Early Childhood Factors Associated With Peer Victimization Trajectories From 6 to 17 Years of Age

This study is based on the QLSCD, a population-based birth cohort that tracks the development of 2120 children born in the Canadian province of Quebec in 1997–1998 and followed-up until 2015. The sample was drawn through a stratified sampling procedure on the basis of living area and birth rate from the Quebec Master Birth Registry. All mothers giving birth after 24 weeks’ and not later than 42 weeks’ gestation who spoke English or French were eligible. Detailed information on the QLSCD can be found elsewhere.³⁵  The QLSCD protocol was approved by the Institut de la statistique du Québec and the Research Centre of the Sainte-Justine University Hospital ethics committees. Written informed consent was obtained from all participating families at each assessment. The person most knowledgeable about the child (the mother in 98% of the cases) provided data about the child, family, and broader social context at 5 months, 1½, 2½, 3½, 4½, and 5 years after birth through home interviews. The fathers (biological fathers who had contact with the child at least once a month or mother’s partner living in the household) also provided information through a self-administrated questionnaire.

The analytical sample in this study consists of 1760 children followed-up from 5 months to 17 years of age who reported their peer victimization experience at least once between 6 and 17 years: 862 boys (49.0%) and 898 girls (51.0%). More than half of the participants (n = 1038; 59.0%) provided information about peer victimization on >6 waves (ie, 7 or 8 out of 8 assessments) (Supplemental Table 4). The characteristics of the participants included in this study are presented in Table 1.

When the children were aged 6, 7, 8, 10, 12, 13, 15, and 17 years, information on peer victimization was collected by using 6 items of a modified version of the Self-reported Peer Victimization Scale (Supplemental Information).³⁶  Participants reported how often they experienced physical (ie, “pushed, hit or kicked”), verbal (ie, “called names, insulted, said mean things to you,” “teased you in a mean way/made fun of you”), relational victimization (ie, “did not let you be part of his or her group,” “said bad things about you to other children”), and property attacks (ie, “forced you to give something that belonged to you/made you pay them or give them something so they would leave you alone”) (response range: 0 = never, 1 = once or twice, 2 = more often). The wording of the items was adapted to reflect changes in the experience of victimization that could occur with age (eg, the item “did not let me play with his or her group” used when participants were aged 6–12 years was changed to “did not let me be part of his or her group” when children were aged 13 years or older). At each wave, if participants answered at least 4 out of the 6 questions of the Self-reported Peer Victimization Scale, we calculated the mean of the items (range 0–2) and considered the data missing otherwise. The mean score at each wave was rescaled (multiplied by 5) to range from 0 to 10 (with a higher score indicating a higher level of peer victimization). Cronbach’s α ranged from 0.74 to 0.81 across ages.

In Table 2, we provide the description of the measures used to assess family sociodemographic characteristics, parental mental health, parent-child relationship, and children’s behavior. A comprehensive list of the items used to derive the early childhood measures is available (Supplemental Table 5).

The excluded participants were more likely to be male, be of non-Canadian origins, come from socioeconomically disadvantaged families, have a mother with higher depressive symptoms, and have overprotective parents compared with participants retained in the study (Supplemental Table 6). Therefore, these variables were used to derive weights that were applied in all regression models by using the inverse probability weighting procedure. The proportion of missing data was <3.5% for the majority of the variables with the exception of father psychopathology (13.4%) and father-child relationship (20%). To avoid a loss of participants due to listwise deletion, the multivariate models were estimated by using multiple imputation by chained equations (n = 50 data set).

We used group-based trajectory modeling^43,44  to estimate the developmental trajectories of peer victimization from 6 to 17 years of age. Group-based trajectory modeling, a special case of finite mixture models, identifies clusters of individuals who follow similar developmental trajectories. The best-fitting model was identified by estimating models with 2 to 8 latent clusters with quadratic age terms and comparing them by using the Bayesian Information Criterion (BIC) as the primary fit index. As recommended, the size of the clusters was also considered to select the best model (no solution with small group sizes, ie, <5% of the sample, was selected). We assessed the quality of the classification identified by the model using the average posterior probability of cluster membership (good if >0.70 for each trajectory).

In a first step, we used univariate multinomial logistic regression models to estimate the association between trajectory membership and each early childhood variable separately. In a second step, to estimate the unique contribution of each variable over and above the effect of the other variables, we ran multivariate multinomial logistic regression models. We entered in the multivariate model all variables that revealed a significant association at P < .05 with any of the trajectories relative to the reference trajectory in the univariate models. The trajectory with the lowest levels of peer victimization was used as reference category in all the multinomial logistic regression models.

We performed 2 separate subgroup analyses to compare the high-chronic trajectory to the moderate-emerging and childhood-limited trajectories.

We identified 4 distinct developmental trajectories of self-reported peer victimization from 6 to 17 years of age (Fig 1): (1) low peer victimization across the entire period (n = 579; 32.9%); (2) moderate-emerging peer victimization, characterized by steady levels of victimization from age 6 to 12 years and the second highest level of victimization across adolescence (n = 525; 29.8%); (3) childhood-limited peer victimization, characterized by a relatively high level of victimization at age 6, followed by a progressive sharp decline from age 6 to 17 years, and virtually no victimization at age 17 (n = 461; 26.2%); and (4) high-chronic peer victimization, characterized by persistently higher levels of victimization relative to the other groups despite a decline from age 6 to 17 years (n = 195; 11.1%). The fit indices of the models with 2 to 8 trajectories that were compared to determine the optimal solution are presented in Supplemental Table 7.

Univariate analyses revealed that early childhood behavior and family characteristics were associated with peer victimization development (Table 1). Similar to the univariate analyses, in multivariate analyses, we showed that compared with the children following a low trajectory of peer victimization, children in the 3 other trajectories were more likely to exhibit higher levels of externalizing symptoms. Additionally, children following a moderate-emerging or a high-chronic trajectory of peer victimization, compared with those in the low victimization trajectory, were more likely to be boys and have a father with a history of antisocial behavior. Finally, children following a childhood-limited or high-chronic peer victimization trajectory were more likely to come from nonintact families (Table 3).

The associations for maternal and paternal depression and parenting as well as for socioeconomic disadvantage observed in the univariate models were not statistically significant when accounting for children’s behaviors and the other family characteristics in multivariate models (Table 3). The level of internalizing behavior in early childhood and the maternal history of antisocial behavior were similar for children across the 4 peer victimization trajectories (Table 1).

Children in the high-chronic relative to the moderate-emerging and childhood-limited trajectories were more likely to exhibit higher level of externalizing symptoms in early childhood, controlling for other behaviors and family factors. Additionally, children in the high-chronic trajectory were more likely to be boys and have a father with a history of antisocial behavior compared with those in the childhood-limited trajectory and come from a nonintact family relative to those in the moderate-emerging trajectory (Supplemental Table 8).

This was the largest study to describe the developmental trajectories of peer victimization from 6 to 17 years of age and document their associations with early childhood behavior and family characteristics.

We identified 4 distinct peer victimization trajectories: low, moderate-emerging, childhood-limited, and high-chronic. Although the majority of children reported some level of peer victimization at school entry, all groups except the moderate-emerging group reported declining levels in middle childhood. The pattern of severity and stability of peer victimization, the relative size of the low and childhood-limited peer victimization groups, and the higher proportion of boys in the trajectories characterized by persistent peer victimization were findings similar to those described by Ladd et al¹⁶  over the same ages (ie, 6–17 years). Thus, the striking similarities between these 2 studies done in distinct North American cultural settings suggest that they both captured general patterns of perceived peer victimization development throughout the cycle of mandatory education. Moreover, these 2 studies indicate that middle childhood is a period of substantial differentiation in the development of peer victimization. That is, more than half of the children exhibited a change in the rank ordering of the peer victimization group. The childhood-limited group reported the second highest level of victimization at 6 years of age and had, together with the low group, the lowest levels at 17 years of age. In contrast, the moderate-emerging group reported the second lowest levels of victimization at 6 years of age and the second highest level after the chronic group at 17 years of age. Our findings from 12 years of follow-up across childhood and adolescence strengthen the evidence about the existence of primary school–limited and late-onset peer victimization, which was theoretically described⁸  or empirically derived^6,11  in short-term longitudinal studies.

We showed that paternal history of antisocial behavior was associated with persistent peer victimization (ie, high-chronic and moderate-emerging trajectories) when controlling for children’s sex, behavior, maternal factors, parenting, socioeconomic disadvantage, and family structure. This was the first study reporting on the relationship between the fathers’ mental health and the development of peer victimization in the offspring. However, our findings are in line with evidence from studies that revealed that paternal negativism³⁴  and hostility⁴⁵  are associated with peer victimization and bullying, respectively, and with studies on the association between the father’s psychopathology and the offspring’s behavioral problems.^31,46  Furthermore, in line with Brendgen et al,¹³  we showed that living in a nonintact family was associated with high levels of peer victimization at school entry (ie, high-chronic and childhood-limited trajectories). The father’s antisocial behavior distinguished between children in these 2 trajectories. That is, children who escaped high levels of peer victimization in the first years of primary school (ie, childhood-limited trajectory) had a father with better mental health than those who continued to be highly victimized during adolescence (ie, high-chronic trajectory). These findings strengthen the importance of paternal mental health for high-chronic peer victimization.

A genetically informative study has revealed that a father’s antisocial behavior may influence children’s behavioral problems through both genetic and environmental pathways.⁴⁷  Previous twin studies have indicated that genetic factors accounted for an important part of the variation in persistent peer difficulties.^8,22  Moreover, a polygenic risk score study revealed that high genetic risk for mental health problems was associated with increased exposure to bullying.⁴⁸  Future studies are needed to clarify the association between a father’s mental health and the offspring’s persistent peer victimization (eg, genetic, environmental mechanisms).

Consistent with previous research,^7,8,21,24  we found that high externalizing behavior problems during the preschool years were important factors for the development of peer victimization. Children who exhibited the highest levels of externalizing behavior during early childhood endured the highest levels of peer victimization from 6 to 17 years of age. These findings, taken together with the overrepresentation of boys and fathers with a history of antisocial behavior in the trajectories characterized by persistent peer victimization, echo the literature on the profile of bully-victims.²⁷  Similar to other studies among young children,^7,21,24  we found that children in the different trajectories of peer victimization had similar internalizing symptoms before school entry. These findings differ from those among older children and adolescents⁴⁹  probably because internalizing symptoms become more negatively perceived by peers and associated with peer victimization as children grow older.⁹ 

The findings from this study need to be interpreted in the context of its limitations. First, we did not assess the power imbalance between the bully and the victim, which is part of the definition of bullying. However, students’ definition of bullying tends to focus on negative actions by peers and fails to include power imbalance.^50,51  Second, we did not differentiate between children who are only victimized and those who are simultaneously bullies and victims. Thus, the experiences of peer victimization described in this study also capture the experience of bully-victims. Third, we measured peer victimization using self-reports. Despite the advantages of this assessment method in long-term studies (see Ladd et al¹⁶ ), self-reported peer victimization is potentially biased by the self-system, which may be less differentiated and related to actual experiences in younger children.^22,52  Fourth, 83% of the baseline sample was available for the 17-year follow-up. To minimize attrition bias, analyses were conducted by using weights accounting for the probability of being retained in the study at follow-up. To minimize the loss of participants in multivariate models due to listwise deletion, we used imputations. Results with and without weights and imputations were fairly similar, strengthening the internal validity of the study.

These limitations notwithstanding, this is the largest and one of the longest population-based studies to have applied a longitudinal person-centered approach to analyze repeated measures of peer victimization. The external validity of our results is reinforced by the reproduction of the peer victimization trajectories between 6 and 17 years of age described by Ladd et al¹⁶  despite the use of a different statistical method. Moreover, this study is unique through the description of both maternal and paternal factors associated to peer victimization development.

In this study, we identified 4 different developmental patterns of peer victimization across the entire cycle of mandatory education, primarily distinguished by their development during primary school. Some children experienced persistent peer victimization already in the first years in primary school. Early childhood externalizing behaviors and family vulnerabilities were associated with the development of peer victimization. To prevent persistent peer victimization, children who are victimized should be offered targeted interventions that address these individual and family vulnerabilities early in their school careers.

We thank the children and parents of the QLSCD and the participating teachers and schools for their continued commitment to the study, the Institut de la statistique du Québec, and the Research Unit on Children’s Psychosocial Maladjustment for their support in data collection and management.

BIC

Bayesian Information Criterion

QLSCD

Quebec Longitudinal Study of Child Development