from late childhood to early adolescence: grades 5 to 8); and third, during adolescence (i.e., from early to late adolescence: grades 8 to 11).

Developmental continuity and change were hypothesized to be manifested in multiple ways consistent with the early-onset, late-onset, desistance and interactional continuity models depicted in Figure 1. Collectively, these pathways were hypothesized to account for a substantial proportion of the transitional trends between subgroups, thus demonstrating the heterogeneous development of co-occurring aggression and peer victimization. To the extent form-specific groups were identified (in Aim 1), heterotypic pathways (depicted in Figure 2) were also investigated in order to determine whether the heterotypic continuity perspective would provide a more nuanced explanatory viewpoint pertaining to the form-specific development of these subgroups in comparison to the alternative conceptual models discussed.

A final exploratory aim was to examine gender differences. It was hypothesized that boys would be over-represented among groups of aggressors and aggressive-victims. However, considering previous inconsistent findings, no apriori hypotheses were made pertaining to whether gender would moderate the transitions between groups over time.

Method

Participants

Data for this study were part of a larger longitudinal project of children’s social, emotional, and academic development from kindergarten to grade 12. The sample for this study consisted of 482 children (242 females and 240 males) who had aggression and peer victimization data available at some point between grades 1 and 11. From this sample, the majority of participants (n = 383) were recruited upon kindergarten entry (M_age = 5.59) and an additional sample (n = 99) of children were added to the longitudinal project in grade 5, and therefore had no available data from earlier years. Before participant recruitment began, consent was first obtained from multiple school districts in the Midwestern United States, and of the families invited to participate in this study, 95% consented to their child’s participation. School districts were selected which served students from diverse backgrounds and to proportionately represent this locales’ population in terms of geographic, racial and socioeconomic characteristics. The sample contained nearly equal proportions of families from urban, suburban, or rural Midwestern communities. The median total household income was between $30,001 to $40,000 (19.1% low income, i.e., below $20,000; 43.1% middle income or higher, i.e., over $50,001). Children were primarily Caucasian (80.1%) and African American (15.8%), as well as a small percentage of Hispanic, biracial and other backgrounds (4.1%).

Procedure

This study utilized peer report data collected across 4 grades (grades 1, 5, 8 and 11). Participants and their classmates completed peer nomination questionnaires which assessed student’s physical, verbal and relational aggression and peer victimization. Over the course of this project, participants became increasingly dispersed across classrooms and schools. When children changed schools, permission was sought to extend the project into their respective schools, and project children’s classmates were invited and consented to participate. Administration procedures varied across grade levels to account for developmental differences and changing school settings. In grade 1, participants completed individual interviews with a trained staff member and were asked to provide limited nominations (up to three) by pointing to photographs of their classmates that were displayed on a felt board. In grade 5, classroom-wide sociometric procedures were used and students could provide unlimited peer nominations. Before participants completed the peer nomination forms, they used practice criteria to ensure that they knew their classmates’ names and could perform the nomination procedures correctly.

These procedures for collecting peer nominations are well established in the sociometry literature (Coie, Dodge, & Coppotelli, 1982; Parkhurst & Asher, 1992). In higher levels of schooling (grades 8 and 11), because participants spent time in multiple classrooms and it was not feasible to interview all of their classmates in all of their classes, permission was obtained to review participants’ class schedules, and classmates who shared a minimum of one class with the participant were identified. From this pool of classmates, up to 40 students were randomly selected to provide peer nominations. This method of randomly selecting nominators has been found to be a valid and reliable method for collecting sociometry data when it is not feasible to include the entire sample of classmates (see Prinstein, 2007). To ensure that respondents knew the persons they were nominating, respondents were instructed to nominate only those classmates they knew well and could indicate that they “don’t know this person” on the nomination forms. From grades 1 to 11, the total number of nominators ranged from 1,576 to 2,593 students across 53 to 166 classes. The number of respondents per participant ranged from 10 to 39, with 86% to 98% of participants having 15 respondents or more (mean number of nominators per participant ranged from 17.8 to 29.7 across grade levels).

Measures

Peers’ reports of aggression and victimization

To identify children’s aggressor, victim, and aggressive-victim subgroups, indicators of physical, verbal and relational peer-directed aggression and victimization were assessed: (1) “Someone who hits, kicks, or pushes other kids” (i.e., physical aggression), (2) “Someone who teases, calls names, or makes fun of other kids” (i.e., verbal aggression), (3) “Someone who tells other kids they won’t like them or be their friend anymore just to hurt them or get their own way (i.e., relational aggression), (4) “Someone who gets hit, pushed, or kicked by other kids” (physical victimization), (5) “Someone who gets teased, called names, or made fun of by other children” (i.e., verbal victimization), and (6) “Those who other kids gossip about or say bad things about behind their backs” (i.e., relational victimization). Items pertaining to relational aggression and victimization were not collected in grade 1, but were assessed in subsequent grade levels. Standardized scores were computed by classroom to adjust for the number of nominators.

Data Analysis Plan

Models were estimated in Mplus Version 7.2 using full information maximum likelihood estimation with robust standard errors (MLR; Muthén & Muthén, 1998–2012). To investigate the nature (i.e., frequencies and forms) of co-occurring peer aggression and victimization (Aim 1) latent profile analysis (LPA) was used. LPA is a variant of latent class analysis (LCA) based on continuous-scale indicators to identify distinct subgroups (i.e., classes) of subjects who exhibit similar patterns (profiles) across multiple observed indicators (see Collins & Lanza, 2010). This methodology is ideal for differentiating groups based on both the frequency and forms of aggression and peer victimization. Investigators have recommended this methodology as opposed to more traditional methods (e.g., using cut-off scores) because it does not rely on an arbitrary cut-off score, better accounts for measurement error, and provides greater classification accuracy (Giang & Graham, 2008; Nylund et al., 2007). LPA was performed separately at each grade level. To investigate the continuity and changes among the identified subgroups over time (Aim 2), LTA was used. LTA can be conceived as a longitudinal extension to LPA/LCA and is a flexible methodology for investigating time-specific classifications of individuals. In addition to identifying individuals’ class membership (similar to LPA), LTA estimates transition probabilities which indicate the likelihood that children make specific transitions between two latent classes over time. To further explore gender differences, conditional LPAs were specified. Within the LPA framework, because class membership is specified as a latent categorical variable, gender effects were estimated using multinomial logistic regression and interpreted as odds ratios (OR). Finally, to assess whether gender moderated specific subgroup transitions over time, a series of moderated LTAs were specified.

Results

Preliminary Analyses

First, preliminary analyses were performed which included descriptive statistics (Table 1) and bivariate correlations (Table 2). The bivariate correlations indicated that the aggression forms were moderately to highly positively correlated with each other and had small to moderate correlations with peer victimization. Bivariate correlations tended to decrease over longer time periods. Given the longitudinal nature of this study, participant attrition and missing data increased over time. For all study variables, the percentage of missing data is reported in Table 1. Missing data analyses indicated that, for all study variables, 24% of the data were missing. Because additional participants were added to the longitudinal study in grade 5, roughly 20% of the missing data in grade 1 was due to this reason. To assess whether missing data introduced any observable causes of bias, a series of univariate t-test comparisons were performed. First, non-significant t-tests indicated that missing data on the aggression and peer victimization measures were not associated with children’s demographic characteristics (gender, race, family income). Second, given the higher rates of attrition by grade 11, additional comparisons were made to see if this attrition was associated with earlier levels of aggression or peer victimization. Results indicated that students who had dropped out of the study in grade 11 were not more likely to be aggressive or victimized in grades 1 or 5.

Table 1.

Descriptive and Missing Data Statistics for All Study Variables

	N	Min	Max	M	SD	% Missing┼
Grade 1
1. Physical aggression	368	−1.43	4.10	−0.05	0.95	0.24
2. Verbal aggression	368	−1.61	3.79	−0.06	0.93	0.24
3. Physical victimization	368	−2.03	3.00	0.04	0.98	0.24
4. Verbal victimization	368	−2.11	3.19	0.00	1.01	0.24
Grade 5
1. Physical aggression	419	−1.33	4.09	0.09	1.06	0.13
2. Verbal aggression	419	−1.51	3.93	0.11	1.04	0.13
3. Relational aggression	418	−1.49	3.47	0.07	1.04	0.13
4. Physical victimization	418	−1.11	4.00	0.15	1.12	0.13
5. Verbal victimization	419	−1.69	3.60	0.16	1.12	0.13
6. Relational victimization	418	−1.60	3.62	0.16	1.08	0.13
Grade 8
1. Physical aggression	404	−1.55	4.15	0.05	1.06	0.16
2. Verbal aggression	404	−1.56	4.79	0.03	1.05	0.16
3. Relational aggression	403	−1.62	4.59	−0.01	1.02	0.16
4. Physical victimization	403	−1.35	5.52	0.16	1.14	0.16
5. Verbal victimization	404	−1.37	4.13	0.15	1.09	0.16
6. Relational victimization	404	−1.71	5.22	0.11	1.04	0.16
Grade 11
1. Physical aggression	270	−1.40	5.25	0.18	1.17	0.44
2. Verbal aggression	270	−1.20	4.87	0.19	1.11	0.44
3. Relational aggression	270	−1.05	4.08	0.09	1.02	0.44
4. Physical victimization	269	−3.07	5.09	0.01	0.99	0.44
5. Verbal victimization	270	−1.33	3.74	0.08	0.95	0.44
6. Relational victimization	270	−1.30	3.81	0.05	0.98	0.44

^┼Because additional participants were added into the longitudinal study in grade 5, roughly 20% of the missing data in grade 1 was attributable to this reason.

Table 2.

Bivariate Correlations among Study Variables

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21
1. PA (G1)
2. VA (G1)	.80**
3. PV (G1)	.02	.04
4. VV (G1)	.27**	.29**	.30**
5. PA (G5)	.52**	.50**	.01	.15**
6. VA (G5)	.50**	.49**	.11*	.24**	.81**
7. RA (G5)	.20**	.20**	.06	.19**	.42**	.59**
8. PV (G5)	.28**	.28**	.07	.16**	.24**	.19**	.15**
9. VV (G5)	.21**	.23**	.06	.16**	.21**	.21**	.21**	.82**
10. RV (G5)	.26**	.28**	.12*	.20**	.23**	.33**	.37**	.64**	.76**
11. PA (G8)	.44**	.40**	.07	.14*	.60**	.54**	.27**	.10	.07	.12*
12. VA (G8)	.36**	.31**	.13*	.21**	.49**	.54**	.41**	.11*	.09	.22**	.76**
13. RA (G8)	.14*	.09	.08	.15**	.21**	.32**	.41**	.00	.07	.24**	.40**	.63**
14. PV (G8)	.17**	.24**	.07	.06	.20**	.17**	.09	.58**	.49**	.41**	.24**	.20**	.01
15. VV (G8)	.13*	.18**	.09	.09	.12*	.16**	.16**	.55**	.57**	.56**	.18**	.24**	.17**	.70 **
16. RV (G8)	.09	.08	.11	.16**	−.01	.15**	.31**	.26**	.34**	.46**	.14**	.37**	.45**	.29 **	.59**
17. PA (G11)	.27**	.27**	.13	.11	.40**	.24**	.11	.16*	.14*	.21**	.60**	.49**	.25**	.15 *	.09	.09
18. VA (G11)	.18**	.19**	.20**	.23**	.38**	.39**	.28**	.06	.08	.18**	.49**	.59**	.37**	.16 *	.11	.25 **	.63**
19. RA (G11)	.00	−.01	.24**	.21**	.08	.14*	.29**	−.05	−.04	.14*	.14*	.32**	.33**	−.14 *	−.02	.28 **	.32**	.57**
20. PV (G11)	.16*	.15*	.06	.01	.10	.05	−.03	.18**	.13*	.04	.26**	.15*	−.02	.36 **	.21**	.07	.38**	.22**	.11
21. VV (G11)	.04	.09	.06	.13	.07	.15*	.18**	.20**	.23**	.30**	.23**	.32**	.23**	.29 **	.39**	.39 **	.35**	.46**	.39 **	.41 **
22. RV (G11)	.00	.03	.17*	.18**	−.05	.14*	.30**	.00	.05	.27**	.05	.27**	.39**	−.01	.20**	.48 **	.26**	.45**	.61 **	.17 **	.61**

G= Grade, PA = physical aggression, VA = verbal aggression, RA = relational aggression, PV = physical victimization, VV = verbal victimization, RV = relational victimization.

^*p < .05;

^**p < .01

Identifying Subgroups of Aggressors, Victims and Aggressive-victims (Aim 1)

To identify subgroups of aggressors, victims, and aggressive-victims, LPA was performed at each grade level. Models with varying numbers of classes were specified, and to determine the optimal solution, multiple fit indices were examined including the BIC, AIC, sample-size adjusted BIC (SABIC), and Lo-Mendell-Rubin adjusted Likelihood Ratio Test (LMR-aLRT). Models with smaller AIC, BIC and SABIC values indicate better fitting solutions (Collins & Lanza, 2010). Significant p values on the LMR-aLRT indicate that a model with k classes has better fit to the data than a model with k – 1 classes. Entropy and class assignment probabilities were also assessed which measure classification precision (values ranging from 0 to 1 with values closer to 1 indicative of greater precision). Class-specific means were plotted to assess the qualitative nature of the classes and to determine whether each class was conceptually meaningful and interpretable in consideration of extant empirical findings and theory.

In grade 1, it appeared that the relative decrease in the AIC, BIC and SABIC was smaller for models with three or more classes and the results for the LMR-aLRT were discrepant (see Table 3). In addition to the model fit indices, the qualitative nature of classes was examined in each solution (note that the relational aggression and victimization items were not administered in grade 1 and the LPA was based on the four available indicators). Compared to the classes that were identified in the three-class model, the additional classes in the four-and five-class models appeared to be qualitatively similar and were not conceptually distinct (e.g., the four-class model identified two classes of children which could both be characterized as aggressive-victims). Thus, the three-class model was selected as the most parsimonious solution and consisted of 10.3% of children who were highly aggressive with high levels of verbal victimization (labelled aggressive-victims), 14.1% who were aggressive with low levels of victimization (labelled aggressors), and 75.5% with low aggression and victimization (labelled uninvolved).

Table 3.

Model Fit Indices for Latent Profile Analyses in Grades 1, 5, 8 and 11

Model	LogL	AIC	BIC	SABIC	Entropy	LMR-aLRT
Grade 1
One-class	−2038.00	4092.00	4123.27	4097.89	-	-
Two-class	−1771.75	3569.49	3620.30	3579.05	0.96	515.07***
Three-class	−1701.82	3439.63	3509.98	3452.87	0.95	135.28
Four-class	−1676.48	3398.96	3488.84	3415.87	0.95	49.02
Five-class	−1651.98	3359.97	3469.40	3380.56	0.90	47.39
Grade 5
One-class	−3743.62	7511.24	7559.70	7521.62	-	-
Two-class	−3274.05	6586.10	6662.82	6602.52	0.96	917.44***
Three-class	−3037.67	6127.33	6232.32	6149.81	0.96	461.84*
Four-class	−2896.59	5859.18	5992.42	5887.71	0.98	275.64**
Five-class	−2807.17	5694.34	5855.86	5728.93	0.96	174.70
Six-class	−2735.64	5565.27	5755.05	5605.91	0.96	139.77**
Grade 8
One-class	−3589.78	7203.55	7251.57	7213.49	-	-
Two-class	−3271.85	6581.71	6657.73	6597.45	0.94	621.06***
Three-class	−3102.47	6256.93	6360.97	6278.47	0.94	330.90
Four-class	−3007.08	6080.16	6212.21	6107.49	0.95	186.33
Five-class	−2914.87	5909.75	6069.80	5942.88	0.96	180.13
Six-class	−2856.19	5806.39	5994.46	5845.32	0.96	114.63
Grade 11
One-class	−2347.69	4719.39	4762.57	4724.52	-	-
Two-class	−2128.47	4294.95	4363.32	4303.07	0.92	427.53
Three-class	−2025.61	4103.23	4196.79	4114.35	0.93	200.60*
Four-class	−1969.06	4004.13	4122.88	4018.24	0.95	110.29
Five-class	−1924.68	3929.36	4073.30	3946.47	0.96	86.56
Six-class	−1888.64	3871.28	4040.40	3891.38	0.96	70.29

Note. LogL = Loglikelihood, AIC = Akaike information criteria, BIC = Bayesian information criteria, SABIC = Sample-size adjusted Bayesian information criteria; LMR-aLRT = Lo Mendell Rubin adjusted likelihood ratio test.

^*p < .05;

^**p < .01;

^***p < .001

In grades 5, 8 and 11, the analyses revealed a similar set of findings. With respect to the information criteria (i.e., AIC, BIC, SABIC), the results consistently indicated that the criteria values decreased as the number of classes increased; however model fit indices did not indicate a single solution when examining the information criteria in combination with the LMR-aLRT (see Table 3). With respect to the qualitative nature of classes that were identified at each grade level, the five-class models consistently revealed five distinct and conceptually meaningful classes. However, the additional class included in the six-class models was not qualitatively distinct from other classes, and consisted of a very small proportion of children. Thus, across grades 5, 8 and 11, the five-class models were selected as the most parsimonious solutions. In Figure 3, class-specific means are illustrated, and class proportions and sample sizes are reported. Collectively, approximately 4% of children and adolescents in grades 5, 8, and 11 had high levels of multiple forms of aggression and victimization (aggressive-victims). The proportion of those with high relational aggression and victimization scores (relational aggressive-victims) increased from 7% to 14% over time. The proportion of those with high aggression and low victimization scores (aggressors) decreased from about 16% to 8% over time. Similarly, those with low aggression and high victimization scores (victims) decreased over time from about 12% to 6%. About 61% to 70% had low aggression and victimization scores (uninvolved).

Figure 3.

LPA solutions for grades 1, 5, 8 and 11. PA = physical aggression, VA = verbal aggression, RA = relational aggression, PV = physical victimization, VV = verbal victimization, RV = relational victimization.

Continuity and Changes in Aggression-Victimization Subgroups over Time (Aim 2)

To assess the continuity of the subgroups that were identified in Aim 1, LTA was performed such that transition probabilities were estimated across adjacent time waves (i.e., grades 1 to 5, 5 to 8 and 8 to 11). Transition probabilities may be reflective of both continuity and change; however, the notion that continuity can be measured assumes that the nature of latent classes that are identified are qualitatively similar over time. If this assumption is not met (e.g., the nature of the subgroups changes over time), then it would not be possible to measure continuity per se. To meet this assumption, measurement invariance was imposed when it appeared that the nature of the latent classes were similar over time (i.e., the class-specific means of similar classes were constrained to be equal). To perform LTA with measurement invariance, the manual 3-step approach was used (see Asparouhov & Muthén, 2014).

Based on the findings from the cross-sectional analyses (i.e., Aim 1), in grades 5, 8 and 11 (but not grade 1), the nature and number of classes appeared to be comparable. Although the nature of the victim group in grade 11 was somewhat distinct from the victim groups identified in grades 5 and 8, these groups were fairly similar on five of the six indicators (with the exception of relational victimization), and they nonetheless consisted of children who were primarily victimized and not aggressive. To evaluate the impact of imposing measurement invariance from grades 5 to 11, a model in which similar latent classes were constrained to be identical over time (i.e., constrained model: Logl = −7768.40; AIC 15656.79, BIC = 15905.70, SABIC = 15715.27) was compared to a second model in which the classes were identified in parallel, but independently of one another, without constraining them to be equal (i.e., unconstrained model: Logl = −7646.73; AIC 15533.46, BIC = 16031.27, SABIC = 15650.42). The unconstrained model had a lower log-likelihood, AIC and SABIC, and the constrained model had a lower BIC. Although the model fit indices did not consistently indicate that either model had better fit, in order to facilitate the interpretation of the stability estimates, measurement invariance from grades 5 to 11 appeared to be a reasonable determination.

Transition probabilities across adjacent grades are reported in Table 4. Aggressors and aggressive-victims in grade 1 were significantly more likely to be aggressors (32.2% and 46.2%, respectively) and aggressive-victims (10.7% and 16.6%, respectively) in grade 5 compared to children who were initially uninvolved; however, they were not more likely to be victims or relational aggressive-victims. In contrast, most children who were uninvolved in grade 1 remained uninvolved in grade 5 (74.3%). In grades 5 to 8 and 8 to 11, aggressors exhibited moderate rates of stability (from 41.2% to 51.8%); however, a substantial portion were also able to transition out of risk to the uninvolved group (from 31.2% to 43.2%). Compared to the uninvolved group (about 1%), aggressors were significantly more likely to become aggressive-victims (about 7%). Rates of stability among aggressive-victims were modest and appeared to decline in late adolescence (from 47.2% to 30.7%). They were more likely to transition to relational aggressive-victims in late adolescence (22.2%) compared to earlier adolescence (0%). Moreover, about one in four aggressive-victims (from 24.8% to 28.0%) transitioned out of risk. Relational aggressive-victims had high stability estimates (from 63.4% to 73.2%), and were most likely to transition to the uninvolved group if they did not exhibit continuity (from and 21.9% to 24.7%). Transitions for victims reflected substantial rates of instability during late adolescence (48.3% of victims remained victims from grades 5 to 8, but this rate dropped to 26.7% from grades 8 to 11). Most victims who transitioned out of this group showed signs of recovery and transitioned into the uninvolved group, however, some victims transitioned into aggressive-victims (about 10%). Among the uninvolved group, there were high stability estimates (from 81.7% to 82.2%); however, for those who exhibited late-onset pathways, it was highly unlikely to become aggressive-victims, and a minority exhibited transitional patterns to other risk groups.

Table 4.

Latent Transition Probabilities Examining Continuity and Change from Grades 1 to 5, 5 to 8 and 8 to 11

		Aggressors (A)	Victims (V)	Aggressive-Victims (AV)	Relational Aggressive-Victims (RA-RV)	Uninvolved (U)
Grade 5
Grade 1	A	32.2%	13.5%	10.7%	7.5%	36.0%
	AV	46.2%	13.0%	16.6%	5.8%	18.5%
	U	6.5%	13.1%	0.0%	6.2%	74.3%
Grade 8
Grade 5	A	41.2%	0.0%	9.4%	6.2%	43.2%
	V	7.3%	48.3%	7.9%	1.0%	35.5%
	AV	7.9%	20.0%	47.2%	0.0%	24.8%
	RA-RV	0.0%	11.9%	0.0%	63.4%	24.7%
	U	8.4%	4.9%	0.3%	4.1%	82.2%
Grade 11
Grade 8	A	51.8%	3.8%	6.8%	6.3%	31.2%
	V	0.0%	26.7%	9.5%	0.0%	63.8%
	AV	0.0%	19.1%	30.7%	22.2%	28.0%
	RA-RV	0.0%	0.0%	4.8%	73.2%	21.9%
	U	7.1%	2.8%	1.3%	7.2%	81.7%

Gender Differences

To explore gender differences, two approaches were used. First, gender was specified as a covariate (based on the classes identified in the measurement invariant models when applicable), and results from these conditional LPAs consistently revealed gender differences in group membership. Specifically, aggressive-victims and aggressors were more likely to be boys than children in the uninvolved group in grade 1 (OR = 14.4, p < .001 and OR = 10.8, p < .001, respectively). Aggressive-victims, aggressors and victims were significantly more likely to be boys in grade 5 (OR = 19.4, p < .01, OR = 8.5, p < .001 and OR = 2.6, p < .01, respectively) and grade 8 (OR = 4.5, p = .01, OR = 9.3, p < .001 and OR = 2.5, p < .01, respectively). In grade 11, aggressors and victims were significantly more likely to be boys (OR = 10.5, p < .01 and OR = 10.9, p = .03, respectively). In grades 5, 8 and 11, relational aggressive-victims were less likely to be boys (OR = 0.2, p = .02, OR = < .01, p < .001 and OR = 0.2, p < .01, respectively).

Second, a series of moderated LTAs were specified using procedures outlined by Muthén and Asparouhov (2011) to test for moderation effects in conjunction with the manual 3-step approach. Because covariate effects revealed that risk group membership was significantly related to gender, this limited the ability to assess transition effects between risk groups. The only moderated effects estimated were those from the uninvolved group to the aggressor and victim groups (it was highly unlikely for uninvolved children to transition to the aggressive-victim group and the relational aggressive-victims were primarily girls, thus estimating moderating effects for these transitions were not possible). Nonetheless, during each transitional period, boys had an increased risk of transitioning to the aggressor group from being initially uninvolved (OR = 3.99, p = .02 from grade 1 to 5, OR = 8.11, p < .001 from grade 5 to 8, and OR = 7.74, p = .04 from grade 8 to 11). Moreover, from grade 1 to 5, boys had an increased risk of transitioning to the victim group from being initially uninvolved (OR = 2.71, p < .01).

Discussion

Taken together, the findings provided several novel insights into the co-occurring development of physical, verbal, and relational peer aggression and victimization in childhood and adolescence. Findings highlighted the utility of using person-centered methods to identify subgroups of children with similar patterns of aggression and victimization. Four of the groups identified (aggressive-victims, aggressors, victims, and uninvolved) corroborated extant evidence as hypothesized. These groups were distinguishable by their frequencies of aggression and victimization, rather than forms, such that they displayed relatively similar levels of physical, verbal, and relational forms. There was one notable exception, however, a group that was more form-specific, consisted of mostly girls, and involved primarily in relational forms. Although previous studies have focused on the identification of groups during late childhood and early adolescence (Bettencourt et al., 2013; Giang & Graham, 2008; Williford et al., 2011), the findings reported here were based on a considerably longer developmental period than has been previously assessed, and allowed for a more nuanced examination of the nature (i.e., forms and frequencies) of aggressive-victimization in childhood and adolescence. Towards this end, findings demonstrated how the identified groups were consistent, with a few exceptions, in middle childhood (grade 1), late childhood (grade 5), early adolescence (grade 8) and late adolescence (grade 11). These findings indicated that the qualitative nature of the identified groups was not conditional on the developmental period assessed. Thus, physical, verbal and relational forms of aggression and victimization remain developmentally salient across the childhood and adolescent years and do not appear to be age-specific manifestations.

Perhaps the most significant contributions of this study pertain to its implications about the developmental continuity and change (i.e., stability and instability) of multiple forms of peer aggression and victimization. Taken together, the findings across subgroups elucidate the heterogeneous development of aggression and peer victimization in childhood and adolescence. Indeed, none of the conceptual models alone was able to accurately depict this heterogeneity; however, by considering multiple developmental pathways for each group, it was possible to account for a significant amount of the transitional rates that were observed. Among aggressors, pathways reflecting early-onset, social-interactional and heterotypic continuity and desistance accounted for nearly 86% to 100% of the transitional probabilities over time. Among aggressive-victims, pathways reflecting interactional and heterotypic continuity and desistance accounted for nearly 41% to 81% of the transitional probabilities among children in this group, and among relational aggressive-victims, interactional continuity and desistance accounted for nearly 88% to 95% of the transitional probabilities. Pathways reflecting late-onset also emerged, however, most children who were initially uninvolved maintained low levels of aggression and victimization.

These findings indicate that patterns of developmental heterogeneity appear to be far more common than has been conceived in many previous studies which have typically relied on latent class growth modeling methodologies to investigate developmental trajectories and intra-individual continuities in aggression. Latent growth models appear to be better suited for assessing stability than more transient or sporadic patterns, and are not as sensitive at identifying those who shift in and out of aggression over time. Furthermore, developmental theories on aggression and peer victimization need to more comprehensively recognize these observed patterns of heterogeneity in childhood and adolescence (Burt, 2012).

Identification of Subgroups Based on Physical, Verbal and Relational Aggression and Peer Victimization (Aim 1)

Three of the groups identified over time exhibited some or multiple forms of aggression. Similar to prior investigations, those who were labelled as aggressors or aggressive-victims were not specialized in any particular form, but rather displayed high levels (frequencies) of multiple forms (Bettencourt et al., 2013; Nylund et al., 2007; Williford et al., 2011). Contrary to expectations, it was unclear why aggressive-victims in grade 1 exhibited low levels of physical victimization in combination with higher levels of physical-verbal aggression and verbal victimization. Perhaps, this finding reflects on how physical victimization during this period may be more normative, and it did not sufficiently differentiate children within the identified groups.

In contrast to aggressors and aggressive-victims, those who were labelled as relational aggressive-victims were more form-specific in nature. Although the proportion of aggressors had declined over time, there was a corresponding increase in relational aggressive-victims such that compared to late childhood and early adolescence, this group nearly doubled in proportion by late adolescence. The findings for this group, in combination with the gender differences found among the other aggressive risk groups, imply that boys were more likely to exhibit high levels of physical, verbal and relational aggression and victimization, however, among girls, it was more common to display a form-specific developmental trend characterized by higher relational and lower physical forms (Card et al., 2008). It is unclear why a group of this nature has not been more consistently identified in other investigations (Bettencourt et al., 2013; Giang & Graham, 2008; Williford et al., 2011). By the time children reach adolescence, they have likely developed the social-cognitive skills to engage in different manifestations of relational aggression, and also recognize that it can be as harmful to their victims as physical aggression with fewer risks of retribution from peers or punishment from adults (e.g., from teachers, parents, or school administrators). Thus, this is a group that is theoretically plausible assuming that relational aggression is becoming more normative in adolescence, especially among adolescent girls, and there is a corresponding decline in physical aggression (Björkqvist, 1994; Björkqvist et al., 1992; Karriker-Jaffe, Foshee, Ennett, & Suchindran, 2008).

In contrast to the three aggressive groups, the two other groups that were identified were non-aggressive in nature (i.e., victims and uninvolved). Consistent with expectations and previous investigations, the majority of children and adolescents identified at every grade level had persistently low levels of peer aggression and victimization across multiple forms (i.e., uninvolved). The findings for victims revealed some unexpected developmental differences. Exactly why this group was not identified in grade 1 is unclear. Perhaps, this reflects on how experiences of victimization may be more prevalent and normative during the early grade school years (Ladd, Ettekal, & Kochenderfer-Ladd, 2017), a period when children are making considerable gains in their self-regulation and behavioral inhibition. Moreover, the nature of victims was somewhat unexpected in late adolescence in that they had greater rates of physical and verbal, but not relational, victimization. Although it is unclear why the nature of this group changed, these findings are similar to the victim group identified by Giang and Graham (2008).

Developmental Continuity and Change (Aim 2)

Early-onset

Consistent with the developmental taxonomic perspective on early onset-aggression which posits it is a stable behavioral style that begins in childhood and exhibits continuity into adolescence (Moffitt, 1993), a substantial proportion of children who engaged in high levels of aggression by middle childhood maintained this behavioral style through adolescence. Although initial formulations of this perspective did not explicitly articulate the development of different forms of aggression, the findings demonstrated that children with early-onset aggressive pathways exhibited continuity in co-occurring physical, verbal and relational aggression. Rates of continuity increased from childhood to adolescence; however, there are several possible explanations for this pattern. First, it is possible that those who are chronically aggressive over longer periods of time are more likely to maintain this behavioral style and have a more difficult time recovering (desisting). Second, early aggressors (identified in grade 1) had lower levels of aggression compared to those in later years, thus it is possible that continuity is more pronounced among those who are more severely or highly aggressive (Burt, 2012).

Late-onset

Late-onset pathways to aggression were assessed by the transitions from the uninvolved group to the aggressive groups. In contrast to most uninvolved children who remained uninvolved in childhood and adolescence, a relatively small percentage exhibited late-onset pathways such that they transitioned into aggressors and relational aggressive-victims. Transitional patterns from middle to late childhood indicated that, compared to early aggressors and aggressive-victims, uninvolved children were about equally likely to transition to relational aggressive-victims. Considering that the uninvolved group was considerably larger than aggressors or aggressive-victims, these findings suggest that most relational aggressive-victims in late childhood did not initially exhibit patterns of aggression, contrasting with heterotypic continuity perspectives, and rather exhibited a late-onset relational aggression pathway. Conceptualizations of late- or adolescent-onset antisocial behavior have typically focused on increases in non-aggressive rule-breaking behaviors (Moffitt, 1993, Burt, 2012). These findings suggest that in addition to rule-breaking behaviors, the late-onset antisocial pathway may also be characterized by increases in non-physical forms of aggression. However, because relational aggression was not measured in middle childhood, it was not possible to rule out an early-onset relational aggression pathway and interpretations of late-childhood onset relational aggression should be made with some reservations. Indeed, considering the evidence that certain manifestations of relational aggression begin in early childhood (Cote et al., 2007; Ostrov, 2010), an early-onset pathway is possible. Nonetheless, that some relational aggressive-victims engaged in early physical and verbal aggression, but most did not, suggests that there is considerable developmental heterogeneity in the etiology of this group.

In contrast to these trends for relational aggressive-victims, rates of transitions from initially uninvolved children to (physical, verbal, and relational) aggressive-victims were very low throughout the developmental periods assessed. Thus, most children in this latter group exhibited early-onset risk profiles, and among children who were initially uninvolved, the likelihood of having a late-onset pathway to (multiple forms of) aggressive-victimization was highly unlikely. Collectively, these findings have important conceptual implications pertaining to how there may be a distinct etiological course which differentiates children who engage in chronic physical aggression with those who are chronic relational (but not physical) aggressors. Findings pertaining to late-onset pathways were further qualified by gender differences. Boys were more likely than girls to exhibit a transition from uninvolved to aggressors. Thus, they were at increased risk for a late-onset pathway to multiple forms of aggression. In contrast, considering that relational aggressive-victims were disproportionately girls indicates that both genders are at risk for late-onset aggression, but manifest different forms of aggression when they exhibit this developmental pathway.

Social interactional continuity

Several transitional effects assessed social interactional continuity perspectives. Among children who were early aggressive-victims, it was expected that co-occurring aggression and peer victimization would reinforce each other, thus promoting continuity into late childhood and adolescence. However, rates of continuity among aggressive-victims varied over time. Perhaps their lower rates of physical victimization in grade 1 attenuated this transactional effect and reduced some of their risks for continuity. More consistent with interactional continuity perspectives, chronicity among relational aggressive-victims was corroborated such that the majority persisted in this group from late childhood to late adolescence. Among aggressors, a small proportion made the transition to aggressive-victims, consistent with social interactional continuity perspectives, and supporting the premise that there are bidirectional and transactional associations between aggression and peer victimization concurrently and longitudinally (Giesbrecht et al., 2011; Ostrov, 2010; Reijntjes et al., 2011).

Heterotypic continuity

Considering that a group of relational aggressive-victims was consistently identified from late childhood to late adolescence, it was possible to investigate patterns of heterotypic continuity by examining the transitions into this group over time. Some evidence for this developmental pathway emerged. For instance, during the transition from early to late adolescence, aggressive-victims showed some risk for becoming relational aggressive-victims. Moreover, a small proportion of aggressors consistently transitioned into relational aggressive-victims. However, the overall support for heterotypic continuity was limited and should be weighed in consideration of several qualifications. First, most of these transition probabilities were small in magnitude. More specifically, during the transition from late childhood to early adolescence, it was highly unlikely for aggressive-victims to become relational aggressive-victims. Second, aggressive-victims and aggressors also had higher rates of relational aggression. Thus, although this transitional pattern indicated that some aggressive children reduced their uses of physical aggression over time, they did not replace relational for physical aggression per se. Third, in light of the gender differences, heterotypic continuity appears to be a more applicable developmental pattern among adolescent girls than boys. Taken together, although the findings consistent with heterotypic continuity support the premise that there is considerable heterogeneity in the development of different forms of aggression from middle childhood to late adolescence, this developmental pattern alone was limited in its predictive utility as the majority of aggressive children exhibited pathways more consistent with the other developmental processes discussed.

Desistance

Finally, across all subgroups, patterns of desistance were pervasive and reflect the premise that many children initially at risk were able to recover from aggression and victimization over time (Barker & Maughan, 2009; Burt, 2012). In contrast to the other perspectives which postulated continuity or onset of risk, desistance pathways accounted for substantial rates of change among the three groups of aggressors. In light of social interactional continuity models which posit that aggression and victimization reinforce each other over time, further research is needed to evaluate the mechanisms by which some children are able to desist from aggressive-victimization, while others are not.

Implications for Intervention

These findings have several implications for intervention efforts aimed to reduce peer-directed aggression and victimization. Considering the prevalence of the risk groups, the need for intervention programs is well-supported. However, the findings raise questions about the ideal timing of such programs. On the one hand, the greater instability that characterized the middle to late childhood years may suggest that this is an ideal time to intervene before children’s risk group membership becomes more established. On the other hand, considering the rates of continuity that emerged in late childhood and adolescence, intervention efforts may also need to be in place throughout these periods to disrupt pathways of risk continuity and enhance the likelihood that children transition into less risky groups. Another question that arises is whether schools should implement universal programs targeting the entire school population or more selective or indicated programs focusing on those most at risk. Given that aggressive-victims showed low rates of desistance and high rates of continuity, they are in need of targeted interventions and support. However, given the risks for late-onset pathways among children who were initially uninvolved, there remains a need for universal prevention.

Limitations and Future Directions

In order to further clarify the extent to which the identified subgroups are qualitatively and etiologically distinct, future studies should continue to explore how other factors, besides gender, are associated with group membership and the developmental transitions between and within these groups. It is plausible that there are many other factors, outside the scope of this study, that contributed to the observed developmental heterogeneity among these risk groups and differentiated (i.e., moderated) patterns of continuity and change over time. Although this study investigated multiple forms of aggression and victimization, there are other behavioral manifestations, in particular of relational forms, which were not assessed. For instance, the measure of relational aggression reflected a more direct form of social exclusion. However, it is conceivable that more indirect and sophisticated forms become more prevalent in adolescence while direct forms begin to occur earlier in development. Thus, patterns of heterotypic continuity may have been more pronounced had the measurement of relational aggression more clearly differentiated these multiple manifestations. Moreover, future investigations should include measures of cyber-aggression which may have implications pertaining to the developmental and heterotypic continuity of aggression. In addition to peer reports, future studies should also incorporate multi-informant data. Because peer nominations measure the extent to which children are identified by peers as being aggressive or victimized, higher scores are not necessarily reflective of greater severity per se; however, it is plausible that children who received more nominations were also likely to engage in aggression, or experience victimization, more frequently and severely. Replication studies in large scale longitudinal samples are also warranted and would validate these findings, given that certain transitional patterns consisted of a small number of children.

Conclusion

This study contributes to ongoing research on the development of different forms of aggression and peer victimization in childhood and adolescence. For the most part, the findings demonstrated that children who were either perpetrators (aggressors) or victims of peer aggression, or both (aggressive-victims), engaged in multiple forms of these behaviors (i.e., physical, verbal, and relational). However, by late childhood, a subgroup emerged (mostly girls), which was more ‘specialized,’ and primarily engaged in and experienced relational forms. Across all groups, intra-individual continuity was observed such that many children maintained the same group status over time. However, multiple transitional patterns emerged which reflected interactional, heterotypic, late-onset and desistance pathways, supporting the premise that the development of aggression and peer victimization is characterized by heterogeneity.