The larger search went through five phases, reducing the number of records from 47,241 to 140, which was summarized in a PRISMA diagram for systematic review (see Fig. 2; Page et al., 2021). Studies were excluded if they did not meet all inclusion criteria as described above. The number of studies excluded at each phase was as follows: de-duplicating (n = 13,705), title screening (n = 28,731), abstract screening (n = 4,069), full-text screening (n = 520), and data extraction (n = 76). Among the final 140 records with data extracted, 51 on bullying were reported in this study, and 92 on general peer relationships were described in J. H. Wang et al. (2023).

PRISMA flowchart of the search

A summary of the 51 included studies is provided in Table 1 (see Supplemental Table S1 for more details). There was an increasing trend in the number of publications since the first study was published in 2005 (see Fig. 3). The studies were conducted in 22 countries or regions across continents. The US had the most studies (n = 14), followed by Brazil (n = 5), Canada (n = 3), South Korea (n = 3), and Sweden (n = 3). The following countries or regions had two studies each: China, Finland, Norway, South Africa, Taiwan, and the UK. Further, the following countries contributed one study each: Cambodia, France, Ireland, Israel, Japan, Malawi, Mexico, Spain, The Netherlands, and Vietnam. Lastly, one study had participants from both the US and Puerto Rico.

Number of included articles per year

The quality assessment total score ranged from 0.59 to 1.00 across studies, with a mean of 0.81. Readers are recommended to interpret the total score in the context of a study’s methodology (e.g., design and analyses) and the quality scores at the item level (see Table 2). Some studies had a high total quality score due to their high quality, whereas others received a high score because their methodology was simple and multiple items were not applicable. The scores for the 11 items ranged from 0.82 (outcome measure) to 1.98 (study objective).

The results are organized in two sections: (1) the associations between ACEs and bullying (i.e., the main effect) and (2) the mediating and moderating roles of bullying. The main effect section is further organized by the three bullying behaviors (perpetration, victimization, and bully-victims) and then by ACEs (first combined ACEs and then singular ACEs) when appropriate. Results are described in a separate subsection when there are at least two studies on the association between an ACE and bullying behavior. ACEs with only one study are presented at the end of one subsection. Additionally, the main effect section reported links between the independent variables (i.e., ACEs) and mediators (i.e., bullying) of the mediation models. The term “special sample” refers to samples recruited from non-school or community settings or a unique cultural context.

Across the 11 singular and six combined ACEs (see Table 3), physical abuse was most studied in relation to bullying (n = 16), followed by domestic violence (n = 13), sexual abuse (n = 11), emotional abuse (n = 8), divorce and separation (n = 8), abuse (n = 6), and family violence (n = 6). The remaining ACEs had three or fewer studies, and household dysfunction had no studies.

There were 34 studies, primarily cross-sectional (cross-sectional, n = 27; longitudinal, n = 7), that examined the associations between ACEs and bullying perpetration. Some studies investigated multiple ACEs in relation to bullying perpetration. For instance, Williamson et al. (2018) examined the relations of four ACEs (i.e., physical abuse, sexual abuse, emotional abuse, and domestic violence) with bullying perpetration. This led to a total of 59 associations analyzed (i.e., 41 positive, 17 nonsignificant, and one negative) across all studies. Across all ACEs (see Table 3), cumulative ACE score (or composite ACEs), maltreatment, and family violence were positively associated with bullying perpetration across all studies. Abuse, physical abuse, and domestic violence appeared to be positively related to bullying perpetration, as most studies found positive associations, and the few studies that reported nonsignificant associations could be explained by methodological features. The associations between the remaining ACEs and bullying perpetration were not studied at all or appeared to be inconclusive either because the total number of included studies was too small or the results were split between positive and nonsignificant relations.

Overall, sample sizes appeared adequate, considering the complexities of most models. Samples were mainly recruited from schools or communities, with a few studies recruited from special groups, including a psychiatric youth sample (Mustanoja et al., 2011), youth involved with child welfare (Sterzing et al., 2020), and youth offenders (Viljoen et al., 2005). The predominant analytic strategy was logistic regression, although other analyses were used, such as the Chi-square test, the Kruskal–Wallis H test, linear regression, multilevel modeling, manifest path analysis, and structural equation modeling (SEM) with latent variables. The primary method of assessing bullying perpetration was self-report, except for three studies that included either parent report (Morcillo et al., 2015; Odar Stough et al., 2016) or peer nomination (Wei & Lee, 2014). Further, many studies utilized single-item measures (e.g., Duke et al., 2010) or lacked psychometric information (e.g., CDC, 2011), although multi-item measures were used in other studies (e.g., Espelage et al., 2014). Lastly, only seven of the 34 included studies were longitudinal, with three examining family violence (Espelage et al., 2014; Grant et al., 2019; Low & Espelage, 2013), two on domestic violence (Knous-Westfall et al., 2012; Le et al., 2017), one on abuse (Hong et al., 2017), and one on divorce and separation (Wei & Lee, 2014). Thus, although a pattern of positive associations was reported for multiple ACEs (e.g., cumulative ACE score), it is largely unknown the extent to which ACEs are associated with bullying perpetration over time.

Composite ACEs. All three cross-sectional studies reported positive associations between cumulative ACE scores and bullying perpetration. Specifically, Duke et al. (2010) reported that the cumulative ACE score was positively related to bullying perpetration. Further, the likelihood of perpetration increased two fold for girls and 2.7 fold for boys when youth had four or more ACEs, compared to peers with no ACEs. Further, Reisen et al. (2019) found that bullying perpetrators were more likely to report total adversity (i.e., OR = 5.8 for 4–10 ACEs, OR = 2.3 for 1–3 ACEs). Finally, Forster et al. (2020) reported that an increase from 0 to 6 ACEs was associated with higher physical bullying perpetration probability, although an increase in ACEs was not significantly related to relational bullying perpetration.

Maltreatment. Positive relations between maltreatment and bullying perpetration were found across all three cross-sectional studies. The definition of maltreatment (i.e., a combination of abuse and neglect) differed slightly across studies as G.-F. Wang et al. (2019) included sexual abuse, and the other two studies did not (Holt et al., 2009; X. Wang et al., 2017). Specifically, G.-F. Wang et al. (2019) reported that maltreatment was associated with higher odds of being in the bully group compared to the non-involved group for each form of school bullying (i.e., physical, verbal, relational, and cyber). Holt et al. (2009) found via a Chi-square test that bullies were more likely to report maltreatment. Lastly, using SEM, X. Wang et al. (2017) reported that maltreatment was positively related to concurrent bullying perpetration.

Family Violence. Positive associations between family violence and bullying perpetration were reported in all five studies (three longitudinal; two cross-sectional) that utilized various analyses (i.e., multilevel modeling, SEM, regression, logistic regression). Two cross-sectional studies conducted logistic regression and reported that family violence was associated with increased odds of bullying perpetration (Fitzpatrick et al., 2007; Fujikawa et al., 2016). Of note, Fitzpatrick et al. (2007) analyzed a predominantly African American sample and Fujikawa et al. (2016) examined a Japanese sample.

In a longitudinal study (3 waves across 3 semesters) with baseline bullying perpetration controlled, Low and Espelage (2013) found that Wave 1 family violence positively predicted Wave 3 non-physical bullying perpetration. However, this association became nonsignificant when additional control variables were added to the model (i.e., depression, substance use, hostility). Wave 1 family violence was not significantly related to Wave 3 cyberbullying perpetration (Low & Espelage, 2013). In another longitudinal study (3 waves across 3 semesters) that examined the mediating role of bullying perpetration on the relation of family violence with youth substance use, Espelage et al. (2014) conducted SEM and reported that family violence was positively associated with later bullying perpetration. In the last longitudinal study (4 waves across 2 years) with multilevel modeling, Grant et al. (2019) found that between-person family violence predicted increases in bullying perpetration over time, and within-person family violence was positively associated with concurrent bullying perpetration. Of note, these three studies were conducted by the same research group. Further, Low and Espelage (2013) and Espelage et al. (2014) analyzed the same dataset.

Abuse. Abuse had positive relations with bullying perpetration in two studies but nonsignificant associations in one study that used a limited measure for perpetration. Specifically, Shamu et al. (2016) found via SEM that abuse (i.e., physical, emotional, and sexual) was positively related to concurrent bullying perpetration. Hong et al. (2017) reported in a longitudinal study through path analysis that abuse (i.e., physical and emotional) was positively associated with later bullying perpetration. Lastly, Lepistö et al. (2012) suggested that family abuse (i.e., physical and emotional) was not related to concurrent bullying perpetration utilizing multivariate analyses. Of note, a single-item dichotomized measure was used to assess bullying perpetration in Lepistö et al. (2012), whereas Hong et al. (2017) and Shamu et al. (2016) both utilized multi-item measures.

Physical Abuse. Physical abuse was positively related to concurrent bullying perpetration across six studies that analyzed school or community samples but was not significantly associated with concurrent bullying perpetration in the three studies that had special samples. Specifically, physical abuse was associated with increased odds of bullying perpetration in the three studies utilizing logistic regression (CDC, 2011; Duke et al., 2010; Lucas et al., 2016). Physical abuse was also positively related to bullying perpetration in studies utilizing SEM (Min et al., 2015), path analysis (Williamson et al., 2018), and regression (I. R. de Oliveira et al., 2018). These six studies included various control variables in their models, including witnessing domestic violence (CDC, 2011), neglect and emotional abuse (Min et al., 2015), and demographic variables (Duke et al., 2010). All six studies analyzed school or community samples.

The remaining three studies conducted logistic regression in special samples and showed nonsignificant relations between physical abuse and bullying perpetration (Ameli et al., 2017; Mustanoja et al., 2011; Sterzing et al., 2020). Of note, Ameli et al. (2017) was the only one of the nine studies that recruited a sample from an African country (i.e., Malawi). Possible cultural and linguistic differences in the measurement of physical abuse and bullying perpetration might explain the nonsignificant finding. Mustanoja et al. (2011) analyzed a psychiatric sample, and Sterzing et al. (2020) examined a female-only sample with half of the participants involved with child welfare. Such sample characteristics differ from the school or community samples in the six studies above that reported positive associations and may help explain the observed differences.

Sexual Abuse. Eight cross-sectional studies examined the associations between sexual abuse and bullying perpetration, four of which reported positive relations, and four reported nonsignificant findings. The four studies reporting positive associations utilized Chi-square tests or logistic regression and did not control ACEs other than sexual abuse in analyses, which might have impacted the results. The diverse sample characteristics (e.g., school, community, youth offenders, child welfare, and psychiatric) and measurement strategies across the eight studies might have also contributed to the heterogeneity in the findings. Specifically, using a Chi-square test, Viljoen et al. (2005) found in a sample of youth offenders that bullies were more likely to report sexual abuse than non-involved youth. Holmberg and Hellberg (2010; Chi-square test) reported in an all-female school sample that girls who had been sexually abused were more likely to report bullying perpetration than their non-abused counterparts. Duke et al. (2010) found in a school sample that sexual abuse was related to increased odds of bullying perpetration. Further, Mansbach-Kleinfeld et al. (2015) reported in a community sample that sexual abuse was related to higher odds of bullying perpetration.

The remaining four studies indicated nonsignificant associations between sexual abuse and bullying perpetration. Specifically, Mustanoja et al. (2011) found that sexual abuse was not significantly related to the odds of bullying perpetration in a psychiatric youth sample with physical abuse and witnessing interparental violence (i.e., another two ACEs) controlled. Williamson et al. (2018) indicated that sexual abuse was not associated with bullying perpetration through path analysis with multiple ACEs (i.e., domestic violence exposure and emotional and physical abuse) controlled. Sterzing et al. (2020) discovered that youth with sexual abuse were not more likely to be bullies compared to non-involved youth in an all female sample, with approximately half involved with the child welfare system. Lastly, Mignot et al. (2018) found that sexual abuse was not related to bullying perpetration. Of note, single-item measures were used to measure sexual abuse and bullying perpetration by Mignot et al. (2018).

Emotional Abuse. The associations between emotional abuse and bullying perpetration appeared mixed across the six cross-sectional studies (positive, n = 3; nonsignificant, n = 3). Of note, the studies showing positive associations utilized univariate regression strategies (i.e., linear regression, logistic regression), whereas studies indicating nonsignificant findings used multivariate analyses (i.e., SEM, path analysis) or a special sample. Specifically, Ameli et al. (2017) reported that emotional abuse was associated with higher odds of bullying perpetration among girls but not boys, with other ACEs (i.e., domestic violence, physical abuse) controlled. Lucas et al. (2016) found that emotional abuse was related to higher odds of bullying perpetration. Further, I. R. de Oliveira et al. (2018) reported that emotional abuse was positively associated with being a bully via linear regression with demographic variables but no other ACEs controlled.

The remaining three studies suggested nonsignificant associations. Specifically, Williamson et al. (2018) reported that emotional abuse was not associated with bullying perpetration in a path analysis with other ACEs controlled (i.e., domestic violence exposure, physical and sexual abuse). Min et al. (2015) found that emotional abuse was not related to bullying perpetration via SEM with neglect and physical abuse controlled. Lastly, Sterzing et al. (2020) reported that youth with emotional abuse were not more likely to be bullies via logistic regression in an all-female sample, with about half involved with the child welfare system.

Domestic Violence. Findings were fairly consistent that domestic violence was positively associated with bullying perpetration, as positive associations were reported in nine of the 12 studies with different control variables included (e.g., no control variables, demographic variables only, and demographic and other ACE variables). However, it is unclear how witnessing domestic violence affects youth bullying perpetration across adolescence, as the predominant design was cross-sectional (n = 10). Further, most studies utilized simple analyses like logistic regression (n = 8) or Chi-square tests (n = 2).

Specifically, three cross-sectional studies conducted logistic regression with various ACEs controlled in the models (i.e., physical abuse, CDC 2011; parental mental health, Odar Stough et al., 2016; physical and emotional abuse, Lucas et al., 2016), all of which reported a positive association between witnessing domestic violence and the concurrent odds of bullying perpetration. Two cross-sectional studies conducted logistic regression without controlling any other ACEs but also reported a positive relation between witnessing domestic violence and the odds of bullying perpetration (W. A. de Oliveira et al., 2016; Duke et al., 2010). Two additional cross-sectional studies found that bullies were more likely to report domestic violence exposure via Chi-square tests (Holt et al., 2009; Viljoen et al., 2005).

Further, two longitudinal studies showed positive associations between domestic violence exposure and bullying perpetration. Le et al. (2017) explored bullying stability through a short-term longitudinal design (i.e., 2 waves across 6 months) and logistic regression in a Vietnamese school sample. They found that witnessing interparental violence (IPV) was associated with decreased odds in the stable low group of bullying perpetration and increased odds in the declining group but not related to the increasing group. The increased odds in the declining group may have been due to the high baseline level of bullying perpetration in participants who had witnessed IPV. In another longitudinal study that spanned 6–7 years, Knous-Westfall et al. (2012) reported through regression that witnessing severe IPV where an injury was present predicted higher rates of relational bullying perpetration in boys and girls, and higher rates of overt bullying in boys only, although witnessing any type of IPV was not significantly associated with later bullying perpetration (i.e., overt or relational). Of note, Knous-Westfall et al. (2012) did not control for baseline levels of bullying perpetration.

The remaining three studies showed nonsignificant associations between witnessing domestic violence and concurrent bullying perpetration. Specifically, Williamson et al. (2018) reported that domestic violence exposure was not significantly related to bullying perpetration in a path analysis with three other ACEs controlled (i.e., physical, sexual, and emotional abuse). Similarly, Ameli et al. (2017) reported in a Malawi sample that domestic violence exposure was not associated with the odds of bullying perpetration with other ACEs (i.e., physical and emotional abuse) controlled. Lastly, Mustanoja et al. (2011) reported in a psychiatric sample that witnessing IPV was not related to the odds of bullying perpetration, with physical and sexual abuse controlled.

Caregiver Substance Misuse. Mixed results (i.e., positive and nonsignificant) were suggested regarding the associations between caregiver substance use and bullying perpetration in two cross-sectional studies with logistic regression. Specifically, Duke et al. (2010) reported in a US school sample that family alcohol and drug use was associated with increased odds of bullying perpetration, with demographics controlled. Morcillo et al. (2015) found in a sample of Puerto Rican children living in the US or Puerto Rico that parental substance use was related to higher odds of bullying perpetration with demographic variables controlled. However, this association became nonsignificant once child characteristic variables (i.e., early aggression, academic achievement) were added to the model. Thus, the positive association reported by Duke et al. (2010) may have been an artifact of only including demographic controls.

Household Mental Illness. Two studies reported mixed results (i.e., negative and positive) regarding the associations between household mental illness and bullying perpetration. Both studies conducted cross-sectional logistic regression and used parent-reported single-item measures of bullying perpetration. However, they used different definitions and measurements of parental mental health, which may explain the discrepant findings. Specifically, in a national sample of normal weight, overweight, and obese youth, Odar Stough et al. (2016) found that youth whose mothers had poor mental health were less likely to be a bully, with multiple variables controlled (e.g., domestic violence exposure). However, paternal mental health was not associated with bullying perpetration. In a sample of Puerto Rican children living in the US or Puerto Rico, Morcillo et al. (2015) reported that maternal depression was related to higher odds of bullying perpetration, with demographic variables controlled. However, this association became nonsignificant once child characteristic variables (e.g., early aggression) were added to the model (Morcillo et al., 2015). In the same study, authors found that parental antisocial behavior was related to higher odds of bullying perpetration. This relation remained significant when other control variables (e.g., child characteristics and maternal depression) were added, except in the final model, where cultural factors were included (Morcillo et al., 2015).

Divorce and Separation. Mixed findings (positive, n = 2; nonsignificant, n = 1) were reported across the three studies examining parental divorce and separation on bullying perpetration. All studies used family structure in analyses, which indicated possible divorce and separation. However, they were conducted in three different cultures, which might explain the discrepant findings. Specifically, in a Taiwanese sample with longitudinal multilevel models (5 waves across 1 year), Wei and Lee (2014) found that living with two married biological parents was negatively associated with later physical bully centrality (i.e., less likely to be nominated as a bully). In a sample of Puerto Rican children living in the US or Puerto Rico with Chi-square tests, Morcillo et al. (2015) found that a higher percentage of children from single-parent families reported bullying perpetration than two-parent families. However, in a predominantly African American sample, Fitzpatrick et al. (2007) reported that alternative family structures (i.e., no biological parents; one biological parent; one biological parent/stepparent) were not significantly associated with bullying perpetration, compared to intact families (i.e., living with two biological parents) via logistic regression with multiple variables controlled (e.g., family violence).

Child Welfare Involvement. The associations between child welfare involvement and bullying perpetration appeared mixed (i.e., positive and nonsignificant) across the two cross-sectional studies. Specifically, Mohapatra et al. (2010) found that child protective services involvement was related to higher odds of bullying perpetration among girls but not boys. However, Edwards and Batlemento (2016) reported that youth in out-of-home placement did not differ significantly from youth in the six home-living arrangements regarding bullying perpetration via the Kruskal–Wallis H test.

Other ACE. Min et al. (2015) reported that neglect was positively related to bullying perpetration in a cross-sectional SEM controlling for physical and emotional abuse.

There were 36 studies, primarily cross-sectional (cross-sectional, n = 30; longitudinal, n = 6), that investigated the relations of ACEs with bullying victimization. Some studies examined more than one ACE in relation to bullying victimization. For example, CDC (2011) investigated the associations of physical abuse and domestic violence exposure with bullying victimization. This led to the number of associations examined to 58 (i.e., 43 positive, 15 nonsignificant, and 1 negative). Across all ACEs (see Table 3), the cumulative ACE score (i.e., composite ACEs), maltreatment, family violence, neglect, and child welfare involvement were positively associated with bullying victimization across all studies. Physical abuse, emotional abuse, and witnessing domestic violence appeared to be positively associated with bullying victimization as most studies showed positive associations, and the few studies reporting nonsignificant findings could be explained by their methodological features. The associations between the remaining ACEs and bullying victimization were not studied at all or inclusive as the total number of included studies was too small, or the results were split between positive and nonsignificant.

Overall, sample sizes were adequate, with three studies having relatively small samples (N = 129, Knous-Westfall et al., 2012; N = 125, Sterzing et al., 2016; N = 125, Wei & Lee, 2014). Most samples were recruited from schools or communities, except for five samples recruited from special groups, including youth from disadvantaged neighborhoods (Cluver et al., 2010), psychiatric youth (Mustanoja et al., 2011), youth involved with child welfare (Sterzing et al., 2020), sexual minority youth (Sterzing et al., 2016), and youth offenders (Viljoen et al., 2005). The predominant analysis was logistic regression, although other analyses were also used, including Chi-square tests, Kruskal–Wallis H tests, linear regression, multilevel modeling, path analysis, and SEM. Further, the predominant assessment of bullying victimization was self-report, except for three studies that included either parent report (Bowes et al., 2013; Mansbach-Kleinfeld et al., 2015) or peer nomination (Wei & Lee, 2014). Multiple studies used single-item measures (e.g., Mustanoja et al., 2011) or lacked information on measures (e.g., CDC, 2011), although multi-item measures were used in other studies (e.g., Williamson et al., 2018). Lastly, only six of the 36 studies were longitudinal, with one study on each of the following ACEs: abuse (Hong et al., 2017), physical abuse (Bowes et al., 2013), sexual abuse (Hébert et al., 2016), emotional abuse (Calvete et al., 2018), domestic violence exposure (Knous-Westfall et al., 2012), and divorce and separation (Wei & Lee, 2014). Thus, it is unclear across all ACEs how they affect bullying victimization longitudinally, even if they were reported as positively related to bullying victimization above (e.g., family violence).

Composite ACEs. Positive relations between the cumulative ACE scores and bullying victimization were reported in two cross-sectional studies that conducted logistic regression, with gender differences to consider. Specifically, Reisen et al. (2019) found that victims were more likely to report total adversity than non-victims (i.e.,