Introduction
Child maltreatment, including sexual, physical, and emotional abuse and neglect, is a prevalent issue and a global problem. It is estimated that approximately 1 in 4 children experience physical punishment or psychological violence from parents and caregivers (Hillis et al., Reference Hillis, Mercy, Amobi and Kress2016; Stoltenborgh et al., Reference Stoltenborgh, Bakermans-Kranenburg, Alink and van Ijzendoorn2015) while 1 in 7 experience incidents of sexual abuse (Stoltenborgh et al., Reference Stoltenborgh, van Ijzendoorn, Euser and Bakermans-Kranenburg2011). The consequences of child maltreatment can be devastating. Individually, it can lead to short and long-term impacts across various domains of functioning, including neurodevelopment, physical health, risk-taking behaviors, and mental health (Hughes et al., Reference Hughes, Bellis, Hardcastle, Sethi, Butchart, Mikton, Jones and Dunne2017; Luby et al., Reference Luby, Tillman and Barch2019; Madigan, Deneault, et al., Reference Madigan, Deneault, Racine, Park, Thiemann, Zhu, Dimitropoulos, Williamson, Fearon, Cénat, McDonald, Devereux and Neville2023; Murphy et al., Reference Murphy, Steele, Dube, Bate, Bonuck, Meissner, Goldman and Steele2014; Norman et al., Reference Norman, Byambaa, De, Butchart, Scott and Vos2012; Zhu et al., Reference Zhu, Exner-Cortens, Dobson, Wells, Noel and Madigan2024), as well as relational health (e.g., caregiver-child attachment; Cyr et al., Reference Cyr, Euser, Bakermans-Kranenburg and Van Ijzendoorn2010; Madigan, Fearon et al., Reference Madigan, Fearon, van Ijzendoorn, Duschinsky, Schuengel, Bakermans-Kranenburg and Verhage2023). Moreover, child maltreatment is a leading public health concern; it is estimated that countries spend 4% of their GDP annually on child maltreatment-related expenses such as health and social services (Hughes et al., Reference Hughes, Ford, Bellis, Glendinning, Harrison and Passmore2021). Due to these individual- and societal-level impacts, there is a pressing need to more precisely identify risk factors and intervention targets to prevent and address child maltreatment (Alink et al., Reference Alink, Cyr and Madigan2019; Berthelot et al., Reference Berthelot, Lemieux and Maziade2019).
Intergenerational risk of child maltreatment
Theoretical models propose that multiple factors, nested within various contextual layers, can influence children’s risk of experiencing maltreatment. These contexts include individual (e.g., age, temperament, and disabilities; Connelly & Straus, Reference Connelly and Straus1992; Maclean et al., Reference Maclean, Sims, Bower, Leonard, Stanley and O’Donnell2017; Pekdoğan & Kanak, Reference Pekdoğan and Kanak2022); family (e.g., number of children in the home, domestic violence, parenting stress, parent mental health issues; Chan, Reference Chan1994; Kelley et al., Reference Kelley, Lawrence, Milletich, Hollis and Henson2015; Kitamura et al., Reference Kitamura, Takauma, Tada, Yoshida and Nakano2004; McGuigan & Pratt, Reference McGuigan and Pratt2001); neighborhood (e.g., poor social support, housing instability, violence, high crime; Lynch & Cicchetti, Reference Lynch and Cicchetti1998; Marcal, Reference Marcal2018; Morris et al., Reference Morris, Marco, Maguire-Jack, Kouros, Bailey, Ruiz and Im2018); community (e.g., poverty, poor healthcare and/or social services access; Berger, Reference Berger2004; Negash & Maguire-Jack, Reference Negash and Maguire-Jack2016); and cultural and social factors (e.g., norms that condone violence and/or discourage help seeking; Klika et al., Reference Klika, Haboush-Deloye and Linkenbach2019).
In a recent umbrella review, a parent’s own history of child maltreatment was identified as one of the strongest risk factors for child maltreatment in the next generation (van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg, Coughlan and Reijman2020). This association aligns with early theories on the cycle of maltreatment– where Garbarino and Gilliam (Reference Garbarino and Gilliam1980) suggested that “the premier developmental hypothesis in the field of abuse and neglect is the notion of intergenerational transmission, the idea that abusing parents were themselves abused as children and that neglect breeds neglect” (p. 111). A meta-analysis of 142 studies (Madigan et al., Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019) explicitly tested the cycle of maltreatment hypothesis and the pooled association was an r of 0.22 (medium effect; Funder & Ozer, Reference Funder and Ozer2019), suggesting that parents with a history of maltreatment are twice as likely to have children who experience maltreatment, compared to those without such a history. This finding reflects heterotypic continuity of maltreatment, indicating that any type of childhood maltreatment experienced by a parent may elevate the risk of maltreatment for their child, regardless of the specific type of maltreatment (Berzenski et al., Reference Berzenski, Yates, Egeland, Korbin and Krugman2014). However, Madigan et al. (Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019) also found support for homotypic continuity in their meta-analysis. Homotypic continuity refers to continuity within subtypes of child maltreatment (i.e., physical abuse, sexual abuse, emotional abuse, and neglect). For example, the child of a parent who experienced physical abuse may also experience physical abuse in their childhood (Berzenski et al., Reference Berzenski, Yates, Egeland, Korbin and Krugman2014).
Patterns of continuity and discontinuity of childhood maltreatment across generations
The meta-analysis by Madigan et al. (Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019) established that parents’ history of child maltreatment is a risk factor for their own children’s potential for experiencing maltreatment. However, for several reasons, a more nuanced approach to studying the intergenerational transmission of maltreatment is needed. First, the child maltreatment literature has been critiqued for placing a disproportionate focus on the continuities of maltreatment (i.e., “abuse begets abuse”; Garbarino & Gilliam, Reference Garbarino and Gilliam1980; McKenzie et al., Reference McKenzie, Thompson, Hurren, Tzoumakis and Stewart2021; St-Laurent et al., Reference St-Laurent, Dubois-Comtois, Milot and Cantinotti2019; Thornberry et al., Reference Thornberry, Knight and Lovegrove2012) across generations, as opposed to discontinuities (i.e., cycle breakers and initiators). An overly deterministic view of this phenomenon may be detrimental, as it can lead to social policies and services that overlook the potential for change in the context of adversity.
Second, an exclusive focus on continuity fails to capture the multinominal nature of intergenerational pathways, which can be categorized into four patterns: (1) cycle maintainers, where both parents and children experience child maltreatment; (2) cycle breakers, where parents experience maltreatment but their child does not; (3) cycle initiators, where parents do not experience maltreatment but their child does; and (4) unaffected families, or controls, where neither generation experiences maltreatment.
In the largest study to date with 32,574 biological parent-child dyads conducted using Australian child protective service record data, the percentages of cycle maintainers, breakers, initiators, and controls were 2.2%, 6.9%, 2.8%, and 88.0%, respectively (McKenzie et al., Reference McKenzie, Thompson, Hurren, Tzoumakis and Stewart2021). Thus, while a parent’s history of maltreatment may be a risk factor for intergenerational maltreatment, parents may be more likely to break versus maintain cycles of maltreatment, and many may initiate as well as maintain, highlighting the complexities of intergenerational maltreatment.
Third, a scoping review of the literature by Langevin et al. (2021) highlighted that (dis)continuity varies considerably across individual studies (7% to 88%). This wide range of estimates suggests the need to explore potential moderating factors that may explain when continuity or discontinuity are more or less likely to occur. Several factors may help to explain differences in estimates of intergenerational maltreatment across studies, including demographic characteristics of the parent and child, as well as measurement and study design factors. Although child maltreatment occurs across all socioeconomic strata, socioeconomic status (SES) is an important moderator to consider, as evidence indicates higher incidences of child maltreatment in low SES groups (e.g., Hussey et al., Reference Hussey, Chang and Kotch2006; Madigan, Deneault, et al., Reference Madigan, Deneault, Racine, Park, Thiemann, Zhu, Dimitropoulos, Williamson, Fearon, Cénat, McDonald, Devereux and Neville2023, Madigan et al., Reference Madigan, Thiemann, Deneault, Fearon, Racine, Park, Lunney, Dimitropoulos, Jenkins, Williamson and Neville2025). Additionally, parent gender and age may be relevant to the prevalence of intergenerational transmission. Mothers are more often perpetrators than fathers, likely because they are more often the primary parents (van Berkel et al., Reference van Berkel, Prevoo, Linting, Pannebakker and Alink2020). Moreover, children of younger aged parents are at greater risk of child maltreatment (Boyer & Fine, Reference Boyer and Fine1992; Madigan et al., Reference Madigan, Wade, Tarabulsy, Jenkins and Shouldice2014; Zhu et al., Reference Zhu, Deneault, Turgeon and Madigan2025).
Child characteristics may also play a role in understanding the prevalence of intergenerational risk. For example, child sex is relevant as girls are at greater risk for specific types of maltreatment, such as child sexual abuse (Stoltenborgh et al., Reference Stoltenborgh, van Ijzendoorn, Euser and Bakermans-Kranenburg2011). Child age is another key factor, as maltreatment is typically defined as having occurred before the age of 18. When intergenerational transmission is examined in younger (vs. older) children, transmission rates may be underestimated due to a shorter exposure period (Madigan et al., Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019; Thornberry et al., Reference Thornberry, Knight and Lovegrove2012).
Measurement factors may further contribute to variability in estimates of intergenerational transmission of maltreatment (van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg, Coughlan and Reijman2020; Zhu et al., Reference Zhu, Deneault, Turgeon and Madigan2025). The type of assessment used is particularly relevant. Some studies rely on file review methods, often based on child protective services reports, while others use questionnaire measures or interviews. Retrospective recall measures, such as questionnaires and interviews, can lead to unreliable estimates, typically due to underreporting (e.g., Widom & Shepard, Reference Widom and Shepard1996). For example, in a meta-analysis by Baldwin et al. (Reference Baldwin, Reuben, Newbury and Danese2019) across 16 studies (25,471 participants), only 52% of individuals with historical reports of childhood maltreatment corroborated these experiences in retrospective self-reports. This underreporting can significantly influence estimates of intergenerational maltreatment. Measurement methods are also often linked to study design. Cross-sectional studies tend to use questionnaire-based methods, while longitudinal studies more frequently use file review methods or prospective data collection. Another methodological factor is whether the child maltreatment is substantiated or not. Substantiated maltreatment may be more severe and subject to frequent monitoring by child welfare, which may lead to higher prevalence estimates compared to unsubstantiated cases.
Taken together, examining moderators based on parent and child characteristics, as well as methodological factors, is important for understanding when and how the prevalence of maltreatment in cycle maintainers, breakers, and initiators may vary across studies. Advancing our understanding in this regard can inform policy decisions and guide the development of targeted prevention and intervention strategies, which could in turn more effectively support families and protect children.
Current study
To gain a better understanding of patterns of continuity and discontinuity in experiences of childhood maltreatment across generations, the current systematic review uses a framework for meta-analyzing multinomial data to provide reliable estimates of the distribution of cycle maintainers, breakers, initiators, and controls. Consistent with past studies, we examine patterns of heterotypic (i.e., G1 maltreatment including two or more types of maltreatment to G2 any maltreatment type) and homotypic continuity in our analyses (i.e., G1 physical abuse to G2 physical abuse; G1 sexual abuse to G2 sexual abuse; G1 emotional abuse to G2 emotional abuse; G1 neglect to G2 neglect).
In the meta-analysis by Madigan et al. (Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019), as well as others (Assink et al., Reference Assink, Spruit, Schuts, Lindauer, van der Put and Stams2018), both instances of victim-to-perpetrator maltreatment (i.e., parent victims of child maltreatment later become the perpetrators of their child’s maltreatment) and victim-to-victim maltreatment (i.e., parent victims of child maltreatment and their children experience maltreatment perpetrated by someone other than the parent) were included within the same analysis. This was primarily because few studies distinguish between these types of direct and indirect maltreatment transmission types (for an exception, see Widom et al., Reference Widom, Czaja and DuMont2015). Thus, we adopt the same approach herein of including instances of both victim-to-perpetrator and victim-to-victim maltreatment in our analyses.
In sum, the primary objective of this study was to synthesize data on the patterns of continuity and discontinuity of generational cycles of child maltreatment broadly, as well as specifically for subtypes of maltreatment, including child sexual abuse, physical abuse, emotional abuse, and neglect. The secondary aim was to conduct moderator tests to determine whether demographic factors (parent gender and age, child age and sex, family income), methodological factors (assessment method, substantiation of maltreatment), and study design (cross-sectional vs. longitudinal vs. file review) influence estimates of the prevalence of these forms of maltreatment continuity and discontinuity.
Methods
Definitional constructs
Our definitional criteria were guided by the international Consultation of Child Abuse Prevention (World Health Organization, 1999).
-
Physical abuse is defined as “that which results in actual or potential physical harm from an interaction or lack of an interaction, which is reasonably within the control of a parent or person in a position of responsibility, power or trust” (p. 15).
-
Neglect is defined as “the failure to provide for the development of the child in all spheres: health, education, emotional development, nutrition, shelter, and safe living conditions, in the context of resources reasonably available to the family or caretakers and causes or has a high probability of causing harm to the child’s health or physical, mental, spiritual, moral or social development. This includes the failure to properly supervise and protect children from harm as much as is feasible” (p. 15).
-
Sexual abuse is defined as “the involvement of a child in sexual activity that he or she does not fully comprehend, is unable to give informed consent to, or for which the child is not developmentally prepared and cannot give consent, or that violate the laws or social taboos of society” (p. 15).
-
Emotional or psychological abuse is defined as “the failure to provide a developmentally appropriate, supportive environment, including the availability of a primary attachment figure, so that the child can develop a stable and full range of emotional and social competencies commensurate with her or his personal potentials and in the context of the society in which the child dwells. There may also be acts towards the child that cause or have a high probability of causing harm to the child’s health or physical, mental, spiritual, moral or social development. Acts include restriction of movement, patterns of belittling, denigrating, scapegoating, threatening, scaring, discriminating, ridiculing or other non-physical forms of hostile or rejecting treatment” (p. 15).
-
Some studies examined general maltreatment experiences, characterized by the presence of two or more of the maltreatment subtypes above.
Search strategy
Searches were conducted in PsycINFO, MEDLINE, Embase, Social Work Abstracts, Web of Science and the Cochrane Central Register of Controlled Trials by a science librarian on April 1, 2018 and updated July 15, 2023, with no language restrictions. Database subject heading and text fields were searched for the concept of “maltreatment” and “intergenerational” (see eTables 1-3 for lists of key words used). We used truncation symbols and adjacency operators, when appropriate, to capture variations in spelling and phrasing. Another method of searching included the review of reference lists of articles meeting our inclusion criteria. This study was registered with PROSPERO (CRD42023473603) and reported using the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PRISMA; Page et al., Reference Page, McKenzie, Bossuyt, Boutron, Hoffmann, Mulrow and Moher2021; see Figure 1).
PRISMA flow diagram.
Selection criteria
Two coders reviewed all titles and abstracts (97% intercoder agreement) and discrepancies were resolved through consensus. They used the following inclusion criteria: (1) a measure of childhood maltreatment among parents and their offspring which aligned with the definitional criteria outlined by the international Consultation of Child Abuse Prevention by the World Health Organization; (2) child maltreatment ≤ 18 years of age; prevalence data (proportions with/without maltreatment history) for Generation 1 (G1; parents) and Generation 2 (G2; children); and (3) study written in the languages spoken by our team (English or French).
Exclusion criteria included: (1) nonempirical publications, such as descriptive reports, case studies, or book and narrative reviews; (2) studies focused on witnessing or committing intimate partner violence; and (3) studies examining potential for, but not perpetration of, maltreatment.
Full-text screening was conducted by two coders for all relevant studies (86.8% intercoder agreement) and discrepancies were resolved through consensus.
Data extraction
Data was extracted using a pre-developed protocol that included coding of sample size, and proportions of cycle maintainers, breakers, initiators, and controls for overall maltreatment and any subtype of child maltreatment examined. Sociodemographic, methodological, and measurement factors were also coded to test for potential moderators. Specifically, family socio-economic status (SES) was operationalized as low or mixed according to the information provided in each study. The mean age of caregivers and children (in years) was also extracted. As for parent gender and child sex, the proportion (%) of mothers in G1 and the proportion (%) of females in G2 was recorded. Regarding methodological factors, the maltreatment measure in each generation was documented as follow: single question, questionnaire, interview, or child protective services records. We also specified whether the G2 maltreatment reports had been substantiated or not. Finally, with respect to measurement factors, each study was classified as using a cross-sectional, longitudinal, or retrospective file review of official records research design.
When studies provided insufficient information for the calculation of prevalence estimates, study authors were contacted directly. Of the authors contacted (N = 62), 26 responded and 12 provided data for our meta-analysis. Two coders conducted data extraction, with 20% coded for inter-rater reliability (94% intercoder agreement). Discrepancies were resolved via consensus.
Study quality assessment
The 22-item Strengthening the Reporting of Observational Studies in Epidemiology Statement Guidelines for Reporting Observational Studies checklist was used to assess study quality (National Heart, Lung, & Blood Institute, 2014). Studies were given a score of 0 (no) or 1 (yes) for each of the 22 items (see eTable 4). One coder evaluated all studies for study quality, and 20% were coded for inter-rater reliability (87% intercoder agreement. Discrepancies resolved through consensus.
Data preparation and analysis
Unlike traditional meta-analyses of means or correlations, meta-analyses of prevalence effectively allow the analyst to recover the raw case-level data. Frequency counts of each of the four groups were extracted from the original papers, and used as frequency weights using STATA’s fweight command, with each group coded as 0, 1, 2 and 3 respectively in a single multinomial variable. Random effects were estimated to capture between-study variability in the four maltreatment groups. We extended the standard multilevel approach to meta-analysis to the multinomial case by using a multilevel logit model with random intercepts (Skrondal & Rabe-Hesketh, Reference Skrondal and Rabe-Hesketh2003). In this framework, the between-study variability is estimated by k-1 random intercepts, reflecting variance in the contrast between one category and a reference category (in this study, we selected the controls as the reference category). The k-1 random effects were assumed to correlate. Estimation was by maximum likelihood and adaptive Gauss–Hermite quadrature using the gsem package of STATA version 17. All estimates are presented with 95% confidence intervals (CIs). Due to the number of analyses conducted, we set a more conservative p-value of .01 to assess the significance of moderators (Lakens et al., Reference Lakens, Adolfi, Albers, Anvari, Apps, Argamon, Baguley, Becker, Benning, Bradford, Buchanan, Caldwell, Van Calster, Carlsson, Chen, Chung, Colling, Collins, Crook and Cross2018). We did not conduct publication bias testing as it is not recommended for analyses of proportion meta-analyses (Barker et al., Reference Barker, Migliavaca, Stein, Colpani, Falavigna, Aromataris and Munn2021).
We chose a range of moderators based on methodological and substantive considerations, as noted above, and included moderator variables in the analyses if at least two group sizes were k = 3 or more. Moderator categories of k < 3 were excluded. Analyses were considered statistically significant if p < .01. Categorical moderators were dummy variable coded, and all group contrasts were tested using STATA’s margins command, when an omnibus Wald test of any group differences in prevalence was significant. Continuous variables were treated as such in the analysis, and estimates obtained at ±1 SD of the moderator for interpretation purposes.
Results
The search strategy identified 2771 non-duplicate titles/abstracts, of which 463 articles met initial screening for inclusion and their full texts were reviewed. Ultimately, 65 independent studies and 80 samples met full inclusion criteria.
Study characteristics
Details regarding study characteristics are provided in Table 1. Of the 65 independent studies, 30 examined cycles of maltreatment generally, while 27 reported prevalence data for G1 and G2 physical abuse, 17 for G1 and G2 sexual abuse, 5 for G1 and G2 neglect, and 1 study for G1 and G2 emotional abuse. Sample size varied considerably between studies, ranging from 18 to 49,730 dyads (mean = 2,394; median 382.5). Studies were mostly conducted in North America (k = 42, 64.6%) or Europe (k = 10, 15.4%), with the rest being conducted in Asia (k = 6, 9.2%), South America (k = 1, 1.5%), and Australia/New Zealand (k = 4, 6.2%). One study (k = 1, 1.5%) reported on data from North America and South America, and one study did not report on study geographical location (k = 1, 1.5%). The mean score across studies on study quality was 19.8 (range: 13-22; see eTable 4).
Characteristics of studies included
Note. CS = Cross-sectional; L = longitudinal; RFR = retrospective file review; GM = general maltreatment; PA = Physical Abuse; SA = Sexual Abuse; N = neglect; EA = Emotional Abuse. aSome studies are included more than once if data were provided for more than one maltreatment category.
Sample characteristics
Among the 80 samples, the parent generation (G1) had a mean age range between 19.7 and 56.9 years (mean = 35.1), and the average proportion of females was 81.6%. The child generation (G2) had a mean age range between 0.4 and 23.7 years (mean = 10.8), and the average proportion of females was 60.9%.
Most studies (k = 46; 57.5%) used a cross-sectional design, whereas 28 (35%) used a longitudinal design, and 6 (7.5%) used a retrospective file review methodology. Different measures were used to assess child maltreatment experiences, including questionnaires (G1 n = 39, 49%; G2 n = 25; 31%), single questions (G1 n = 9, 11%; G2 n = 4; 5%), interviews (G1 n = 20, 25%; G2 n = 11; 14%), official child protective services records (G1 n = 11, 14%; G2 n = 33; 41%), mixed methods (G1 n = 1, 1%; G2 n = 6; 7.5%), and observations (G2 only, n = 1, 1%). G2 maltreatment data were obtained from different types of informants, such as agency reports in 33 studies, parent reports in 27 studies, self-reports in 12 studies, and at least two informants (mixed reports) in 8 studies.
Child maltreatment across generations: prevalence and moderators
Prevalence. There were 30 studies (76,047 dyads) included in this analysis. In a multilevel multinomial model with no covariates (i.e., moderators), the distribution of the four groups – maintainers, breakers, initiators, and controls – was estimated to be 17.1% (95% CI: 12.1%, 22.1%), 23.6% (95% CI: 18.0%, 29.2%), 11.4% (95% CI: 7.8%, 15.1%) and 47.8% (95% CI: 39.7%, 55.9%), respectively. All three random effects representing between-study heterogeneity [equivalent to tau-squared] in the difference between each group and the control group, were significantly greater than zero (maintainers vs. controls: 2.22, se = .59; 95% CI: 1.32, 3.73; breakers vs. controls: 1.32, se = .36, 95% CI: .77, 2.25; initiators vs. controls: 1.73, se = 0.46; 95% CI: 1.03, 2.90), indicating significant differences in prevalence rates between studies.
Moderator analyses. The multilevel models were re-estimated one at a time with the inclusion of categorical or continuous moderators, to test for differences in the odds of falling into one of the three maltreatment categories, relative to the control category. The results of these analyses are presented in Table 2. Several significant moderators emerged from these analyses.
Results of multinomial hierarchical modeling of prevalences of intergenerational patterns of maltreatment
Firstly, there were reliable differences in the prevalence of the intergenerational maltreatment groups, relative to the control group, between studies deploying different research designs (χ2(6) = 22.87, p = .0008). The estimated marginal probabilities of falling into each of the four groups is plotted in Figure 2. Marginal contrast tests between cross-sectional studies on the one hand, and longitudinal and file review studies on the other, showed that cross-sectional studies estimated higher rates of maintainers than longitiudinal studies (18.4% [95% CI: 9.7%, 27.0%] higher. Cross-sectional studies also produced higher estimates of rates of maintainers than file review studies (10.8% [–3.8, 25.3] higher), but the difference was not significant. File review studies also produced higher estimates than longitudinal studies, although the difference was also non-significant (7.6% [95% CI: –5.1%, 20.3%]). No significant differences in the estimates of breakers or initiators were seen between the three designs.
Estimated group probabilities for overall maltreatment by study design (error bars are 95% CIs). Note. Circles are individual study prevalence estimates; size of circle is proportional to the sample size.
Secondly, the method of assessment in the parent generation was associated with differences in the estimated group prevalence rates (χ2(6) = 28.04, p = .0001). The estimated probabilities are plotted in Figure 3. File review studies produced lower estimates of maintainers than questionnaires (15.7% lower [95% CI: 8.1%, 23.4%]) but not interviews (only 3.3% lower [95% CI: -4.1%, 10.8%]). Questionnaires also produced higher estimates than interviews (12.4% higher [95% CI: 3.7%, 21.1%]). There were no significant differences in estimates of the prevalence of breakers between the different methods. However, file review produced lower estimates of initiators than questionnaires (7.3% lower [95% CI: 2.2%, 12.3%]) and interviews (14.5% lower [95% CI: 4.5%, 24.5%]), while questionnaires and interviews produced similar estimates. There was only very weak (p = .02) indication that the assessment method in the child population affected estimated prevalence rates.
Estimated group probabilities for overall maltreatment by parent assessment method (error bars are 95% CIs). Note. Circles are individual study prevalence estimates; size of circle is proportional to the sample size.
The percentage of the parent generation that was female also emerged as a significant moderator (χ2(3) = 14.23, p < .003). Figure 4 plots the marginal probabilities of the maltreatment categories at two levels of the distribution of the percentage of females in the sample (60 and 100% [+ − 1 SD from the mean). Marginal contrasts showed that a higher percentage of women in the study sample was associated with a significantly higher estimate of maintainers: the prevalence was 9.7% [95% CI 2.9%, 16.5%] higher in the 100% group of studies than the 60% group of studies). The difference was not significant for cycle breakers. The prevalence of initiators was also 8% [95% CI 2.8, 13.2] higher in the 100% female group of studies than the 60% female studies. There was little evidence that the percentage of female participants in the G2 sample made a difference to the estimates of the group prevalences. No other moderators were significant at the p < .01 level.
Estimated group probabilities for overall maltreatment by percent female in parent sample (error bars are 95% CIs). Note. Circles are individual study prevalence estimates; size of circle is proportional to the sample size. Marginal probabilities of the maltreatment categories are illustrated at two levels of the distribution of the percentage of females in the sample (60% and 100% [± 1 SD from the mean]).
Child sexual abuse, physical abuse, emotional abuse, and neglect across generations
To explore the prevalence of the intergenerational maltreatment groups for homotypic transmission within each maltreatment type (i.e., G1 physical abuse to G2 physical abuse), we ran the multinomial models for all studies that report each maltreatment type (see Figure 5).
Prevalence of the intergenerational maltreatment groups for each maltreatment type.
Physical Abuse. Based on 27 studies (104,133 dyads), the estimated prevalence of the four types of intergenerational pattern of G1 to G2 physical abuse were: 18.5% maintainers (95% CI: 11.9%, 25.2%); 23.2% breakers (95% CI: 18.7%, 27.7%); 9.0% initiators (95% CI: 7.0%, 11.1%); and 49.3% controls (95% CI: 40.5%, 58.1%).
Sexual Abuse. Based on 17 studies (6,889 dyads) the estimates from G1 experiences of sexual abuse to G2 experiences of sexual abuse were: 16.2% maintainers (95% CI: 10.8%, 21.7%); 13.9% breakers (95% CI: 9.5%, 18.3%); 18.9% initiators (95% CI: 12.4%, 25.4%); and 51.0% controls (95% CI: 41.8%, 60.2%).
Neglect. Based on 5 studies (4,298 dyads), the estimates from G1 neglect to G2 neglect were 9.7% maintainers (95% CI: 1.6%, 19.2%); 21.4% breakers (95% CI: 8.9%, 33.9%); 13.2% initiators (95% CI: 5.0%, 21.4%); and 55.7% controls (95% CI: 43.3%, 68.1%).
Emotional abuse. Only one study examined G1 to G2 emotional abuse and therefore only a narrative summary of this study can be provided. Rikić et al. (Reference Rikić, Beljan, Milošević, Miškulin, Miškulin and Mujkić2017) demonstrated among 118 G1 participants that emotional/psychological abuse, including shouting, ridiculing, and criticizing, directed towards G2 was predicted by the presence of emotional/psychological abuse experienced by G1 at the hands of their own parents.
Moderators of maltreatment subtype prevalences
The groups of studies that reported on prevalence rates for physical abuse and sexual abuse across generations were large enough to conduct moderator analyses (having moderators with smallest group sizes of k = 3 or more). The parameter estimates from the multilevel multinomial models for physical abuse are shown in Table 3. When estimating the effect of the percentage of females in the first generation the model did not converge. Splitting the variable into two groups, with 100% female samples in one group and <100% female samples in the other led to model convergence. There were marked differences in prevalence estimates between samples where the first generation was 100% female, versus studies that were not (χ 2(3) = 771.6, p < .001). The predicted probabilities are shown in Figure 6. Contrasts tests showed that the initiator group was more prevalent in 100% female samples than in the mixed gender samples (7.6% higher [95% CI 3.9%, 11.3%). There were also differences in prevalence depending on whether the G2 abuse had been substantiated or not (χ 2(3) = 335.0, p < .001). Predicted probabilities are plotted in Figure 7. Rates of maintainers were 20.9% lower (95% CI 2.8%, 39.0%) and rates of initiators were 10% lower [95% CI 5.9%, 14.2%] when reports had been substantiated. The rates of cycle breakers did not vary reliably in relation to whether reports had been substantiated or not. No other moderators were significant at the p < .01 level.
Estimated group probabilities for physical abuse by percent female in parent sample (error bars are 95% CIs). Note. Circles are individual study prevalence estimates; size of circle is proportional to the sample size. Marginal probabilities of the maltreatment categories are illustrated at two levels of the distribution of the percentage of females in the sample (60% and 100% ± 1 SD from the mean]).
Estimated group probabilities for physical abuse by substantiated report status (error bars are 95% CIs). Note. Circles are individual study prevalence estimates; size of circle is proportional to the sample size.
Results of multinomial hierarchical modeling of prevalences of intergenerational patterns of physical abuse
The parameter estimates for the analyses of sexual abuse are shown in Table 4. No moderators were significant at the p < .01 level.
Results of multinomial hierarchical modeling of prevalences of intergenerational patterns of sexual abuse
Discussion
In the current meta-analysis, we estimated prevalence rates for patterns of child maltreatment across generations (G1 to G2). Specifically, for general maltreatment, across 30 studies (76,047 dyads), we found that 17.1% of parents maintained the cycle of maltreatment across generations, whereas 23.6% broke the cycle of maltreatment. This contrast highlights the nuanced nature of intergenerational transmission. While a parent’s history of maltreatment significantly increases the risk for child maltreatment in the next generation (Madigan et al., Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019; van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg, Coughlan and Reijman2020; Zhu et al., Reference Zhu, Deneault, Turgeon and Madigan2025), it is important to recognize that most parents with such histories do not repeat these harmful patterns. These findings underscore both the risks associated with childhood maltreatment and the resilience that many parents demonstrate in breaking cycles of maltreatment. Moreover, our results suggest that the prevalence of cycle initiators for general maltreatment is 11.4%, which is lower than the prevalence of cycle maintainers. Our findings advance understanding of the patterns of maltreatment across generations, which can inform the development of policies and strategies aimed at preventing and addressing child maltreatment.
In the current study, we focused exclusively on one risk factor for child maltreatment - the parent’s own history of child maltreatment. However, there are numerous risk factors at the individual, family, neighborhood, community, and cultural levels that can threaten a child’s safety. Moreover, many of these risk factors intersect, accumulate, and/or can exacerbate each other to predict cycles of intergenerational risk. In a study by McKenzie et al. (Reference McKenzie, Thompson, Hurren, Tzoumakis and Stewart2021) examining risk factors distinct to cycle maintainers, breakers, and initiators, it was observed that cycle maintainers were younger at the time of their first child’s birth, had a greater number of children, and were more likely to be unmarried, compared to cycle breakers. Cycle initiators had a comparable number of risk factors to cycle maintainers in McKenzie et al.’s study; however, being male versus female presented an additional risk for child maltreatment initiation. Interestingly, McKenzie et al., also found that cycle breakers experienced a similar number of risk factors, albeit to a lesser extent, than cycle maintainers.
McEwen (Reference McEwen2012) and others have suggested that a history of maltreatment gets “under the skin,” influencing biological (e.g., altered stress response) and psychological functions (e.g., cognitive and emotional) that shape how parents treat their own children (Alink et al., Reference Alink, Cyr and Madigan2019; Assink et al., Reference Assink, Spruit, Schuts, Lindauer, van der Put and Stams2018; Madigan et al., Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019; Thornberry et al., Reference Thornberry, Knight and Lovegrove2012; van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg, Coughlan and Reijman2020). In line with a developmental psychopathology perspective (Cicchetti & Cohen, Reference Cicchetti and Cohen1995; Masten & Garmezy, Reference Masten and Garmezy1985; Rutter, Reference Rutter1979; Sroufe, Reference Sroufe2009) an important avenue of future research lies in the examination of when and why a history of maltreatment gets under the skin for some, but not all those with a history of maltreatment. Moreover, it will be important to explore how specific risk and protective factors interact to differentiate parents who break versus maintain or initiate cycles of maltreatment. There is considerable research evidence to suggest that children who experience safe, stable, and nurturing relationships may be more likely to break versus maintain the intergenerational cycle of maltreatment (Egeland et al., Reference Egeland, Jacobvitz and Sroufe1988; Jaffee et al., Reference Jaffee, Bowes, Ouellet-Morin, Fisher, Moffitt, Merrick and Arseneault2013; Schofield et al., Reference Schofield, Lee and Merrick2013), and may also be more engaged and responsive to therapy when it is received following maltreatment (Eirich et al., Reference Eirich, Racine, Garfinkel, Dimitropoulos and Madigan2020). However, greater insights into the joint and interactive contributions of risk and protective factors could both advance understanding of the pathways of intergenerational maltreatment and powerfully inform interventions aiming to break harmful cycles of intergenerational risk and toxic stress (Turgeon et al., Reference Turgeon, Racine, McDonald, Tough and Madigan2024).
This meta-analysis also examined continuity of maltreatment across maltreatment subtypes, including physical abuse, sexual abuse, and neglect. Importantly, we found varying prevalence rates based on the type of maltreatment, suggesting different mechanisms are at play for diverse maltreatment subtypes, which warrants consideration in future research. For physical abuse, across 27 studies, the prevalence breakdowns mirrored those of general maltreatment, where cycle breakers (23.2%) were the largest group (after controls, 49.3%), followed by maintainers (18.5%) and initiators (9.0%). This means that despite having experienced physical abuse in their own childhoods, most parents do not perpetrate physical abuse against their children.
Our findings align with previous research showing weaker effects for the intergenerational transmission of physical abuse (Widom et al., Reference Widom, Czaja and DuMont2015). Several protective factors may help to break this cycle, such as having safe, stable, and/or nurturing relationships in childhood (Egeland et al., Reference Egeland, Jacobvitz and Sroufe1988; Jaffee et al., Reference Jaffee, Bowes, Ouellet-Morin, Fisher, Moffitt, Merrick and Arseneault2013; Schofield et al., Reference Schofield, Lee and Merrick2013), or a healthy romantic relationship (Jaffee et al., Reference Jaffee, Bowes, Ouellet-Morin, Fisher, Moffitt, Merrick and Arseneault2013). These supportive relationships can mitigate the risk of perpetrating maltreatment across generations. Additionally, having a child with an easier temperament and high self-control may also reduce the likelihood of parents engaging in physical abuse (Schofield et al., Reference Schofield, Conger and Conger2017). Differential susceptibility theory suggest that parents who were less susceptible to their environments as children may be better able to regulate their emotions and cope with stress, despite their own maltreatment experiences (van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg, Coughlan and Reijman2020). As adults, these individuals may adopt more effective parenting strategies and create supportive environments for their own children, potentially breaking the cycle of maltreatment.
The continuity of sexual abuse across generations across 17 studies showed a different pattern, as the percentage of cycle breakers (13.9%) was lower compared to general maltreatment (23.6%). The percentage of initiators for the sexual abuse subtype was higher (18.9%) as compared to general maltreatment (11.4%). It is important to note that the sexual abuse experienced by G2 was not necessarily perpetrated by the G1 parent. It may be that parents who have experienced their own maltreatment, such as neglect or domestic violence, have greater difficulty in fostering relationships and environments for their children that keep them safe from perpetrators of sexual abuse, which may contribute to the higher number of instances of initiated sexual abuse and a lower number of cycle breakers (Testa et al., Reference Testa, Hoffman and Livingston2011). Although maternal exposure to sexual abuse is a strong predictor of child exposure to sexual abuse, other risk factors likely mediate this association including interparental violence, parental substance abuse, and parent psychopathology (McCloskey & Bailey, Reference McCloskey and Bailey2000). Indeed, previous research has shown that maternal drug use is one of the strongest predictors of child sexual abuse risk (McCloskey & Bailey, Reference McCloskey and Bailey2000). Future research that examines the relationship among child victims and perpetrators of sexual abuse may shed additional light on these intergenerational patterns.
The current study revealed some important findings with regards to the intergenerational patterns of neglect across generations. However, prior to a fuller discussion, one note of caution is that this analysis is based on 5 studies. Accordingly, there is significant heterogeneity both in terms of the prevalence of intergenerational maltreatment found in each individual study and in the method within these 5 studies of assessing neglect, particularly for G1 (which included interviews, official records, questionnaires). With these caveats in mind, our findings suggest that parents who were exposed to neglect were more likely to have children who did not experience neglect in the next generation (9.7% for maintainers). Childhood neglect is the most common form of child maltreatment (Mennen et al., Reference Mennen, Kim, Sang and Trickett2010) and involves failing to provide for a child’s physical, emotional, social, safety, educational, or health needs (Stoltenborgh et al., Reference Stoltenborgh, Bakermans-Kranenburg and van Ijzendoorn2013). Neglect is strongly related to other social factors such as poverty, unemployment, and parental mental health difficulties (Shanahan et al., Reference Shanahan, Runyan, Martin and Kotch2017; Slack et al., Reference Slack, Holl, McDaniel, Yoo and Bolger2004). Although risk factors for neglect are multi-factorial, it may be that shifts in these contributing risk factors are more common and that neglect comes to the attention of systems (e.g., child welfare) that provide support, subsequently decreasing the likelihood of neglect in subsequent generations.
We also tested whether differences in prevalence rates across groups could be explained by study and sample characteristics. Our results showed that cross-sectional studies generated higher estimates for maintainers compared to longitudinal studies and file review studies. In cross-sectional studies, past and current child maltreatment of both G1 and G2 were assessed at the same time point. The higher prevalence of maintainers may be due to possible reporting bias in the studies using the same reporter for both experienced and perpetrated/experienced maltreatment in the two generations (Buisman et al., Reference Buisman, Pittner, Tollenaar, Lindenberg, van den Berg, Compier-de Block, van Ginkel, Alink, Bakermans-Kranenburg, Elzinga and van Ijzendoorn2020; Pears & Capaldi, Reference Pears and Capaldi2001).
Notably, for general maltreatment, file review studies showed the lowest prevalence estimates in which one or both generations were exposed to maltreatment. This may be due to that fact that file reviews typically result in lower prevalence estimates of child maltreatment more broadly (Madigan et al., Reference Madigan, Thiemann, Deneault, Fearon, Racine, Park, Lunney, Dimitropoulos, Jenkins, Williamson and Neville2025; Stoltenborgh et al., Reference Stoltenborgh, Bakermans-Kranenburg, Alink and van Ijzendoorn2015; van Berkel et al., Reference van Berkel, Prevoo, Linting, Pannebakker and Alink2020), as many victims are not reported to child service agencies and therefore do not appear in these official records. This likely leads to an underestimation of intergenerational maltreatment. This may also explain the low prevalence rates for maintainers among studies using official child maltreatment records (in the analyses on assessment method), since these studies likely overlap with file review studies.
This meta-analysis showed that female parents were more likely to be cycle maintainers and initiators for general maltreatment, and more likely to be initiators for sub-analyses on physical abuse. Mothers, more often than fathers, are the main point of contact within child protective services. As such, it may be more common to identify continuity and initiation that is perpetrated by mothers (Wall-Wieler et al., Reference Wall-Wieler, Brownell, Singal, Nickel and Roos2018). In part, this may be explained by the fact that despite considerable changes in social norms over the past few decades, mothers still bear a disproportionate burden of childcare responsibilities (Lee & Hofferth, Reference Lee and Hofferth2017). Increased maternal stress, substance use, and mental health issues are all known risk factors for child maltreatment (Niu et al., Reference Niu, Liu and Wang2018). Research also suggests that women, compared to men, are more likely to experience severe forms of intimate partner violence victimization, including physical and psychological victimization and sexual violence (Caldwell et al., Reference Caldwell, Swan and Woodbrown2012). Greater exposure to such types of intimate partner violence can erode parents’ well-being and affect parenting strategies (Sousa et al., Reference Sousa, Siddiqi and Bogue2021). Taken together, these factors may play a particular role in the higher likelihood of initiation and maintenance among female caregivers.
Lastly, moderator analyses revealed the prevalence for physical abuse uniquely differed depending on whether the abuse had been substantiated or not for G2. Specifically, estimates of cycle maintainers and initiators were 20.9% and 10.0% lower when reports of child physical abuse had been substantiated, while rates of cycle breakers showed no variation. One possibility for the lower prevalence of maintainers is that families with a history of substantiated physical abuse, which is often more readily observed than other forms of maltreatment (e.g., the hidden hurt of neglect), can be subjected to greater monitoring by child protective services, resulting in more supports and thus a lower likelihood of transmission. Conversely, unsubstantiated cases may be less severe or overt, and therefore fail to be detected or treated by social services, but the intergenerational risk would remain.
Study Limitations
Several limitations of this meta-analysis should be noted. First, there were insufficient studies specific to intergenerational patterns of emotional to conduct an analysis on this subtype of maltreatment. Although a narrative report of the one existing study was provided, a future meta-analysis is warranted when sufficient studies become available.
Second, our analyses only examined the presence versus absence of childhood maltreatment experiences, measured by a single question in a handful of studies, which fails to consider the potential impacts of their severity and chronicity. It is plausible that patterns of intergenerational continuity are more pronounced in cases of more severe, repeated, and/or prolonged childhood maltreatment experiences (McKenzie et al., Reference McKenzie, Thompson, Hurren, Tzoumakis and Stewart2021; St-Laurent et al., Reference St-Laurent, Dubois-Comtois, Milot and Cantinotti2019; Thornberry et al., Reference Thornberry, Knight and Lovegrove2012). For example, the duration of maltreatment may amplify its intergenerational effects. Future research should prioritize exploring these dimensions to advance understanding of how varying levels of maltreatment severity and chronicity influence (dis)continuity of intergenerational maltreatment.
Third, due to limited information in individual studies, we were unable to test certain moderator variables that are likely important for understanding child maltreatment exposure across generations. For example, we could not disentangle whether the perpetrator of the abuse was the parent or an intrafamilial or extrafamilial perpetrator. This distinction - between maltreated parents who go on to maltreat their own children versus maltreated parents whose children are maltreated by someone else - may be critical for understanding cycles of generational risk. The implications of this distinction warrant broader exploration in future research, particularly in relation to different risk factors and mechanisms that may underlie child maltreatment by non-parental perpetrators compared to parent (to child) perpetrators. Consideration of these nuanced pathways is essential for supporting the development and evaluation of targeted prevention and intervention efforts.
Lastly, we also had too few studies reporting on the age of the parent at the time of the child’s birth. This could be important as literature suggests that children of younger parents are more likely to have experienced childhood adversity and maltreatment (Madigan et al., Reference Madigan, Wade, Tarabulsy, Jenkins and Shouldice2014; Zhu et al., Reference Zhu, Deneault, Turgeon and Madigan2025) and be reported to child protective services (Fallon et al., Reference Fallon, Ma, Black and Wekerle2011). Moreover, while we aggregated data from as many countries as possible, the majority of studies are from North America and Europe (80%), which limits the global generalizability of our findings. Very few studies provided information about the race/ethnicity of the samples, which limits testing of this important demographic variable as a moderator of diverse patterns of intergenerational continuity.
Conclusions
Examining patterns of intergenerational continuity and discontinuity are essential for effective prevention and intervention efforts. It is equally important to rigorously test the hypothesis—and challenge the potentially inaccurate public assumption—that “abuse begets abuse.” While a parent's history of abuse is a known risk factor for child maltreatment in the next generation (Madigan et al., Reference Madigan, Cyr, Eirich, Fearon, Ly, Rash, Poole and Alink2019; Van IJzendoorn et al., Reference van IJzendoorn, Bakermans-Kranenburg, Coughlan and Reijman2020; Zhu et al., Reference Zhu, Deneault, Turgeon and Madigan2025), results from the current meta-analysis suggest that a greater proportion of parents break the cycle of maltreatment versus maintain it. By providing precise estimates of intergenerational pathways of child maltreatment, the current study can inform the development of policies and strategies to prevent and address these issues, with the ultimate goal of fostering safer and healthier environments for future generations.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0954579425000239.
Data availability statement
The data and code used in this study are available upon request from the corresponding author. This study was registered with PROSPERO (CRD42023473603).
Acknowledgements
The authors would like to thank Cheri Nickel, MLIS (University of Calgary) who conducted the literature search, as well as Anh Ly who helped with data management.
Funding statement
This study has been funded by an anonymous donor and the Calgary Health Foundation, as well as support from the Alberta Children’s Hospital Research Foundation (through the Alberta Children’s Hospital Research Institute and the Owerko Centre for Neurodevelopment and Child Mental Health), as well as the University of Calgary’s Cumming School of Medicine, Faculty of Arts, and Office of the Vice-President (Research).
Competing interests
None.
Open access
