10155022438936Psychol ViolencePsychol ViolencePsychology of violence2152-08282152-081X30151328610808910.1037/vio0000210HHSPA982012ArticleExamining Within-Person and Between-Person Associations of Family Violence and Peer Deviance on Bullying Perpetration among Middle School StudentsGrantNickholasMAUniversity of Illinois, Dept. of Psychology, 603 East Daniel Street, Clinical and Community Division, Champaign, IL 61820-6925, Phone: 217-333-0631MerrinGabriel J.PhDUniversity of Victoria, Dept. of Psychology, 3800 Finnerty Road (Ring Rd), Victoria, British Columbia V8P 5C2, Phone: (250)-472-4823KingMatthew T.MSUniversity of Illinois, Dept. of Educational Psychology, 216 Psychology, 1310 S. Sixth Street, Counseling Psychology Division, Champaign, IL 61820-6925, Phone: ?EspelageDorothy L.PhDUniversity of Florida, Dept. of Psychology, 945 Center Drive, Gainesville, FL 32611-2250, Phone: 352-273-21392572018282018120190112020911827Objective

Family violence and peer deviance have shown to be related to bullying perpetration. Although there are several cross-sectional investigations of these two factors in relation to bullying behavior, no known studies have examined their interactive associations. The current study examines the longitudinal associations of both factors on bullying perpetration using a multi-level approach.

Method

Participants included 1,194 5th, 6th, and 7th grade students from four middle schools in a Midwest county. We examined the main and interactive relations between how individual reports of family violence and peer deviance fluctuated over time (i.e., within-person effects) and how average reported differences between individuals (i.e., between-person effects) were associated with levels of bullying perpetration.

Results

Positive main effects were found for both family violence and peer deviance on levels of bullying perpetration. Within-person effects indicated that, on average, fluctuations from one’s ‘typical’ levels in family violence and peer deviance were associated with contemporaneous increases in bullying perpetration. A statistically significant time-variant interaction revealed that within-person family violence significantly exacerbated the relationship between within-person peer deviance and bullying perpetration. Furthermore, a statistically significant cross-level interaction revealed that the association between within-person peer deviance and bullying perpetration was stronger for individuals with higher average levels of between-person family violence (+1SD) compared to lower levels (−1SD).

Implications

These findings provide a more nuanced lens from which to view the co-occurring relations between family and peer ecologies. Prevention and intervention efforts should target peer relations to reduce the effect of family violence on bullying behavior.

BullyingPeer DevianceDelinquent PeersFamily ViolenceDomestic Violence

Bullying is a pervasive issue today: approximately 21% of students between the ages of 12–18 reported being bullied in 2015 (Musu-Gillette et al., 2018). Bullying perpetration is a multifaceted phenomenon that is characterized as the intentional, unsolicited, and repeated use of physical (e.g., hitting, kicking, pushing, shoving), verbal (e.g., name-calling, teasing), and/or social (e.g., spreading rumors, social exclusion) aggression toward one’s peers to inflict physical, psychological, social, or educational harm (Gladden, Vivolo-Kantor, Hamburger, & Lumpkin, 2014; Olweus, 1997). The relationship between perpetrator and victim is often characterized by an imbalance of strength and power, used to coerce a person of lesser strength or status (Olweus, 1993). If not prevented early, children are at risk for developing psychopathology in adulthood (Copeland, Wolke, Angold, & Costello, 2013).

Over the last four decades, scholars have found that bullying is particularly prevalent in educational settings such as pre-school, primary, and secondary institutions (Hymel & Swearer, 2015). Depending on the student’s race/ethnicity, gender, and age, prevalence rates can vary. For instance, studies found race and ethnicity differences in levels of bullying perpetration, with African-American/Black and Latino youth reporting higher involvement in comparison to their white counterparts (Albdour & Krouse, 2014). Likewise, males have reported bullying others more frequently than females (Pellegrini & Bartini, 2000). Moreover, studies have found significant grade differences, with much higher levels of bullying observed in middle and secondary institutions than in elementary institutions; yet, levels of bullying others tend to decrease once at the secondary level (Espelage, Hong, Rao, & Thornberg, 2015; Pellegrini & Long, 2002; Pellegrini & Bartini, 2000).

One possible explanation for the development of bullying perpetration comes from social learning theory, which suggests that “most human behavior is learned observationally through modeling: from observing others, one forms an idea of how new behaviors are performed, and on later occasions this coded information serves as a guide for action” (Bandura & Walters, 1977, pg. 22). Specifically, in relation to physically aggressive and violent behavior, children who are exposed to violent behaviors within their immediate family and/or peer contexts may learn and accept such behavior as an appropriate method for dealing with conflict or reaching a desired goal (Akers, 2011; Bandura, 1973). Indeed, longitudinal studies have shown that exposure to family violence is related to high levels of bullying behavior (Bauer et al., 2006; Knous-Westfall, Ehrensaft, MacDonell, & Cohen, 2012; Low & Espelage, 2014); however, peer deviance has only demonstrated this association in cross sectional studies (Espelage, Bosworth, & Simon, 2000; Ferguson, Miguel, & Hartley, 2009). As such, there has been no examination of the longitudinal, joint associations of exposure to family violence and peer deviancy on levels of bullying perpetration to date. The current study aims to provide a better understanding of these joint associations across a middle school sample.

Family Violence, Peer Deviance, and Bullying

According to the United States Department of Justice (2015), family violence (also known as domestic violence) is defined as any pattern of violent, threatening, coercive or controlling behavior, including physical injury, direct or indirect threats, sexual assault, emotional and psychological abuse, damage to property, economic control, or social exclusion against current or past family members, a current or former intimate partner, or between individuals within a shared living space.

A recent review by Voisin and Hong (2012) highlighted several cross-sectional studies, yet, few longitudinal studies have found associations between forms of family violence and bullying behavior. For example, Knous-Westfall and colleagues (2012) conducted a 25-year longitudinal study investigating the impact of parental intimate partner violence (IPV) on relational and physical bullying and victimization across three generations. The authors measured both minor forms of physical IPV (e.g., hitting and grabbing) and psychological violence (e.g., physical threat) and more severe measures of IPV (i.e., violence that resulted in serious injuries). When controlling for externalizing and internalizing problems, they found that minor forms of IPV did not significantly predict physical bullying or relational peer bullying; however, they did find that exposure to severe IPV was significantly predictive of physical bullying in males. Another prospective longitudinal, multi-generational study conducted by Bauer and colleagues (2006), examined the relationships among IPV, physical and relational forms of bullying, internalizing behavior problems, and social attention problems in children. They found that physical bullying was related to IPV based on parent report.

Outside the family context, peer relationships also play an important role throughout adolescent development, especially during the transition to middle school. According to Pellegrini and Long (2002), it is a critical period in early adolescent development where youth begin to explore their social identity and status and may engage in negative peer relationships, which can influence problem behaviors (Salmivalli, 2010). For instance, scholars have found that peer deviance—meaning to have peers who engage in criminal (e.g., theft, drug use, weapon carrying) and aggressive and violent behavior (e.g., hitting, fighting, damaging property)—is a common risk factor for the development of externalizing behavior problems (Cotter, Wu, & Smokowski, 2015; Gifford-Smith, Dodge, Dishion, & McCord, 2005). With regard to bullying, studies have found a relation between peer deviance and bullying perpetration (e.g., Espelage et al., 2000; Ferguson et al., 2009). For example, Ferguson and colleagues (2009) found that depressed mood and associating with deviant peers were the strongest predictors of child self-reported bullying and rule-breaking behaviors.

As students begin to matriculate into secondary education, researchers have recognized the dynamic association between family and peer ecologies on the development of bullying behavior (Low & Espelage, 2014; Pepler et al., 2008). For instance, Pepler and colleagues (2008) examined different developmental trajectories—from early to late adolescence—of youth bullying and general aggression behavior in 800 children, while also assessing for individual (i.e., bullying, moral disengagement), family (i.e., parental trust, parental monitoring, and conflict with parents), and peer (i.e., physically aggressive peers, conflict with peers, and susceptibility to peer pressure) factors. Four bullying trajectories were found: high-bullying group, early-moderate bullying group, moderate bullying group, and non-bullying group. When comparing the groups and the associated risk and protective factors, they found that students in the high bullying group showed little regard for the welfare or feelings for others, often came in conflict with parents and peers, and associated with peers who bully, especially at the beginning of mid-adolescence. In addition, a longitudinal study conducted by Low and Espelage (2014) found significant interactions when examining community violence exposure and bullying and victimization, in relation to individual (e.g., impulsivity and delinquent behavior), peer (e.g., deviant peers), and familial (e.g., parental monitoring) influences. They found that parental monitoring buffered the effects of community violence exposure on bullying by reducing delinquent behavior and involvement with deviant peers. Overall, these studies demonstrate that peer and family influences not only predict but having moderating effects on bullying behavior.

Despite previous findings suggesting that children’s social context are important avenues by which to explore the development of bullying behavior, studies to date examining the joints effects of family violence and peer deviance have been limited to cross-sectional studies (Espelage et al., 2000; Ferguson et al., 2009). One limitation of cross-sectional studies is that statistical differences in the variables of interest are typically inferred across people (i.e., between-person effects), while disregarding how variables may vary within individuals’ over time (i.e., within-person effects; (Curran & Bauer, 2012). For instance, when applying psychological theories of behavior (e.g., social learning theory), empirical research investigating these patterns often use between-person data (e.g., cross-sectional assessments) or model between-person effects, even though implicitly or explicitly implying within-person processes (i.e., an error of inference; Curran & Bauer, 2012). However, an underlying presumption of social learning theory is that an individual’s behavior will change (i.e., within-person processes) because of what they learn and observe within their social context (Bandura, 1973). Studies applying this theory often ignore that these factors may fluctuate over time, and that these fluctuations themselves may be important determinants of change in an individual’s behavior (Curran & Bauer, 2012). For example, Merrin, Davis, Berry, D’Amico, and Dumas (2016) examined both the between- and within-person associations of crime, substance use, and social risk, using structural equation modeling (i.e., auto-regressive latent trajectory model with structured residuals). Despite past research showing that crime and substance use have a reciprocal relation, particularly when examining differences across individuals, the authors found that this relation was not evident when taking into consideration how individuals’ reports of crime and substance use fluctuated over time. As such, examining both within- and between-person effects would allow for a more substantive, developmentally meaningful, and statistically robust examination as well as have significant implications for theory (Curran & Bauer, 2012). Thus, the current study utilizes this framework as a tool to investigate the role of adult and peer influences on bullying behavior, while also employing a multilevel approach in hopes of expanding on the literature that applies social learning theory to bullying behavior.

The Current Study

Although findings from Low and Espelage (2014) and Pepler and colleagues (2008) demonstrate that there is a dynamic process between relational antecedents in both the peer and familial contexts that affect the development of bullying behavior over time, no longitudinal study, to date, has examined the interactive association of family violence and peer deviance. By employing a multilevel approach to examine both within- and between-person variables together, and how they may interact with each other (cross-level interaction), the current study examines the extent to which students’ fluctuations in exposure to family violence and peer deviance across time respectively and interactively affect their levels of bullying perpetration. In addition, there is the potential for moderating effects on bullying behavior to be found between family violence and peer deviance, especially because peer influences play a critical role during adolescence (Pellegrini & Long, 2002). For example, for students reporting higher levels of bullying in relation to heightened exposure to family violence over time, this association may also be moderated by reported fluctuations in exposure to peer deviance (and vice-versa with family violence as a moderator). Our findings will provide a more dynamic investigation of how reported levels of bullying may vary at different levels of family and peer deviance exposure, thus, expanding on the literature that applies social learning theory to bullying behavior.

Research Questions

The current study used a large sample of middle school students to examine the longitudinal associations (4 waves) between family violence and peer deviance on bullying perpetration using a multilevel design. We sought to answer several research questions, including (a) on average, are there increases in bullying perpetration across middle school?; (b) on average, are higher levels of family violence and peer deviance predictive of higher levels of bullying perpetration respectively (i.e., between-person effects)?; (c) are time-specific fluctuations from one’s typical level (i.e., within-person effects) of family violence and peer deviance associated with contemporaneous increases in bullying perpetration during middle school?; and (d) are there time-specific fluctuations—both at the within-and between-person levels as well as at the cross level—in family violence and peer deviance associated with contemporaneous increases in bullying perpetration?

MethodsParticipants and Procedures

Participating students in this study were asked to complete a survey seeking information about their experiences with family violence, peer deviance, and bullying perpetration. Participants included 1,194 5th, 6th, and 7th grade cohorts from four middle schools in a Midwest county. The sample was 49.6% female; age 11 to 15 years at baseline (M = 13.46); and 58.7% (n = 701) identified as African American, 30.9% (n = 369) identified as White, and 10.4% (n = 124) identified as Other. Over two years, data were collected in the Spring and Fall semester for a total of four time points. Sample sizes for the 5th (wave 1 = 337, wave 2 = 303, wave 3 = 294, wave 4 = 273), 6th (wave 1 = 411, wave 2 = 379, wave 3 = 358, wave 4 = 341), and 7th (wave 1 = 444, wave 2 = 403, wave 3 = 373, wave 4 = 0) grade cohorts differed slightly over time except for the 7th grade cohort at wave 4. The 7th grade cohort were not followed into high school and thus only contribute three waves of data.

Parental consent

The current study was formally announced in school newsletters, school district newsletters, and e-mails from the principals prior to the Spring of 2008. Upon receiving approval from the institutional review board (IRB) and district school board, a waiver of active consent was distributed to each parent/guardian of the students enrolled in the school. The passive consent included a letter containing information about the purpose of the study, and parents/guardians were also invited to attend information meetings held in each community. Parents/guardians who did not wish to have their child participate in the study were asked to sign the information letter and return it to the researchers.

After parents/guardians turned in these forms, it was determined that 95% of students initially participated in the study. Students were asked to consent to participate in the study through an assent procedure described on the coversheet of the survey distributed to all remaining students. Surveys were later de-identified with code numbers so researchers could track their responses over multiple time points and ensure confidentiality.

Survey administration

Students were initially informed about the nature of the study by one of the six trained research assistants, the principal investigator, or another faculty member who administered the survey. Surveys were conducted each semester (Spring and Fall) in classrooms ranging from 10 to 25 students. The survey took approximately 40–45 minutes to complete. Members of the research team ensured confidentiality by ensuring students were sitting far enough away from one another. The survey was administered and read aloud while students responded individually.

Because the content of the survey could be upsetting to students, researchers assured them that their participation in the study was entirely voluntary and that they could skip any question or stop participating in the survey at any time. At least one appropriately trained doctoral-level psychology student was on site to provide immediate support to any student and direct him or her to the appropriate resources. Students were also provided the contact information of the research team to seek more information about the study. Also, students were reminded about in-school resources available to them (e.g., guidance counselors) should they feel the need to talk to someone as a result of completing the survey.

MeasuresDemographics

Participants completed a demographic questionnaire that asked about their sex, age, grade, and race/ethnicity. Participants were given six options for their race/ethnicity: African American (not Hispanic), White (not Hispanic), Asian, Native American, Hispanic, and other (with an option to provide the most appropriate racial/ethnic descriptor).

Bullying perpetration

The Illinois Bully Scale (Espelage & Holt, 2001) is a nine-item scale assessing the frequency of bullying at school. Students were asked to recall how frequently they teased others, upset others for the fun of it, excluded others from their group of friends, helped harass others, and threatened to hit or hurt another student. Response options range from “never” to “7 or more times” in the past 30 days. Exploratory and confirmatory factor analysis has demonstrated the construct validity of this scale (Espelage & Holt, 2001). In the development sample, factor loadings for these items ranged from .52 to .75, and this factor accounted for 31% of the variance in the factor analysis (Espelage & Holt, 2001). This scale and the Youth Self-Report Aggression Scale correlated moderately (r = .65; Achenbach, 1991), supporting the notion that it was somewhat distinct from general aggression. Cronbach’s alphas ranged from .86 – .90 across assessment waves.

Peer deviance

The Friend’s Delinquent Behavior-Denver Youth Survey is a 7-item scale (Institute of Behavioral Science, 1987) that asks participants to report how many of their friends within the last year “Hit or threatened to hit someone,” “Purposely damaged or destroyed property that did not belong to them,” and “Used alcohol”, to name a few items. Response options were “None,” “Very Few,” “Some of them,” “Most of them,” and “All of them.” Cronbach’s alphas ranged from .85 – .88 across assessment waves.

Family violence

The Family Con ict and Hostility Scale (Thornberry, Krohn, Lizotte, Smith, & Tobin, 2003) was used to measure the level of perceived con ict and hostility in a student’s family environment. This scale contains three items from a larger survey designed for the Rochester Youth Development Study: “How often is there yelling, quarreling, or arguing in your household?”, “How often do family members lose their temper or blow up for no good reason?”, and “How often are there physical ghts in the household, like people hitting, shoving, or throwing things?” Response options ranged from “never” to “always” on a four-point scale. Cronbach’s alphas ranged from .79 – .81 across assessment waves.

Analytic Plan

To address our research questions, we fit a taxonomy of growth models to our data (Singer & Willett, 2003). We began by establishing a plausible functional form for individual’s levels of bullying perpetration over time. We tested a series of unconditional and conditional models that included a null model, random intercept, fixed linear growth, and a random linear growth model. Fixed linear growth was not found to be significant (b = −.006, SE = .008, p = .568) and random linear growth did not improve model fit (M1 to M2; ΔLR = 0.30, Δdf = 1, p = .596), suggesting that there was no significant average change in bullying over the four waves. As such, these growth parameters were removed from subsequent analysis. We also fit random intercept models for our main predictors (family violence and peer deviance) to assess whether there were meaningful variation at both the within- and between-person levels. All models were nested and tested based on significant reductions in −2*log-likelihood using deviance tests.

In proceeding models, we addressed our research questions by examining systematic groups of conditional models. In Model 2, we controlled for age, race, and sex. For Model 3, we tested the within- and between-person main effects of family violence. We then added the within- and between-person main effects of peer deviance on individual levels of bullying perpetration in Model 4. Model 5, the final model in Equation 1, we tested all possible interactions (within-person, between-person, and cross-level) between peer deviance and family violence on bullying perpetration. We excluded non-significant interactions for parsimony, leaving only two hypothesized interactions that examined peer deviance as a moderator of the relation between family violence and bullying in our final model. On levels of bullying perpetration, we estimated the within-person interactive effect of peer deviance and family violence. In addition, we allowed within-person peer deviance to vary as a function of between-person family violence. The stochastic part of the model allowed the intercept and within-person peer deviance to vary randomly across people.

Centering strategies are very important in multilevel models because they help separate the variance into different groups (e.g., within-person and between-person). Between-person, time-invariant predictors (level 2) were grand-mean-centered and again refer to average differences between people over time. Within-person, time-varying predictors (level 1) were person-mean-centered and again refer to time specific fluctuations from an individual’s ‘typical’ level (i.e., individual’s average). By using these centering techniques, we could partition variance in our variables at two respective levels of analysis making them orthogonal to one another (Tofighi & Enders, 2007). As such, individuals are treated as their own control, thereby adjusting for all observed and unobserved between-person (level 2) confounds.

To address non-normality, we used the Robust Maximum Likelihood (MLR) estimator available in Mplus 7.4 (Muthén & Muthén, 1998–2014). More specifically, we evaluated the extent to which our data were skewed and adjusted for the small amount of skewness we found across all variables by using MLR. To check these estimates, we also bootstrapped our standard errors (10,000), reran all models, and found the same results. Attrition in the current sample ranged from approximately 15–20% over time. However, it should be noted that individuals that were in 7th grade at baseline transitioned to high school at wave 3 and thus do not contribute any information at wave 4. To address missing data, we fit all our models using full information maximum likelihood (FIML). Unlike listwise deletion, FIML allows participants to contribute all available information they have available without the need to remove participants due to missing data, unless participants have missing values on all the predictor variables. As such, data from all 1,194 participants were included. We took specific steps to examine the missing data patterns. First, we tested the extent to which our data was missing completely at random (MCAR) using Little’s (1998) random MCAR test (Enders, 2010). The test for bullying perpetration was significant (X2 = 67.17, df = 28, p < .001), and indicated that the data were not MCAR. While there is no formal method for testing the missing at random (MAR) assumption without knowing the values of the missing dependent variable (i.e., bullying perpetration), we then examined the extent to which missing data were associated with sex and race. There were no differences in missing data patterns between males and females on bullying perpetration at wave 1 (X2 = 1.53, df = 1, p = .217), wave 2 (X2 = 0.42, df = 1, p = .515), or wave 4 (X2 = 0.12, df = 1, p = .912). However, females had a larger proportion of missing data on bullying perpetration at wave 3 (X2 = 5.47, df = 1, p = .019). For race, White participants had a larger proportion of missing data on bullying perpetration at wave 1 (X2 = 69.32, df = 2, p < .001), wave 2 (X2 = 54.54, df = 2, p < .001), wave 3 (X2 = 25.28, df = 2, p < .001), and wave 4 (X2 = 33.07, df = 2, p < .001) compared to Black and Other races. There were no differences between Black and Other race categories on bullying perpetration. By including race and sex in our models, we adjust for any bias due to missing data on these variables.

Equation 1:

Level 1: Bullyingji=β0i+β1i(FamilyViolenceij-FamilyViolence¯i)+β2i(PeerDevianceij-PeerDeviance¯i)+β3i(FamilyViolenceij∗PeerDevianceji)+εij

Level 2: β0i=γ00+γ01(Sex)i+γ02(Age)i+γ03(Black)i+γ04(Other)i+γ05(FamilyViolence¯)i+γ06(PeerDeviance¯)i+U0i β1i=γ10 β2i=γ20+γ21(FamilyViolence¯)i+U2i β3i=γ30

ResultsPreliminary models

Table 1 presents a taxonomy of our five nested models. The preliminary models showed support for several of our hypotheses. We fit a null model to our data and calculated an intra-class correlation of .513 that indicated that approximately 51% of the variability in individuals’ levels of bullying perpetration was due to between-person differences while 49% of variability was due to within-person differences across time.

Examining the demographic variables, Model 2 indicated that girls reported significantly lower levels of bullying perpetration across time (b = −.06, SE = .026, p = .027) compared to their boy counterparts. Compared to boys, there was a −.12 standard deviation decrease in bullying perpetration for girls. Further, Black students reported significantly higher average levels of bullying perpetration (b = .10, SE = .026, p < .001), such that Black students generally reported individual levels of bullying perpetration that were .22 of a standard deviation higher compared to their White counterparts. No differences were observed among individuals in the ‘Other’ group when compared to White students. Additionally, age had a positive association with bullying perpetration in preliminary models but was non-significant in subsequent models.

Between-person effects and bullying perpetration

Also, in Table 1, Models 3 and 4 display the main effects of the between-person associations of family violence and peer deviance. Significant between-person main effects were found, with individuals who reported higher levels of family violence (b = .12, SE = .013, p < .001) and peer deviance (b = .32, SE = .02, p < .001) also reported higher averages levels of bullying perpetration. That is, a one standard deviation increase in family violence was associated with a .26 standard deviation increase in bullying perpetration, and a one standard deviation increase in peer deviance was associated with a .50 standard deviation increase in bullying perpetration. Further, individuals who reported higher average levels of family violence and peer deviance showed higher levels of bullying perpetration in comparison to individuals with lower levels of family violence.

Within-person effects and bullying perpetration

In Table 1, Models 3 and 4 present the main effects of the (time-specific) within-person associations. The main effects model indicated that when individuals reported higher levels of family violence, they also reported higher levels of bullying perpetration at the same occasion (b = .05, SE = .016, p = .002). However, this effect was very small. Specifically, a one standard deviation increase in within-person family violence was associated with a .06 standard deviation increase in bullying perpetration. Further, when individuals reported higher peer deviance, they also reported higher levels of bullying perpetration at the same time point (b = .20, SE = .023, p < .001). A one standard deviation increase in within-person peer deviance was associated with a .17 standard deviation increase in bullying perpetration.

Within-and between-person interactive effects

We found two significant interactions: a within-person and cross-level interaction (see Model 5 in Table 1). At the within-person level, the positive relation among within-person family violence and bullying perpetration is especially pronounced at time points when individuals report high peer deviance (b = .12, SE = .04, p = .005). Figure 1 displays the plotted slopes of peer deviance at varying levels of family violence. This pattern is carried on as individuals report higher levels of family violence at any given occasion shown by the increasing slopes. It is noteworthy that at low levels of peer deviance (−1 standard deviation) there is no discernable difference among varying levels of family violence, indicated by the non-significant positive simple slope for low (−1SD) peer deviance. The positive simple slopes for the other two levels were significant (see Figure 1).

A significant cross-level interaction (b = .06, SE = .03, p = .05) showed that within-person peer deviance varied as a function of between-person family violence on increased levels of bullying perpetration. For instance, individuals with higher average levels of family violence had significantly higher levels of bullying perpetration, even at low levels of within-person peer deviance. Moreover, when individuals reported high levels of peer deviance, levels of bullying perpetration were exacerbated particularly for the high between-person family violence group (+1 standard deviation). Positive simple slopes were significant for each level (see Figure 2).

Discussion

Given the dearth of literature on the associated effects of family violence and peer deviance exposure on levels of bullying perpetration, the current study sought to examine the longitudinal associations of both factors on bullying perpetration using a large sample of middle school students. Consistent with previous studies and social learning theory (Espelage et al., 2000; Ferguson et al., 2009), our findings showed that violence exposure within individuals’ intermediate family and peer contexts was associated with higher levels of bullying behavior. Though we cannot conclude that the bullying perpetrators learned and accepted observed violent and aggressive acts as appropriate methods for dealing with conflict or reaching a desired goal, these findings still affirm the notion that there is a relationship between exposure to violence and one’s own perpetration of it.

It is important to underscore that bullying perpetration levels did not increase over the two years of this longitudinal study. This finding was unexpected in that we believed levels of bullying would increase significantly as students transitioned into and moved through middle school (5th and 6th grade), especially because levels of bullying tend to be more pronounced during that time (Espelage et al., 2015; Pellegrini & Long, 2002; Pellegrini & Bartini, 2000). For instance, Pellegrini and Long (2002) found that students’ levels of bullying perpetration increased significantly from 5th to 6th grade, and then began to decrease at the beginning of the 7th grade. However, we in general only found slight changes in bullying levels across waves, which were not significantly different from 0. Also, we did not assess students in elementary school, so we were, unfortunately, unable to assess the elementary-middle school transition.

Furthermore, we found dynamic relationships between exposure to family and peer deviance in how they affect bullying perpetration. Specifically, students with relatively elevated levels of both family violence and peer deviance at a given time reported particularly high levels of bullying perpetration at that time (see Figure 1). In addition, increasing levels of peer deviance at a given time corresponded with higher levels of bullying perpetration at that time, especially for students with generally high levels of family violence across time (see Figure 1).

A similar dynamic relation also existed across levels (cross-level interaction). We found a moderating effect where increasing levels of peer deviance at a given time exacerbated the relation between family violence exposure and levels of bullying perpetration. One plausible explanation for the moderating role of peer deviance on bullying behavior is that research on problem behaviors among adolescents indicate that youth who live in households characterized with high levels of family conflict and low positive family relations are less likely to receive adequate parental monitoring and more likely to associate with deviant peers over time (Ary, Duncan, Duncan, & Hops, 1999). However, more research is needed to test this relation with bullying behavior.

The current study has multiple strengths worth noting. First, previous studies examining associations among peer deviance, family violence, and bully perpetration have largely been cross-sectional. Furthermore, longitudinal studies examining these associations have been limited in their methodological approach, whereas this study appears to be the first examining the longitudinal association among peer deviance, family violence, and bully perpetration using a time-variant, multilevel approach. Second, the current study examined both within- and between-person level effects, allowing researchers to examine average differences between people, and how individuals change respective to their own mean levels of bullying perpetration over time. More specifically, partitioning the variance at multiple levels of analysis allowed us to control for person-level dependencies (i.e., how individual’s scores are correlated over time) and all observed and unobserved between-person confounds (i.e., how individuals differ from one another), which made for a more robust model, thereby adding to the substantive utility of our findings. By factoring in both within- and between–level differences, this added interpretational value in terms of how we interpreted the relation between family violence and peer deviance on bullying perpetration, with non-significant average growth in levels of bullying perpetration and within-person effects showing significant fluctuations in reported levels.

Limitations

There are several limitations to this study that should be noted. First, the current study used data collected from self-reported measures of bullying. While self-report is the most commonly used method in bullying research (Cornell & Brockenbrough, 2004), multi-informant methods of examining incidence of bullying such as peer- and teacher-reports should also be considered for future studies to reduce any threats to validity. Second, the current study examined middle school students from one Midwest county. Thus, the generalizability of these findings is geographically limited. Third, we were only able to examine bullying behaviors across a two-year span (four waves), looking at students who have already transitioned into or passed the 7th grade by the second wave. This may have limited our ability to detect significant average growth because our analysis did not include critical time periods where bullying behavior becomes pervasive, particularly around the 5th grade. Fourth, the MCAR tests was significant and indicated that the data were not missing completely at random. However, we examined MAR assumptions and included sex and race variables in our model to account for any potential bias due to missingness on these variables. Sixth, the 7th grade cohort was not followed into high school and thus do not contribute any information at wave 4. Lastly, although social learning theory provides a plausible explanation for the development of bullying behavior, it is limited in that it does not take into consideration the broader context. Bullying is indeed influenced by various other ecological contexts (e.g., school climate, neighborhood environment, cultural norms; Hong, & Espelage, 2012) that this study did not include. Moreover, the question of why is it that youth who live in households characterized by high levels of family conflict associate more with deviant peers over time is interesting. However, the current study’s theoretical framework and methodological approach do not provide data to answer this question. We can only hypothesize based on previous literature rooted in relational theory (in that an examination of the nature and quality of relationships on thoughts, behaviors, and attitudes is conducted) rather than a social-cognitive framework that focuses on the relation between cognitions (i.e., observation) and one’s social environment on behavior (i.e., the imitation of observed interactions in one’s social environment). Causal models linking pathways between characteristics of the peer and family relational context on bullying behavior is needed.

Research Implications

In sum, these findings support previous research that points to the strong predictors of family violence and exposure to deviant peers on increased levels of bullying perpetration (Espelage et al., 2000; Ferguson et al., 2009). In line with social learning theory, bullying behavior did change in relation to reported levels of violence exposure within individuals’ immediate family and peer contexts. Expanding on social learning theory’s explanation for bullying behavior, we also found a dynamic process between relational antecedents in the peer and family context when factoring in within-person differences. Specifically, we indeed found that fluctuations in violence exposure within individuals’ family and peer contexts were important determinants of change in an individual’s bullying behavior. These findings provide a more nuanced lens from which to view the co-occurring relations between family and peer ecologies. Future research should incorporate statistical modeling that considers both between-person and within-person approaches in order to capture the complex interactions between family and peers on bullying perpetration. Also, adding protective factors in these models is critical to identify how youth exposed to family violence resist joining deviant peer groups.

Clinical and Policy Implications

With regard to prevention and intervention efforts, delinquency and peer relations should be targeted to reduce the effect of family violence on bullying behavior. Interestingly, a recent randomized clinical trial of a middle school social-emotional learning program (Second Step; Committee for Children, 2008) reduced self-reported delinquency across the three-year evaluation, which in turn was associated with reductions in bullying perpetration (Espelage, Van Ryzin, Low, & Polanin, 2015). While few anti-bullying programs target both the peer and family-level, programs that have focused on skill building at the peer-level have had success in reducing bullying perpetration. In the context of the current findings, identifying the skills and programs that build prosocial connections between peers seems is especially important. Two factors in specific – social support and connectedness – have been suggested to be protective factors against bully perpetration (McNeely & Falci, 2004; Resnick, Ireland, & Borowsky, 2004). According to SAMHSA’s National Registry of Evidence-Based Programs and Practices, the Cross-Age Mentoring Program (CAMP) for Children with Adolescent Mentors (Karcher, 2000) is one program that includes these factors and offers a potential intervention for middle school students. CAMP’s purpose is to promote social-emotional and cognitive development in students by fostering connectedness among their peers, school, family, and community and prevent problematic behaviors. Younger (4th to 8th grade) mentees are paired with older (9th to 11th grade) mentors who engage in structured meetings throughout the school year and for ten days over the summer. Though it has not been evaluated as an anti-bullying program, CAMP offers a potential intervention deserving of future research given its association with improved connectedness with parents (Karcher, Davis, & Powell, 2002) and friends, culturally diverse peers, and their schools (Karcher, 2009). Future research may consider evaluating programs focused on connectedness and social support to improve peer relations.

AchenbachTM1991Manual for the Youth Self-Report and 1991 ProfileBurlington, VTUniversity of Vermont Department of PsychiatryAkersRL2011Social learning and social structure: A general theory of crime and devianceTransaction PublishersAlbdourMKrouseHJ2014Bullying and victimization among african american adolescents: A literature reviewJournal of Child and Adolescent Psychiatric Nursing272688224456268AryDVDuncanTEDuncanSCHopsH1999Adolescent problem behavior: The influence of parents and peersBehaviour Research and Therapy37321723010087640BauerNSHerrenkohlTILozanoPRivaraFPHillKGHawkinsJD2006Childhood bullying involvement and exposure to intimate partner violencePediatrics1182e235e24216882768BanduraA1973Aggression: A social learning analysisPrentice-HallBanduraAWaltersRH1977Social learning theoryEnglewood Cliffs, NJPrentice- HallBureau of Just1ice Statistics2014Violence by gang membersU.S. Department of JusticeCarlyleKESteinmanKJ2007Demographic differences in the prevalence, co- occurrence, and correlates of adolescent bullying at schoolJournal of School Health77962362917970866CopelandWEWolkeDAngoldACostelloEJ2013Adult psychiatric outcomes of bullying and being bullied by peers in childhood and adolescenceJAMA Psychiatry70441942623426798CornellDGBrockenbroughK2004Identification of bullies and victims: A comparison of methodsJournal of School Violence36387CotterKLWuQSmokowskiPR2016Longitudinal risk and protective factors associated with internalizing and externalizing symptoms among male and female adolescentsChild Psychiatry & Human Development47347248526341092CurranPJBauerDJ2011The disaggregation of within-person and between-person effects in longitudinal models of changeAnnual Review of Psychology62583619EndersCK2010Applied missing data analysisGuilford PressEspelageDLBosworthKSimonTR2000Examining the social context of bullying behaviors in early adolescenceJournal of Counseling and Development783326333EspelageDLHoltML2001Bullying and victimization during early adolescence: Peer influences and psychosocial correlatesJournal of Emotional Abuse23123142EspelageDLHongJSRaoM*ThornbergR2015Social-ecological factors associated with bullying perpetration among early adolescents across the elementary- middle school transitionViolence & Victims3047048810.1891/0886-6708.VV-D-14-0004626118267EspelageDLLowSRaoMAHongJSLittleTD2014Family violence, bullying, fighting, and substance use among adolescents: A longitudinal mediational modelJournal of Research on Adolescence242337349FergusonCJSan MiguelCHartleyRD2009A multivariate analysis of youth violence and aggression: The influence of family, peers, depression, and media violenceThe Journal of Pediatrics155690490819683724Gifford-SmithMDodgeKADishionTJMcCordJ2005Peer influence in children and adolescents: Crossing the bridge from developmental to intervention scienceJournal of Abnormal Child Psychology33325526515957555GladdenRMVivolo-KantorAMHamburgerMELumpkinCD2014Bullying surveillance among youths: Uniform definitions for public health and recommended data elements, version 1.0Retrieved from https://www.cdc.gov/violenceprevention/pdf/bullying-definitions-final-a.pdfHongJSEspelageDL2012A review of research on bullying and peer victimization in school: An ecological system analysisAggression and Violent Behavior174311322HymelSSwearerSM2015Four decades of research on school bullying: An introductionAmerican Psychologist70429325961310Institute of Behavioral Science1987Youth Interview Schedule: Denver Youth SurveyBoulder, CoUniversity of ColoradoKarcherMJ2000Children with Adolescent Mentors Program (CAMP): The developmental mentoring trainer’s guideUnpublished manualUniversity of Wisconsin-MadisonKarcherMJDavisCPowellB2002The effects of developmental mentoring on connectedness and academic achievementSchool Community Journal1223550KarcherMJ2009Increases in academic connectedness and self-esteem among high school students who serve as cross-age peer mentorsProfessional School Counseling124292299Knous-WestfallHMEhrensaftMKMacDonellKWCohenP2012Parental intimate partner violence, parenting practices, and adolescent peer bullying: A prospective studyJournal of Child and Family Studies215754766LittleRJ1988A test of missing completely at random for multivariate data with missing valuesJournal of the American Statistical Association8340411981202LowSEspelageDL2014Conduits from community violence exposure to peer aggression and victimization: Contributions of parental monitoring, impulsivity, and deviancyJournal of Counseling Psychology61222124635595McNeelyCFalciC2004School connectedness and the transition into and out of health- risk behavior among adolescents: A comparison of social belonging and teacher supportJournal of School Health74728429215493705MerrinGJDavisJPBerryDD’AmicoEJDumasTM2016The longitudinal associations between substance use, crime, and social risk among emerging adults: A longitudinal within and between-person latent variables analysisDrug and Alcohol Dependence165717827242288Musu-GilletteLZhangAWangKZhangJKempJDilibertiMOudekerkBA2018Indicators of School Crime and Safety: 2017 (NCES 2018-036/NCJ 251413)National Center for Education Statistics, U.S. Department of Education, and Bureau of Justice Statistics, Office of Justice Programs, U.S. Department of JusticeWashington, DCMuthénLKMuthénBO1998–2014Mplus User’s Guide7Los Angeles, CAMuthén & MuthénOlweusD1993Bullying at schoolOxford, EnglandBlackwellOlweusD1997Bully/victim problems in school: Facts and interventionEuropean Journal of Psychology of Education7495510PellegriniADBartiniM2000A longitudinal study of bullying, victimization, and peer affiliation during the transition from primary school to middle schoolAmerican Educational Research Journal373699725PellegriniADLongJD2002A longitudinal study of bullying, dominance, and victimization during the transition from primary school through secondary schoolBritish Journal of Developmental Psychology202259280PeplerDJiangDCraigWConnollyJ2008Developmental trajectories of bullying and associated factorsChild Development79232533818366426ResnickMDIrelandMBorowskyI2004Youth violence perpetration: What protects?what predicts? findings from the national longitudinal study of adolescent healthJournal of Adolescent Health355424433SalmivalliC2010Bullying and the peer group: A reviewAggression and Violent Behavior152112120SingerJDWillettJB2003Applied longitudinal data analysis: Modeling change and event occurrenceNew YorkOxford University PressThornberryTPKrohnMDLizotteAJSmithCATobinK2003Gangs and delinquency in developmental perspectiveNew YorkCambridge University PressEndersCKTofighiD2007Centering predictor variables in cross-sectional multilevel models: a new look at an old issuePsychological Methods12212113817563168United States Department of Justice2015Domestic violenceRetrieved July 19, 2016, from http://www.justice.gov/ovw/domestic-violenceVoisinDRHongJS2012A meditational model linking witnessing intimate partner violence and bullying behaviors and victimization among youthEducational Psychology Review244479498

Within-Person Family Violence and Within-Person Peer Deviance (Level 1 Interaction)

Simple Slopes: +1SD WP Peer Deviance: β = 0.09, SE = 0.02, p < .001; Mean Peer Deviance: β = 0.05, SE = 0.02, p < .001; −1SD WP Peer Deviance: β = .01, SE = 0.02, p = .578.

Within-Person Peer Deviance and Between-Person Family Violence (Cross Level Interaction)

Simple Slopes: +1SD BP Peer Deviance: β = 0.14, SE = 0.02, p < .001; Mean Peer Deviance: β = 0.12, SE = 0.01, p < .001; −1SD BP Peer Deviance: β = .10, SE = 0.02, p < .001.

Estimates of Fixed and Random Effects from a Series of Individual Multilevel Models Predicting Bullying Perpetration

Parameter Estimates (SE)
Model 1Model 2Model 3Model 4Model 5
Fixed Effects
Intercept.435***(.014).338***(.026).369***(.025).413***(.023).408***(.023)
Age.038*(.013).019(.013)−.016(.012)−.016(.012)
Sex (Ref = Male)−.034(.026)−.085***(.025)−.050*(.023)−.049*(.023)
Black (Ref=White).206***(.09).197***(.027).099***(.025).100***(.026)
Other (Ref=White)−.011(.047).011(.044)−.035(.041)−.032(.041)
WP Family Violence.075***(.016).050**(.016).050***(.016)
BP Family Violence.201***(.013).117***(.013).116***(.013)
WP Peer Deviance.204***(.023).212***(.032)
BP Peer Deviance.319***(.020).315***(.020)
WP Family Violence* WP Peer Deviance.120***(.006)
WP Peer Deviance* BP Family Violence.063*(.032)
Random Effects
BP Intercept.183***(.011).169***(.010).138***(.009).106***(.008).121***(.008)
WP Intercept.165***(.006).165***(.006).159***(.006).152***(.006).126***(.006)
WP Peer Deviance.142***(.036)
Fit Indices
−2LL4856.764757.854403.304075.774007.80
AIC4862.764771.854421.314097.774035.80
BIC4880.824813.994475.284163.714119.72
#Parameters3791114

Note: Model 1 is an unconditional null model. Model 2 added the control variables of age, gender, and race (M1 to M2; ΔLR = 98.9, Δdf = 4, p < .001). Model 3 added the main effects of within and between-person family violence (M2 to M3; ΔLR = 354.55, Δdf = 2, p < .001). Model 5 added the main effect of within and between-person time peer deviance (M3 to M4; ΔLR = 327.54., Δdf = 2, p < .001). Model 6 added the interactions of within-person family violence and within-person peer deviance, and within-person peer deviance and between-person family violence (M4 to M5; ΔLR = 67.97, Δdf = 3, p <.001).

p < .05,

p < 01,

p < .001.