Development of Bullying and Victimization: An Examination of Risk and Protective Factors in a High-Risk Sample

Early Substance and Violence Exposure

Maternal cocaine and other substance use in pregnancy is often a marker for living in high-risk neighborhoods and associated exposure to community violence (Veira et al., 2014). Maternal exposure to violence is also inextricably linked to continued substance use as a coping mechanism (Eiden et al., 2007), and may have cascading effects on children. Indeed, in early childhood, children are in close proximity to their mothers and are more likely to witness maternal experiences of violence that include violence perpetration and victimization. Social-ecological theories of development (Bronfenbrenner, 1979) highlight the salient role of maternal exposure to violence in creating a context for continued aggression and engagement with violence victimization in children. With increasing age, children’s exposure to violence is more likely to involve other contexts. Early childhood exposure to violence is a significant, unique predictor of later outcomes including aggressive behavior and peer victimization (e.g., Schwartz et al., 1997) and across a number of domains in substance exposed cohorts (Buckingham-Howes et al., 2013).

Prior Aggression and Victimization

It is developmentally typical to observe an increase in aggressive behavior up until 24–36 months, with a decline from early to middle childhood (see Eisner & Malti, 2015). However, groups displaying specific aggressive behavioral trends, especially those exhibiting early, chronic elevated levels (i.e., life course persistent; Moffitt, 1993) compared to their peers, are more likely to continue on the trajectory of exhibiting higher levels of physical, direct aggression, and violence throughout development. Prior peer victimization (i.e., the receipt of aggression from peers) and exposure to violence within the home may also increase the risk for bullying (e.g., Bauer et al., 2006) and further victimization (Knous-Westfall et al., 2012), making them important dyadic and broader contextual factors to consider within the social-ecological model. Aggression and victimization are inter-related across development. Engagement in aggression is associated directly with future victimization (e.g., Giesbrecht et al., 2011), and vice versa—victimization may predict general aggression and bullying behavior, potentially as a retaliatory response to victimization (e.g., (Yeung & Leadbeater, 2007)). Therefore, victimization may create socialization contexts which inhibit the development of social skills, promote retaliation, or lead to engagement with delinquent peers (McDougall & Vaillancourt, 2015) and thus increase the likelihood of continued victimization and bullying.

Positive Parenting

Research has documented the ability of positive family factors to buffer children from the impact of stressful life events (Labella & Masten, 2018). Maternal sensitivity is linked to increased social and emotional well-being in children (Kim-Cohen et al., 2004), and has repeatedly been shown to buffer at-risk children from developing externalizing problems. For example, maternal warmth and sensitivity moderated the association between negative environmental and familial factors such as prenatal adversity, maternal substance use during pregnancy, negative discipline (Alink et al., 2009), exposure to intimate partner violence (Manning et al., 2014), and child externalizing behaviors. Prior work with the current sample has supported the moderating role of maternal sensitivity such that children experiencing higher levels of maternal warmth/sensitivity in the context of higher exposure to community violence (Veira et al., 2014) had lower behavioral problems.

Positive Peers

Peers may also serve as a protective factor within the school context for reducing bullying and victimization. Indeed, bystanders challenging a bully may reduce the frequency of bullying (Salmivalli et al., 2011). Levels of bullying and victimization may be low when attitudes at the classroom level support defending victims (Karna et al., 2010) and there is peer pressure to intervene instead of distance from episodes of bullying (Pozzoli & Gini, 2010), underscoring the protective nature of peers. However, the friends of those who are frequently bullied or victimized may not be perceived as popular, dominant, or powerful (Hodges et al., 1997), or may use retaliatory methods to try to reduce the victimization of their friends (Hodges & Perry, 1999). This may increase the risk of future victimization for the defender, and therefore, in some contexts positive peers who have lower social status and are ineffective in defending others may exacerbate risk. Higher social status or preferred children may be more likely to successfully intervene to defend a peer (e.g., Pöyhönen et al., 2010).

School Connectedness

In addition to the proximal protective effects of positive peers, connection to the school context may also be protective. School connectedness refers to experiences of perceived belonging, safety, and support in the school environment (Kim et al., 2020). There are a dearth of longitudinal studies examining school connectedness as a moderator (i.e., protective factor) with respect to bullying victimization and perpetration. Two recent studies found that school connectedness did not function as a protective factor in the longitudinal association between peer victimization and later aggression or between exposure to violence and later psychological functioning (Duggins et al., 2016; Ozer, 2005). More recently however, Kim et al. (2020) found that school connectedness served a protective role in moderating the relation between electronic forms of bullying victimization and suicide risk behaviors for adolescents. Research on the role of school connectedness as a protective factor in preventing perpetration and victimization within contexts of high risk (under-resourced schools, violent neighborhoods) is needed.

Participants

The sample included 216 caregiver–child dyads (49% or 106 boys) participating in a longitudinal study from the child’s birth, recruited from two urban hospitals serving a predominantly Black/African-American low-income population. Adolescents self-identified their race/ethnicity as 57.6% Black/African American, 17.9% Multiple Races, 7.9% Hispanic/Latinx, 6.6% White, 2.0% American Indian, 0.7% Asian, and 7.3% other during EA and as 62.4% Black/African American, 16.5% Multiple Races, 6.4% Hispanic/Latinx, 7.3% White, 2.8% American Indian, and 4.6% other during LA. Biological mothers ranged from 18 to 42 years old (M = 29.53; SD = 6.06) at recruitment. The majority (76%) were receiving Temporary Assistance for Needy Families, 66% were single, 69% were on Medicaid, 27% were unemployed, and 71% had high school or below education. About 72% of mothers were Black/African-American, 8% Hispanic/Latinx, 19% White, and 1% other.

All mothers were screened post-delivery to identify a sample of participants with high rates of prenatal cocaine exposure (PCE). Exclusionary criteria consisted of maternal age < 18 years, use of illicit substances other than cocaine or marijuana during pregnancy, multiple births (e.g., twins), and significant infant medical complications/issues. The sample included 116 infants exposed to cocaine and other substances (alcohol, cigarettes, marijuana) in utero, and 100 not exposed to cocaine but the majority using substances other than cocaine (Eiden et al., 2014). Caregiver-infant dyads were group matched on maternal race/ethnicity and education (Eiden et al., 2014). As part of a prospective longitudinal study of risk and resilience, caregiver-infant dyads were assessed beginning at 1 month of infant age (corrected for prematurity), with follow-up assessments every 6 months until and including when the participants were in kindergarten (60 months), in second grade (84 months), in EA (M = 13.26 years, SD = 0.83), and LA (M = 15.08 Years, SD = 0.83). Children born at or before 37 weeks gestation were scheduled for appointments at chronological age corrected for prematurity until the 24 month assessment.

Measures

Early and Late Adolescence

Violence Victimization Self-Report. The Survey of Exposure to Community Violence Checklist (CECV) is a 35-item measure (Richters & Saltzman, 1990) completed in EA. The CECV measures the child’s exposure to several types of severe violence (e.g., experiencing or witnessing assault, stabbings, or robberies) on a five-point Likert scale from 0 (never) to 4 (> 10 times). A sum assessed total exposure to violence and was reliable (Cronbach’s α = .89). Self-Report of Bullying and Victimization. In EA and LA, bullying victimization and perpetration in the previous 30 days were assessed with the California Bullying Victimization Scale (Felix et al., 2011). Each subscale consists of eight items assessing the frequency of bullying victimization or perpetration on five-point Likert scale from 1 (not in the past month) to 5 (several times a week). Scores in the current study were obtained by averaging across items (Hase et al., 2015), with Cronbach’s αs = .76/.78 for bullying victimization and .73/.62 for perpetration in EA/LA. School Connectedness. School connectedness was measured in EA using a six item self-reported scale. Items such as “I feel part of my school” and “I feel that teachers treat student fairly at school” were measured on a five-point Likert scale from 1 (strongly agree) to 5 (strongly disagree). All items were then reverse coded so that higher scores indicated higher levels of school connectedness. Internal consistency of the scale was good, Cronbach’s α = .78. Positive Peers. The extent to which adolescents had a positive influence peer group was measured using the prosocial peers scale (Fredricks & Eccles, 2005), which included six items such as “How many of your friends encourage you to do your best in school?” “How many of your friends plan to go to college?” and “How many of your friends think it is important to do volunteer work in their community.” Adolescents self-reported on their peer group using a seven-point Likert scale from 1 (none of my friends) to 7 (all of my friends), thus higher scores indicated a more positive peer group. Internal consistency of the scale was good, Cronbach’s α = .78.

Procedures

The procedures were approved by the local university IRB (see Eiden et al. (2014) for additional details on this longitudinal sample and procedures). Informed active written consent (and child/youth assent) was obtained from all recruited participants, and they were compensated for their time. All assessments were conducted with the primary caregiver of the child at that time, although for ease of presentation the terms mother and maternal are used throughout when referring to the primary caregiver, as mothers were the typical respondent. Specifically, at least 84% of respondents were biological mothers at each timepoint where parental data is used, with a smaller minority of foster caregivers or relatives. Biological mothers were interviewed at the 4 to 8 week assessment in addition to the foster mothers in order to obtain accurate information about prenatal substance use. The primary caregiver was identified as the adult who had legal guardianship of the child and accompanied the child at all appointments.

Data Analysis

First, descriptive data and correlations of the measures were obtained. Outliers were modified by adjusting the value to +/- three SDs from the mean. For key study variables, skew values (-1.76–2.21) and kurtosis values (-0.44–4.67) were elevated but within accepted ranges for normally distributed variables (Kline, 2015).

As with all longitudinal studies and especially due to the high-risk nature of the sample, missing data were expected. There were complete data for 90.3% of participants in early childhood, 74.1% of participants in middle childhood, 70.4% of participants in EA, and 66.6% in LA. Results from χ² tests and ANOVAs demonstrated that early non-biological parent care status was related to missing data in EA [χ² (1) = 7.46, p = .01, Cramer’s V = .19] and LA [χ² (1) = 13.73, p < .001, Cramer’s V = .25]. Additionally, adolescent non-biological care status was related to missing data in LA [χ² (1) = 7.97, p = .01, Cramer’s V = .23]. Missing data at earlier timepoints was not related to any demographic or key study variable. Taken together, this indicated that the data were most likely Missing At Random (MAR) and the identifiable variables related to missingness were included in model testing (Little, 2013).

All models were estimated in Mplus version 8.4 (Muthén & Muthén, 1998-2020). Maximum likelihood estimation with robust standard errors (MLR) was used due to a slight skew in some of the variables. Missing data were accommodated by using full information maximum likelihood (FIML). Several demographic covariates (child sex, SES adversity, early and adolescent non-biological care, maternal children, maternal age, and adolescent age) were considered, but were only included in the model if they were correlated with outcomes at p < .10. A path analysis was used to test the main hypotheses, with EC variables regressed on the covariates, MC variables regressed on EC variables, EA variables regressed on MC variables, and LA variables regressed on EA variables. Modification indices (MI) were examined to see if any direct paths from earlier time points to EA or LA variables could be added. After examining the initial model, indirect effects were tested using bias corrected bootstrap confidence intervals in conjunction with standardized indirect effect estimates (i.e., Preacher & Kelley, 2011). Next, the moderators and interaction terms were added to the model. All pertinent variables were standardized prior to examining moderation. Given the number of parameters, there were issues with model convergence when including all moderators in the model so two models were run, the first with maternal sensitivity as a moderator of EC to MC paths and the second with the EA protective factors as a moderator of EA to LA paths.

A final model included significant interaction terms and their lower order components, controlling for the other interaction terms with those moderator variables. Interactions from this model were then probed at low (i.e., one SD below the mean), average, and high (i.e., one SD above the mean) values of the moderator (Preacher et al., 2006). Maternal sensitivity was not tested as a moderator of the path from early childhood maternal harshness to middle childhood aggression given that maternal sensitivity and harshness were measured using the same observational paradigm at the same assessment period and therefore, had a high level of overlap.

A likelihood ratio χ² test was used to test overall model fit where p > .05 indicated good model fit. The following alternative fit indices were also considered: (a) comparative fit index (CFI), where values greater than .95 suggest good fit, (b) standardized root mean-square residual (SRMR) where values less than .08 represent mediocre fit, and (c) root mean square error of approximation (RMSEA), where values less than .08 suggest mediocre fit (Hu & Bentler, 1999).

Preliminary Analyses

Bivariate correlations and descriptive statistics of the continuous covariates and key variables are provided in Table 1. Several correlations add support to the validity of the constructs (e.g., teacher-reported aggression in fifth grade and self-reported bullying behavior in early adolescence). In addition, child sex was related to MC aggression [F (1, 158) = 5.29, p = .02, Adjusted R² = .03], such that males (n = 79; M = 7.17, SE = 0.72) had higher aggression scores than females (n = 81, M = 4.85, SE = 0.71). Additionally, adolescent foster care status was related to EA victimization [F (1, 147) = 4.02, p = .02, Adjusted R² = .02], such that adolescents in non-biological care endorsed higher rates of bullying victimization than those in biological care. Child sex, prenatal substance use, SES adversity, early foster care status, adolescent foster care status, and EA age were included in the model as covariates (see Table 1). EA control variables were regressed on earlier control variables.

OPEN IN VIEWER

Table 1.

Descriptive Statistics and Bivariate Correlations of Key Study Variables.

	1	2	3	4	5	6	7	8	9	10	11	12	13
1. Maternal harshness, 7 to 36 months-OBS
2. Maternal PTSD symptoms, 7 to 36 months-MR	.08
3. Maternal violence vict, 1 to 36 months-MR	.01	.33**
4. Pure physical aggression, 18 to 48 months-MR	.14†	.32**	.15*
5. Aggression, grade 2-TR	.15†	.10	.05	.07
6. Bullying, EA-SR	.01	.09	−.01	.14†	.29**
7. Bully vict, EA-SR	−.03	.20**	−.03	.12	.09	.45**
8. Violence vict, EA-SR	.08	.13	.06	.11	.18*	.24**	.37**
9. Bullying, LA-SR	−.15†	−.05	.003	−.07	.17†	.31**	.28**	.29**
10. Bully vict, LA-SR	−.03	.18*	−.09	.04	.09	.02	.26**	.24**	.30**
Covariates
11. SES adversity	.23**	.05	.03	.04	.08	.03	.05	.01	.01	−.05
12. Prenatal maternal substance use	.17*	.18*	.21**	.05	−.13	−.03	.01	.02	.00	.02	.23**
13. EA age	−.07	−.18*	−.20*	−.11	−.07	−.05	−.14†	.09	.02	−.03	−.04	.19*
M	1.46	10.97	1.08	2.96	5.99	1.19	1.31	7.56	1.16	1.30	0.64	4.48	13.26
SD	.59	14.56	1.29	1.69	6.46	0.30	0.42	6.01	0.27	0.45	0.25	3.06	0.82
Range	.50–3.30	0–57.38	0–4.95	0–7.35	0–25.50	1–2.34	1–2.70	0–23.31	1–2.03	1–2.78	0–1.00	0–13.00	12–15.00

Primary Analyses

The structural model provided a good fit to the data [χ² (54) = 67.43, p = .10, CFI = .95, RMSEA= .03, SRMR = .06; see Figure 1 for estimates and effect sizes]. MI were examined to see if any direct paths from earlier time points to EA or LA variables could be added. The path from early maternal harsh parenting to LA bullying had the highest MI (MI = 4.94). A model with this path freed improved model fit significantly [Δχ²(1) = 10.34, p = .001], and this model provided a good fit to the data [χ² (53) = 61.95, p = .19, CFI = .96, RMSEA = .03, SRMR = .06]. No other significant MI remained. Of the EC variables, maternal harsh parenting was associated with MC aggression (p = .003). MC aggression was then subsequently related to higher levels of EA violence victimization (p = .01) and bullying (p = .001). EA violence victimization was then associated with higher levels of bullying (p = .003) and bullying victimization (p = .04) in LA. There was also stability between EA and LA bullying victimization (p = .02) and bullying (p = .03). Finally, there was a direct negative association from EC harsh parenting and LA bullying (p = .01). The significance of longitudinal indirect cascading paths was examined. The total indirect path from EC harsh parenting to LA bullying was significant [β = .02, p = .04, 95% CI (.001, .05)] although the specific indirect paths through MC aggression and EA bullying [β = .01, p = .11, 95% CI (−.003, .03)] and MC aggression and EA violence victimization [β = .01, p = .05, 95% CI (−.004, .03)] were not significant. The indirect path from EC harsh parenting to EA bullying through MC aggression was significant [β = .06, p = .01, 95% CI (.01, .10)]. Bootstrapping was not available with the MLR estimator. The indirect effects were re-tested using bootstrapping with the maximum likelihood (ML) estimator and yielded the same findings.

OPEN IN VIEWER

Figure 1.

Main Structural Model. Note. MR = maternal report, OBS = observation, SR = self-report, Vict = victimization, m = months, PTSD = Posttraumatic stress. One-month socioeconomic adversity, non-biological care status through kindergarten, early adolescence non-biological care status, prenatal maternal substance use, and early adolescence age were controlled. *p < .05. **p < .01.

Next, the interaction terms were tested. First, a model with EC maternal sensitivity moderating the paths from the EC variables to MC aggression was tested. A significant interaction emerged for EC aggression and maternal sensitivity (b = -1.12, SE = 0.48, p = .02) and for EC maternal PTSD symptoms and maternal sensitivity (b = 0.14, SE = 0.06, p = .02) in predictions of MC aggression. Next a model was tested with EA school connectedness and positive peers moderating links between EA variables to LA bullying victimization and bullying. There was a significant interaction between EA violence victimization and positive peers in predicting LA peer victimization (b = 0.11, SE = 0.05, p = .02) but none of the school connectedness interaction terms were significant. School connectedness and its associated interaction terms were dropped from the final model.

A final model was examined with the maternal sensitivity and positive peer interaction terms, which provided a good fit to the data [χ² (152) = 154.06 p = .44, CFI = .99, RMSEA= .01, SRMR = .06]. There was a nonsignificant trend for stability between EC aggression and MC aggression for mothers low on maternal sensitivity (b = 1.57, SE = 0.88, p = .07). For mothers high on maternal sensitivity there was a nonsignificant trend for a negative relation between early and later child aggression (b = -1.20, SE = 0.71, p = .09). There was no evidence of stability when maternal sensitivity was average (b = 0.19, SE = 0.53, p = .72). Notably, at moderately low levels of maternal sensitivity (i.e., 1.5 SDs below the mean) there was significant stability (b = 2.27, SE = 1.13, p = .046) and at moderately high levels of maternal sensitivity (i.e., 1.5 SDs above the mean) there was a negative relation between EC and MC aggression (b = -1.89, SE = 0.94, p = .04), underscoring the potential protective nature of maternal sensitivity.

The interaction between EC maternal PTSD symptoms and maternal sensitivity on MC aggression was also probed. There was no relation between maternal PTSD symptoms and MC aggression at low (b = -0.56, SE = .93, p = .55), or average levels of maternal sensitivity (b = 0.90, SE = .65, p = .17), but at high levels of maternal sensitivity, EC maternal PTSD symptoms positively predicted MC aggression (b = 2.36, SE = .84, p = .005).

Finally, the interaction between EA violence victimization and positive peers was probed. At low levels of positive peers (b = -0.02, SE = .06, p = .76) and at average levels of positive peers (b = 0.07, SE = .04, p = .10), there was no relation between EA violence victimization and LA bullying victimization. At high levels of positive peers, there was a positive relation between EA violence victimization and LA bullying victimization (b = 0.15, SE = .06, p = .01), suggesting that in the context of high levels of community violence, high levels of positive peers (i.e., they value academic pursuits) may exacerbate a child’s risk for subsequent bullying victimization.

Limitations

Despite the noted strengths of this longitudinal study with multiple time points, comprehensive prospective assessments from multiple methods and informants, and relatively high retention rate for an at-risk sample, there are several limitations. First, given the gap in assessments between MC and EA, there is the possibility of unmeasured intervening processes that the present study could not test. Moreover, even though we did control for earlier aggression and related behaviors, we were unable to control for earlier (prior to EA) bullying perpetration and victimization, which may undermine our ability to appropriately model change in these constructs. It should be noted that the reliance on self-report of bullying and bullying victimization for both EA and LA assessments may have been appropriate to support valid assessments, but raises possible concerns about shared method variance. In addition, the failure to examine the unique form and function of victimization and bullying behavior may have undermined our ability to see gender-linked pathways and is a clear direction for future work. Prior research has demonstrated that mothers’ and fathers’ behavior (e.g., parental monitoring) may have differential effects on bullying experiences among African-American adolescents (Hong et al., 2020), and future work will be needed to replicate the present findings with fathers. Finally, we did not have an assessment of the quality of the relationship with positive peers within this project and that may have impacted our ability to test the role of this moderator.

Future Directions

Importantly, this work demonstrates the power of maternal sensitivity in early childhood in promoting resilient outcomes. Future intervention work should continue to explore the impact of programs that foster warm, responsive, and sensitive caregiving approaches among families at-risk and especially those with substance exposure. Given our findings about harsh parenting in EC having a cascading influence on later aggression, violence victimization, and bullying perpetration, yet a negative direct relation from harsh parenting to late adolescent bullying, future longitudinal studies with high-risk and racially and ethnically diverse samples are needed to study multiple dimensions of parenting (e.g., power-assertive, harsh, consistent enforcement of rules, and monitoring) across development to disentangle these effects. Future studies with high-risk samples exposed to violence victimization should also assess the moderating effect of peers by assessing this construct in more depth (e.g., quality of these relationships, peers’ social status and dominance, prosocial and delinquent behavior) to more fully understand whether they may serve as risk or protective for bullying victimization. Finally, although we statistically controlled for changes in caregiving situations in this sample, it is possible that changes in primary caregivers over time or stability of maladaptive care situations conferred unique risk for a subset of children. Person centered analyses examining changing contextual risks and how they relate to adolescent outcomes may promote greater understanding of these processes and their relations with violence and bullying victimization and perpetration.

Implications for Policy and Practice

The present study has several policy and practice implications. First, a clear clinical and policy implication of the moderating role of maternal sensitivity is the importance of supporting and fostering positive parenting skills and behavioral interventions among families that may be at-risk (e.g., substance exposed) for problematic parent–child interactions and to do so as early as possible. Moreover, this work highlights the long-term influence of early maternal sensitivity for supporting resilient pathways and underscores the importance of assessing for positive/protective variables in studies and clinical practice with children that are probabilistically at risk for maladaptive outcomes. Second, knowing that experiences during the early months of life (e.g., caregiver harshness) directly predict subsequent aggressive behavior also provides justification for early intervention to reduce caregiver harshness and promote positive parent-child interactions. Third, given the current documented stability of aggressive behavior over a long period of development, our findings support early identification and intervention for aggressive behavior and in particular for physically aggressive behavior that occurs during a period in development when it is no longer normative (i.e., after 60 months; NICHD ECCRN, 2004). Finally, exposure to violence early in life and community violence later during development were associated concurrently and prospectively with bullying and bullying victimization and thus policymakers must understand that efforts to alleviate community violence have cascading influences on children and may help support resilient pathways.