Abstract
-
Purpose
This study aimed to identify ecological risk and protective factors associated with membership in latent classes characterized by heterogeneous associations between parental adverse childhood experiences (ACEs) and the intergenerational transmission of child maltreatment (ITCM).
-
Methods
This secondary analysis used data from the 2017 Family with Children’s Life Experience Survey. Participants were parents who had a spouse and children younger than 18 years. Guided by Belsky’s ecological model, the covariates were categorized as parent-, family/household-, and community-level contextual factors. Latent class regression was used to identify classes with differing associations between parental ACEs and child maltreatment, and multinomial logistic regression was used to examine factors associated with class membership.
-
Results
Three classes were identified: the “None-ACE-Association” class (58.5%), the “Partial-ACE-Association” class (23.3%), and the “Comprehensive-ACE-Association” class (18.2%). Factors across multiple ecological domains differentiated class membership. Acceptance of corporal punishment significantly distinguished the None-ACE-Association class from the Partial-ACE-Association class. In comparisons involving the Comprehensive-ACE-Association class, membership was significantly associated with several parent- and family/household-level factors, including depressive symptoms, parenting stress, partner violence/control, and family relationship quality. Significant associations were also observed for the number of children, the age group of the youngest child, and residential area.
-
Conclusion
Distinct latent class profiles characterized by heterogeneous associations between parental ACEs and child maltreatment support the need for tailored, risk-stratified prevention strategies. Combining the assessment of parental ACEs with the assessment of modifiable factors could improve early identification and inform nursing and community-based interventions intended to interrupt the ITCM.
-
Key words: Adverse childhood experiences; Child abuse; Intergenerational relations; Latent class analysis; Risk
INTRODUCTION
Child maltreatment is increasingly being treated as a major public concern and addressed through an expanded government role in South Korea’s child protection system [
1]. Nevertheless, physical discipline continues to be widely perceived as an effective child-rearing practice, often discussed in relation to Confucian-influenced norms, which may increase maltreatment risk [
2]. A national survey in Korea reported that approximately 83% of child maltreatment perpetrators were parents, underscoring the need to identify parent-related determinants and inform family focused prevention efforts [
3].
Among parent-related risks, adverse childhood experiences (ACEs) are critical for understanding the current maltreatment risk [
4]. Parental ACEs have been linked to intergenerational risk processes that may compromise sensitive and responsive caregiving and increase the likelihood of child maltreatment and related adverse outcomes [
5]. These patterns are broadly discussed in the literature as intergenerational transmission of child maltreatment (ITCM), defined as the continuity of maltreatment-related risk across generations, whereby parents’ childhood maltreatment histories or other adverse childhood experiences are associated with an increased likelihood of maltreatment-related parenting toward their children [
6,
7]. Theoretically, ITCM has been examined through multiple frameworks, with attachment theory and trauma-related models being particularly prominent, while ecological, social, developmental, and biological perspectives have also contributed to understanding how intergenerational risk is shaped by early caregiving experiences, learned relational patterns, and broader developmental and contextual processes [
8]. Importantly, this continuity should not be understood as deterministic; rather, meta-analytic and review evidence indicates that parental maltreatment history is a consistent but non-uniform risk factor, underscoring substantial heterogeneity in ITCM [
6,
7,
9]. Such heterogeneity is reflected in variation in maltreatment subtypes, chronicity, developmental timing, and psychosocial risk and protective factors operating across individual, relational, contextual, and historical levels [
6,
7,
10].
Reported ITCM prevalence varies widely (approximately 7%–70%), likely due in part to methodological differences in the operational definition of ITCM, the measurement of parental adversity and child maltreatment, the maltreatment subtypes included, the potential for self-report bias, and the analytic strategies used [
11,
12]. To better address such heterogeneity, person-oriented analytic approaches have gained attention in child maltreatment research [
10]. Unlike variable-oriented approaches, which estimate average associations across the full sample, person-oriented approaches are designed to identify subgroups of individuals who show distinct patterns within the phenomenon of interest [
11]. In child maltreatment research, person-oriented methods such as latent class analysis and latent profile analysis have often been used to classify individuals according to patterns of maltreatment or adversity exposure [
10]. By contrast, latent class regression was particularly appropriate for the present study because it enabled us to examine whether the associations between parental ACEs and current child maltreatment differed across subgroups, rather than assuming a single average pattern across the full sample [
11-
13]. However, many ITCM studies have relied primarily on variable-oriented approaches (e.g., linear regression or structural equation modeling) [
11,
12], which may obscure meaningful heterogeneity. In the present study, latent class regression was used to identify subgroups characterized by graded levels of maltreatment involvement and differential ACE-domain associations, rather than clearly distinct transmission pathways [
13].
In addition to heterogeneity in ACEs–ITCM pathways, ITCM is shaped by multilevel risk and protective factors. Therefore, ecological frameworks are widely used to conceptualize ITCM as arising from interacting influences embedded within nested contexts [
14]. The ecological model of child maltreatment by Belsky [
15] emphasizes that child maltreatment reflects the interplay of influences ranging from parental functioning and family interaction patterns to broader social stressors and cultural norms. Guided by this perspective, and given the constraints of the secondary dataset, we operationalized ecological influences according to their proximity to the parent–child relationship and organized covariates into parent-level, family/household-level, and community-contextual factors. Nonetheless, the evidence base remains limited in integrating ecological, multilevel influences alongside research on risk and protective factors in the ITCM [
6]. In the present study, ITCM is used to refer to intergenerational continuity of risk reflected in the association between parental ACEs and current child maltreatment, rather than confirmed causal or subtype-specific transmission.
Therefore, this study integrates a person-oriented analytic strategy with a multilevel ecological lens to advance the understanding of ITCM in the Korean context. Specifically, this study aims (1) to identify latent classes of parents based on differential associations between parental ACEs and ITCM and (2) to examine parent-level, family/household-level, and community-contextual risk and protective factors that differentiate class membership. By linking heterogeneous ACEs–ITCM pathways to potentially modifiable ecological influences, this study seeks to inform targeted prevention and intervention efforts to disrupt the intergenerational cycles of maltreatment.
METHODS
Ethical statements: This study was a secondary analysis of existing data and was therefore exempt from the Human Ethics Committee of Yonsei University (IRB No. 4-2022-0041).
1. Study Design
This study was a secondary analysis of cross-sectional data and was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. The survey was conducted by the Korea Institute for Health and Social Affairs in 2017 and is the first national survey to examine the experiences of maltreatment and violence in the present (in the past year), which were divided into childhood and adulthood violence and abuse experiences. No follow-up national survey using comparable instruments to assess childhood adverse experiences and child maltreatment has been conducted since 2017. Although several national child-related surveys have been administered after 2017, their survey frameworks and item definitions differ substantially from the 2017 survey, making them unsuitable for direct comparison or as substitute data sources in the present study.
2. Study Setting and Sample
The survey period was from August 23 to October 13, 2017, during which data collectors visited households directly and conducted interviews using a tablet. The census output area was selected by applying the Population and Housing Census, and samples were allocated based on the adult population aged 19–59 years to construct representative data on a national scale. A stratified sampling method was employed, in which the respondents’ gender ratio was considered and proportionally distributed according to the population composition by region and age. A nationwide sample of 4,008 adults (fathers or mothers) who had children under the age of 18 at the time of the survey was obtained, and only 1 father or 1 mother in each household was allowed to participate in the survey. Only 3,945 participants with spouses (married, cohabitated, and separated partners) and children were included in this study. Considering the sensitivity of the survey questions, the participants directly responded to the tablet for questions related to child maltreatment and violence. To address potential response distortion in self-report data, a social desirability scale was included in the original survey [
16]. There was no missing data for any of the variables included in this study.
Based on Belsky’s ecological model of child maltreatment, the study variables were organized across ecological levels according to their proximity to the parent–child relationship. Specifically, covariates were operationalized into 3 domains: parent-level factors, family/household-level factors, and community-contextual factors. As macrosystem-level indicators were limited to the secondary dataset, the present analysis focused on indicators available within these 3 domains.
Supplement 1 summarizes the survey variables and instruments used to measure them.
1) Intergenerational transmission of child maltreatment
In the present study, ITCM was operationalized as intergenerational continuity of risk, reflected in the association between parental ACEs and current child maltreatment. Accordingly, the analysis was intended to identify heterogeneous patterns of intergenerational risk across latent classes rather than to test confirmed causal or subtype-specific transmission
(1) Parental adverse childhood experiences
The translated and modified Adverse Childhood Experiences International Questionnaire (ACE-IQ) was adopted for the survey [
17]. This scale contains items on experiences before the age of 18 years and comprises 30 items categorized into 13 domains. Notably, among the domains of direct harm and witnessing collective violence, only the “Experience of victims and witnesses of collective violence” domain was modified to suit the Korean context. Following the ACE-IQ scoring guidelines [
18], each item was coded as “yes (1)” if it had been experienced at least once and “no (0).” To examine the effects of specific types of ACEs based on prior research, ACE domains were summed and grouped into neglect, abuse, family dysfunction, and violence [
18]. Cronbach’s α was .80 in a previous study [
19] and .84 in this study.
(2) Child maltreatment
The Parent–Child Conflict Tactics Scale (CTS-PC) was translated and modified for this survey [
20]. The CTS-PC assesses the occurrence of maltreatment-related behaviors during the past year. The original scale comprised 32 items across non-violent discipline, psychological aggression, physical assault, and neglect. The present survey included 4 domains (14 items): psychological aggression (2 items), physical assault (7 items), neglect (4 items), and sexual abuse (1 item). A total score of 0 indicates no maltreatment, whereas a total score of 1 or higher indicates the presence of maltreatment. In the present analysis, this score is best interpreted as indexing the number of maltreatment acts/items reported on the CTS-PC rather than the frequency, chronicity, or severity of repeated maltreatment. Accordingly, the latent classes derived in this study should be understood in relation to a gradient of reported maltreatment acts rather than differences in maltreatment intensity or chronicity. At the time of development, Cronbach’s α ranged from .22 to .60 [
20]; in this study, Cronbach’s α was .57.
2) Parent-level factors
Parent-level factors included parents’ general characteristics (gender, age, educational level, and work status) and psychological/attitudinal characteristics (depression, self-esteem, and acceptance of corporal punishment).
(1) Depression
Parental depression was measured using the 11-item Center for Epidemiologic Studies–Depression scale (CESD-11), an abbreviated form of Radloff’s CES-D that assesses depressive symptoms over the past week [
21]. Score range from 0 to 33, with higher scores indicating greater depressive symptoms. Cronbach’s α was .81 for the CESD-11 compared with .86 for the CES-D [
21]; in this study, Cronbach’s α was .68.
(2) Self-esteem
Parental self-esteem was measured using Rosenberg’s Self-Esteem Scale [
22], which was translated and modified for this survey. The scale comprises 10 items rated on a 4-point Likert scale, with higher scores indicating higher self-esteem. In a study with parents, Cronbach’s α was .83 [
23]; in this study, it was .75.
(3) Acceptance of corporal punishment
Parental acceptance of corporal punishment was assessed using a single item (“What do you think about corporal punishment of children?”) rated on a 4-point Likert scale (1=never allowed to 4=always acceptable).
3) Family/household-level factors
Family/household-level factors captured characteristics of the immediate caregiving context, including household composition (number of children, child age, and gender), parent-child interaction (parenting stress), household socioeconomic conditions (family monthly income), and marital and family relationships (partner violence or control and family relationship).
(1) Parenting stress
Parenting stress was measured using a modified version of Abidin’s Parenting Stress Index–Short Form for the Family Life Experiences Survey [
24]. Five items were rated on a 4-point Likert scale and reverse-coded (3=strongly agree to 0=not at all). Scores range from 0 to 15, with higher scores indicating greater parenting stress. Cronbach’s α was .76–.80 in a previous study using the same tool and .85 in this study [
20].
(2) Partner violence or control
Partner violence/control was measured using the Conflict Tactics Scale (CTS2) [
25], translated and modified for this survey, and items assessing economic violence/control from the National Domestic Violence Survey. The CTS2 includes 10 items covering 5 types of spousal violence (negotiation, emotional, physical, injury, and sexual violence). Three additional items assessed economic violence/control. Responses were coded as “yes (1)” to indicate any violence/damage and “no (0)” to indicate none; negotiation items were reverse coded to align directionality with other subscales. Cronbach’s α at development ranged from .79 to .95 [
25]. In this study, it was .85.
(3) Family relationship
Family relationship quality was measured using the Family Relationship Scale based on the 10 items used in the 2010 National Survey of Elderly Abuse [
16]. Items were rated on a 4-point Likert scale (1=strongly agree to 4=not at all) and reverse-coded, with higher total scores indicating more positive family relationships. The Cronbach’s α was .91 in this study.
4) Community-contextual factors
Community-contextual factors included residential area and social support.
(1) Residential area
Residential areas were categorized as urban, suburban, or rural.
(2) Social support
Social support was measured using the translated and modified Lubben Social Support Scale [
26]. The scale includes 6 items assessing the perceived availability of support from family/relatives as well as from neighbors/friends. Example items include: ‘How many people do you meet or contact with at least once a month?’, “How many people are close enough for you to ask for help?”, and “How many people can you comfortably talk to privately?” The response options were scored as 0 (none), 1 (1 person), 2 (2 people), 3 (3–4 people), 4 (5–8 people), and 5 (9 or more people). Higher total scores indicate greater social support. In a Korean validation study, Cronbach’s α was .75 [
27]; in this study, it was .84.
After sequentially applying the design weight, non-response correction weight, and extreme weight adjustment, the post-stratification weights were computed. All analyses incorporated sampling weights to account for the complex survey design. Because partner violence/control variables pertained to marital relationships, the analyses were restricted to respondents with spouses.
Data analyses were conducted using IBM SPSS ver. 28.0 (IBM Corp.) for descriptive statistics, chi-square tests, and multinomial logistic regression; LatentGOLD ver. 6.0 (Statistical Innovations Europe) for latent class regression (LC regression); and STATA ver. 17.0 (Stata Corp.) for estimating and graphing marginal effects using the “margins” and “marginsplot” commands.
First, descriptive statistics were used to summarize child maltreatment, parental ACEs, and covariates across the 3 ecological domains (parent-, family/household-, and community-contextual factors). To assess potential response distortion, we examined bivariate correlations between social desirability scores and key study variables (parental ACEs and child maltreatment) (
Supplement 2). Given the small magnitude of these correlations, social desirability was not included as an additional covariate in the multinomial logistic regression models predicting class membership. As a sensitivity check, models additionally adjusting for social desirability yielded materially unchanged estimates and conclusions, supporting the robustness of the presented results.
Second, LC regression was employed to identify latent classes characterized by differential associations between parental ACEs and child maltreatment.
Third, the marginal effects of ACE domains on the predicted probabilities of child maltreatment were examined by class membership and are presented graphically.
Fourth, multinomial logistic regression was conducted to examine whether parent-, family/household-, and community-contextual factors differentially predicted class membership.
RESULTS
1. Description of Study Factors
The average age of parents was 40.8 years (standard deviation=6.23). Of the participants, 2,038 (51.7%) were mothers and 68.4% had a university degree or higher.
Supplement 3 summarizes the study factors, including parent-level, family/household-level, and community contextual factors.
The most prevalent types of parental ACEs were direct harm and witnessing collective violence (49.4%), violence in the community (49.3%), witnessing domestic violence (46.8%), and emotional abuse (46.3%). Parents who reported at least 1 ACE accounted for 77.7% of all participants, and 38.6% reported 4 or more ACEs.
Psychological aggression (38.5%) was the most prevalent form of maltreatment. Neglect and sexual abuse were reported by fewer than 2% of parents, with sexual abuse reported by only 2 parents. The most frequent child maltreatment score was zero, reported by 2,314 parents (58.5%) (
Table 1). Given the very low prevalence of neglect and sexual abuse, the latent class solution was likely shaped primarily by psychological aggression and, to a lesser extent, physical assault. Accordingly, the identified classes should be interpreted primarily as reflecting heterogeneity in the more prevalent maltreatment forms rather than stable subtype-specific differences in the less prevalent maltreatment subtypes.
1) Model selection for intergenerational transmission of child maltreatment
To determine the best-fitting solution, latent-class regression models with increasing numbers of classes were compared for up to 4 classes. The fit indices are presented in
Table 2. Although Akaike information criterion (AIC) and Bayesian information criterion (BIC) favored the 4-class solution, improvements in AIC, BIC, and Vuong–Lo–Mendell–Rubin likelihood ratio test, and increases in entropy diminished from the 3-class solution to the 4-class solution. The log-likelihood also showed substantial improvement from the 1-class to the 3-class model, followed by leveling-off for the 4-class model. In addition, the 4-class solution yielded a class comprising less than 10% of the sample, limiting interpretability.
In the 3-class solution, Class 1 comprised 58.5% of the sample, followed by Class 2 (23.3%), and Class 3 (18.2%). The entropy of the 3-class model was .99, indicating an excellent classification. Overall, the 3-class model provides the optimal and most interpretable solutions.
2) Description of the 3-class solution
We identified 3 classes of parents characterized by heterogeneous associations between parental ACEs and child maltreatment (
Table 3). The 3-class solution was retained, and the classes were labeled as the None-ACE-Association class (58.5%), the Partial-ACE-Association class (23.3%), and the Comprehensive-ACE-Association class (18.2%). These labels were chosen to provide a descriptive and methodologically neutral summary of the observed ACE-domain association patterns across classes. Class 1 was characterized by a high probability of reporting no child maltreatment acts (score=0), Class 2 by a high probability of reporting 1 child maltreatment act (score=1), and Class 3 by a high probability of reporting 2 or more child maltreatment acts (score≥2). Mean ACE-domain scores increased monotonically across classes (Class 3>Class 2>Class 1).
As shown in
Figure 1, the Comprehensive-ACE-Association class exhibited consistently positive average marginal effects across all 4 ACE domains. The Partial-ACE-Association class showed positive average marginal effects for abuse, family dysfunction, and violence, but not for neglect. By contrast, the None-ACE-Association class showed comparatively attenuated associations across ACE domains. These differences indicate variation in ACE-domain marginal-effect patterns across the 3 classes, but they are more appropriately interpreted in the context of the graded distribution of reported child maltreatment acts than as evidence of distinct intergenerational transmission mechanisms. These class differences should also be interpreted in light of the low prevalence of neglect and sexual abuse, suggesting that the 3-class solution likely reflects variation driven mainly by the more prevalent maltreatment forms.
3) Factors on class membership
Multinomial logistic regression was conducted to identify the factors associated with class membership in the relationship between parental ACEs and child maltreatment. Predictors were entered as parent-level, family/household-level, and community-contextual factors. In
Table 4, Exp (B) represents the odds ratio (OR) of membership in the comparison class (left-hand column header) relative to the reference class (right-hand column header); ORs <1 indicate lower odds and ORs >1 indicate higher odds of membership in the comparison class.
Several parent-level factors (parental age and gender, depression, acceptance of corporal punishment) and family/household-level factors (number of children, parenting stress, partner violence/control, family relationship, and the age group of the youngest child) differentiated the “None-ACE-Association” class from the “Comprehensive-ACE-Association” class. Older parents and parents with more children were less likely to be in the “None-ACE-Association” class than in the “Comprehensive-ACE-Association” class (OR, .96; p=.004; OR, .56; p<.001). Mothers were less likely than fathers to be in the “None-ACE-Association” class (OR, .66; p=.002). Higher levels of depression, parenting stress, and partner violence/control were associated with lower odds of membership in the “None-ACE-Association” class relative to the “Comprehensive-ACE-Association” class (OR, .96; p=.001; OR, .83; p<.001; OR, .77; p<.001), whereas more positive family relationship quality was associated with higher odds (OR, 1.04; p<.001). Compared with parents of infants and toddlers, parents whose youngest child was a preschooler, school-aged, or an adolescent were less likely to be the “None-ACE-Association” class than in the “Comprehensive-ACE-Association” class (OR, .44, .28, and .29, respectively; all p<.001).
Regarding community contextual factors, residential area was significant. Compared with parents living in urban areas, those living in suburban and rural areas were more likely to be in “None-ACE-Association” class than in the “Comprehensive-ACE-Association” class (OR, 2.04; p<.001; OR, 1.55; p=.020).
The pattern of significant predictors differentiating “Partial-ACE-Association” class from the “Comprehensive-ACE-Association” class was largely similar. However, employment status emerged as an additional parent-level factor and the patterns for the youngest child’s age group and residential area differed. Compared with regular employees, unemployed parents were more likely to be in the “Partial-ACE-Association” class than in the “Comprehensive-ACE-Association” class (OR, 1.54; p=.007). Relative to parents of infants and toddlers, parents whose youngest child was an adolescent were less likely to be in the “Partial-ACE-Association” class than in the “Comprehensive-ACE-Association” class (OR, .53; p=.012). Regarding community-contextual factors, parents living in suburban areas were more likely to be in the “Partial-ACE-Association” class than in the “Comprehensive-ACE-Association” class than in those living in urban areas (OR, 1.43; p=.002).
Finally, acceptance of corporal punishment was a significant parent-level factor differentiating the “None-ACE-Association” class from the “Partial-ACE-Association” class. Parents who reported corporal punishment were less likely to be in the “None-ACE-Association” class than in the “Partial-ACE-Association” class compared with those who never allowed it (OR, .25; p<.001).
Social desirability showed only small correlations with the key study variables (
Supplement 2). In sensitivity analyses additionally adjusting for social desirability, the main associations remained materially unchanged, indicating that the findings were robust to potential response distortion.
DISCUSSION
This study examines the ITCM by combining a person-oriented analytic strategy with an ecological perspective. Using latent class regression with a nationally representative Korean sample, we identified 3 distinct classes: None-ACE-Association, Partial-ACE-Association, and Comprehensive-ACE-Association, suggesting heterogeneity in the association between) ACEs and child maltreatment. This person-oriented approach suggests that subgroup-specific association patterns may be obscured in variable-oriented models that assume a single average pathway from parental ACEs to child maltreatment.
Only a limited number of studies have examined parental ACEs specifically in the context of ITCM [
6,
7,
10]. This study extends prior work by modeling heterogeneity in the associations between parental ACEs and child maltreatment using a person-oriented approach. Building on this framework, the latent class regression findings are more appropriately interpreted as reflecting a graded pattern of maltreatment involvement across classes, with differential ACE-domain associations by class, rather than clearly distinct transmission pathways or a single uniform ACE-to-maltreatment process. The class-specific marginal-effects pattern supports domain-level heterogeneity, with abuse-related ACEs emerging as a potentially salient factor distinguishing classes characterized by more extensive child maltreatment-act endorsement (
Figure 1). This interpretation is broadly consistent with prior person-centered evidence showing that ACE profiles characterized by abuse-related and more cumulative adversities are associated with elevated risk of child maltreatment compared with low-adversity profiles [
12]. However, these class-specific patterns should be interpreted with caution because neglect and sexual abuse were extremely rare in the present sample, suggesting that the observed class structure was shaped primarily by the more prevalent forms of maltreatment. Clinically, this domain-informed heterogeneity suggests the potential utility of risk stratification and tailored assessment—especially prioritizing parental histories of abuse and related violence contexts—and supports multi-level prevention and early intervention strategies to interrupt intergenerational transmission [
14,
15,
28].
Multinomial logistic regression results further clarified how ecological predictors differentiated the 3 classes—the None-ACE-Association class, the Partial-ACE-Association class, and the Comprehensive-ACE-Association class—suggesting that ITCM prevention should address modifiable risks across multiple ecological levels [
27]. Overall, many predictors operated in similar directions when contrasting the Comprehensive-ACE-Association class with the None-ACE-Association and Partial-ACE-Association class, whereas a smaller subset of factors more specifically differentiated the None-ACE-Association class from the Partial-ACE-Association class. This pattern suggests that some ecological factors may be particularly relevant to distinctions at the lower end of the observed child maltreatment act-count distribution.
At the parent level, predictors showed a more contrast-specific pattern. Unemployment emerged as a differentiator primarily when distinguishing the Partial-ACE-Association class from the Comprehensive-ACE-Association class, whereas greater acceptance of corporal punishment more clearly distinguished the None-ACE-Association class from the Partial-ACE-Association class. These findings support prevention strategies that provide culturally sensitive parenting education and promote non-violent discipline strategies as practical alternatives to physical punishment [
2]. The youngest child’s age group also differentiated class membership, largely in contrast involving the Comprehensive-ACE-Association class, suggesting that developmental timing may shape how ACE-related vulnerabilities are expressed within the family. Recent evidence indicates that caregivers’ adverse childhood experiences are associated with higher levels of adversity among their children, underscoring intergenerational vulnerability to cumulative and potentially multi-type adverse experiences [
9]. Notably, class contrasts involving the Comprehensive-ACE-Association class suggested greater vulnerability among adolescents than among infants/toddlers, a pattern that is broadly consistent with WHO guidance noting heightened vulnerability at both developmental ends, particularly among children under age 4 and adolescents [
29]. This discrepancy may reflect the influence of unmeasured child and family processes (e.g., temperament, behavioral problems, academic difficulties, or parent–adolescent conflict), underscoring the need for future studies to incorporate a broader set of child-level and family process indicators. However, because the dataset did not include child-level mechanisms needed to evaluate these possibilities, this interpretation remains speculative rather than evidentiary.
At the family/household level, in contrast involving the Comprehensive-ACE-Association class versus the None-ACE-Association and Partial-ACE-Association class, both larger family size and elevated parenting stress were associated with membership in the Comprehensive-ACE-Association class. These findings are consistent with prior reviews highlighting the role of family- and caregiver-level psychosocial risk and protective factors in intergenerational cycles of maltreatment [
11]. Having more children may amplify risk by increasing parenting demands while simultaneously reducing available resources per child [
28]. Higher parenting stress, in turn, can contribute to negative parenting behaviors and increase the likelihood of maltreatment [
8]. Taken together, these findings imply that prevention efforts should consider caregiving load when providing parenting support and include coping and stress-management components that help families maintain supportive and nurturing environments.
At the community-contextual level, residential-area differences were observed most clearly in contrasts involving the Comprehensive-ACE-Association class, but evidence on residence-based differences in ITCM remains limited and mixed [
30]. Because residential area was operationalized as a broad proxy for community context, these findings should be interpreted cautiously and should not be taken as evidence of specific community-level mechanisms. Future research should incorporate more granular community indicators (e.g., neighborhood disadvantage, service accessibility, and social resource availability) to clarify how community context may relate to class membership.
Although the present study used data collected in 2017, the findings remain informative for contemporary child health practice in Korea. Notably, Korea’s child protection and child-abuse response system has undergone major reforms, including a transition toward a more public-centered model beginning in 2020 with expanded governmental responsibility for investigation and case management [
1]. These system-level reforms may have influenced detection, reporting, and service-delivery processes; therefore, prevalence estimates should be interpreted within the 2017 policy context. Nevertheless, the primary contribution of this study is not to provide updated prevalence rates but to clarify heterogeneous, class-specific association patterns linking parental ACE domains and child maltreatment within an ecological framework. Accordingly, these findings can support current practice by informing risk stratification and class-tailored prevention and linkage-to-care strategies under the evolving public-centered child protection system. Future studies should examine whether these patterns are replicated in more recent cohorts and across changing policy and service contexts.
These findings have implications for pediatric and community health nurses in South Korea. As mandated reporters of child maltreatment, nurses working in primary care settings, pediatric clinics, and community health centers are well positioned to identify families at risk for ITCM and provide early intervention. Identifying distinct class profiles may support tailored screening protocols and prevention strategies aligned with family-specific risk patterns. In practice, nurses should consider routinely assessing parents’ histories of adversity and current parenting stress, particularly in families with multiple children, and evaluating parental mental health as a potentially modifiable risk factor. In addition, culturally sensitive parenting guidance that emphasizes positive, non-violent discipline may help shift attitudes toward corporal punishment [
2]. Connecting families with more extensive risk associations to mental health services, parenting support programs, and accessible community resources may be especially important when risk accumulates across parent-, family-, household-, and community-contextual domains.
This study had several limitations. First, the CTS-PC maltreatment subtypes were summarized primarily at the overall-sample level, and several subtypes were rare; most notably, neglect was reported by fewer than 2% of parents and sexual abuse by only 2 parents. As a result, stable class-stratified subtype comparisons were not feasible, and the latent class solution was likely shaped primarily by psychological aggression and, to a lesser extent, physical assault. Accordingly, the identified classes should be interpreted cautiously and may not fully capture heterogeneity in the less prevalent maltreatment subtypes. Second, although this study was guided by Belsky’s ecological model of child maltreatment, the use of secondary data constrained the range of ecological indicators that could be operationalized; in particular, only a limited set of community-contextual variables was available. Third, because macrosystem-level factors (e.g., policy contexts and broader cultural indicators) could not be comprehensively measured in the dataset, conclusions regarding macrosystem influences on ITCM should be interpreted cautiously. Future studies should incorporate explicit policy/cultural measures or link individual-level data to community- and policy-level indicators to test multilevel pathways more fully. Finally, although the CTS-PC is one of the most widely used measures in child maltreatment research, the modified version used in this sample showed modest internal consistency (Cronbach’s α=.57). Moreover, because child maltreatment and ACEs are socially sensitive topics, underreporting may have introduced additional measurement error. Importantly, because child maltreatment indicators were used to define the latent classes, these measurement limitations may have affected not only effect estimates but also the latent class solution itself. Specifically, imperfect measurement may have contributed to misclassification, obscured smaller or more distinct subgroups, and attenuated contrasts across classes and their associations with ecological predictors. In addition, the relatively low internal consistency may partly reflect the heterogeneous nature of child maltreatment behaviors, the low base rates of some maltreatment subtypes, and restricted variability due to underreporting in self-reports. Therefore, the findings should be interpreted cautiously as potentially conservative estimates. Future research would benefit from using more behavior-specific or subtype-specific measures, incorporating multiple informants or administrative data, and applying longitudinal designs to improve measurement precision and strengthen inferences about intergenerational continuity of risk.
CONCLUSION
Using latent class regression, this study identified heterogeneous patterns in the association between parental ACEs and child maltreatment, distinguishing between the None-ACE-Association, Partial-ACE-Association, and Comprehensive-ACE-Association. Class membership was differentiated by parent-, family/household-, and community-contextual factors, indicating that intergenerational risk is shaped by influences across multiple ecological domains. By linking subgroup-specific ACEs–child maltreatment association patterns to modifiable ecological correlates (e.g., parental depression, parenting stress, partner violence/control, and attitudes toward corporal punishment), these findings can inform screening efforts and support the development of tailored prevention and intervention strategies aimed at disrupting intergenerational cycles of maltreatment.
ARTICLE INFORMATION
Supplementary material
Supplement 3.
Description of factors related to parent-level factors, family/household-level factors, and community-contextual factors.
Figure 1.Average marginal effects of adverse childhood experience (ACE) domains on the predicted probability of child maltreatment across the 3-class solution.
Table 1.Description of parental adverse childhood experiences and child maltreatment
|
Domain |
No. (%) |
Mean±SD |
Possible score range (min, max) |
|
Adverse childhood experiences |
|
|
|
|
Neglect |
|
0.23±0.48 |
0–2 (0, 2) |
|
Neglect of emotion |
654 (16.6) |
|
|
|
Neglect of body |
271 (6.9) |
|
|
|
Abuse |
|
0.73±0.85 |
0–3 (0, 3) |
|
Emotional abuse |
1,828 (46.3) |
|
|
|
Physical abuse |
929 (23.5) |
|
|
|
Sexual abuse |
112 (2.8) |
|
|
|
Family dysfunction |
|
0.59±0.69 |
0–5 (0, 5) |
|
Alcohol and drug use by family members |
186 (4.7) |
|
|
|
Chronic depression of family members |
49 (1.2) |
|
|
|
Imprisonment of family members |
7 (0.2) |
|
|
|
Separation and death of parents |
230 (5.8) |
|
|
|
Witnessing domestic violence |
1,846 (46.8) |
|
|
|
Violence |
|
1.29±1.08 |
0–3 (0, 3) |
|
Peer violence |
1,177 (29.8) |
|
|
|
Violence in the community |
1,944 (49.3) |
|
|
|
Direct harm and witnessing collective violence |
1,950 (49.4) |
|
|
|
Adverse childhood experience score |
|
2.83±2.37 |
0–13 (0, 11) |
|
0 |
878 (22.3) |
|
|
|
1 |
613 (15.5) |
|
|
|
2 |
450 (11.4) |
|
|
|
3 |
480 (12.2) |
|
|
|
4+ |
1,524 (38.6) |
|
|
|
Child maltreatment |
|
|
|
|
Psychological aggression |
1,547 (38.5) |
|
|
|
Physical assault |
397 (10.1) |
|
|
|
Neglect |
78 (1.9) |
|
|
|
Sexual abuse |
2 (0.0) |
|
|
|
Child maltreatment score |
|
0.71±1.05 |
0–14 (0, 14) |
|
0 |
2,314 (58.5) |
|
|
|
1 |
928 (23.4) |
|
|
|
2+ |
716 (18.1) |
|
|
Table 2.Fit Indices for latent class regression
|
Criterion |
Class 1 |
Class 2 |
Class 3 |
Class 4 |
|
Log-likelihood |
–5,436.52 |
2,640.14 |
6,043.81 |
7,664.34 |
|
No. of parameters |
6 |
13 |
20 |
27 |
|
AIC |
10,885.04 |
–5,254.28 |
–12,047.63 |
–15,274.68 |
|
BIC |
10,922.74 |
–5,172.60 |
–11,921.96 |
–15,105.03 |
|
Entropy |
1.00 |
.98 |
.99 |
1.00 |
|
VLMR |
|
16,153.32 (<.001) |
6,807.35 (<.001) |
3,241.05 (<.001) |
|
Class 1 (%) |
100 |
58.2 |
58.5 |
58.5 |
|
Class 2 (%) |
|
41.8 |
23.3 |
23.4 |
|
Class 3 (%) |
|
|
18.2 |
10.3 |
|
Class 4 (%) |
|
|
|
7.8 |
|
Class 5 (%) |
|
|
|
|
Table 3.Child maltreatment and adverse childhood experiences for the 3-class model
|
Class 1 (58.5%) |
Class 2 (23.3%) |
Class 3 (18.2%) |
F |
Schefffe |
|
None-ACE-Association class |
Partial-ACE-Association class |
Comprehensive-ACE-Association class |
|
Child maltreatment |
0±0.00 |
1±0.00 |
2.59±0.91 |
12,780.98*** |
3>2>1 |
|
Adverse childhood experiences |
2.24±2.19 |
3.19±2.08 |
4.75±2.23 |
372.67*** |
|
|
Neglect |
0.17±0.41 |
0.23±0.46 |
0.40±0.61 |
71.08*** |
|
|
Abuse |
0.52±0.77 |
0.88±0.85 |
1.43±0.88 |
357.17*** |
|
|
Family dysfunction |
0.49±0.70 |
0.65±0.61 |
1.02±0.76 |
162.17*** |
|
|
Violence |
1.06±1.02 |
1.42±0.96 |
1.89±0.96 |
198.10*** |
|
Table 4.Multinomial logistic regression of parent-level factors, family/household-level factors, and community-contextual factors on class membership
|
None-ACE-Association vs. Comprehensive-ACE-Association (ref) |
Partial-ACE-Association vs. Comprehensive-ACE-Association (ref) |
None-ACE-Association vs. Partial-ACE-Association (ref) |
|
B |
SE |
p
|
Exp (B) |
B |
SE |
p
|
Exp (B) |
B |
SE |
p
|
Exp (B) |
|
Constant |
6.99 |
915.82 |
<.001 |
|
2.18 |
7.84 |
.014 |
|
4.81 |
0.65 |
<.001 |
|
|
Parent-level factors |
|
|
|
|
|
|
|
|
|
|
|
|
|
Parental age |
–0.04 |
0.01 |
.004 |
0.96 |
–0.04 |
0.01 |
.013 |
0.96 |
–0.003 |
0.01 |
.731 |
1.00 |
|
Gender (ref: male) |
–0.41 |
0.09 |
.002 |
0.66 |
–0.36 |
0.10 |
.010 |
0.70 |
–0.05 |
0.11 |
.660 |
0.95 |
|
Educational level (ref: high school graduate & under) |
0.15 |
0.14 |
.216 |
1.16 |
–0.08 |
0.11 |
.523 |
0.93 |
0.22 |
0.10 |
.021 |
1.25 |
|
Work status (ref: regular employee) |
|
|
|
|
|
|
|
|
|
|
|
|
|
Temporary employee |
–0.29 |
0.17 |
.199 |
0.75 |
–0.18 |
0.20 |
.465 |
0.84 |
–0.12 |
0.21 |
.580 |
0.89 |
|
Own business |
–0.21 |
0.12 |
.144 |
0.81 |
–0.02 |
0.15 |
.893 |
0.98 |
–0.19 |
0.12 |
.115 |
0.83 |
|
Unemployed |
–0.04 |
0.15 |
.817 |
0.97 |
0.43 |
0.25 |
.007 |
1.54 |
–0.47 |
0.13 |
<.001 |
0.63 |
|
Depression |
–0.04 |
0.01 |
.001 |
0.96 |
–0.05 |
0.01 |
<.001 |
0.95 |
0.01 |
0.01 |
.580 |
1.01 |
|
Self-esteem |
–0.02 |
0.02 |
.168 |
0.98 |
0.00 |
0.02 |
.839 |
1.00 |
–0.03 |
0.01 |
.568 |
0.98 |
|
Acceptance of corporal punishment (ref: never allowed) |
–1.67 |
0.03 |
<.001 |
0.19 |
–0.29 |
0.11 |
.052 |
0.75 |
–1.38 |
0.10 |
<.001 |
0.25 |
|
Family/household-level factors |
|
|
|
|
|
|
|
|
|
|
|
|
|
No. of children |
–0.58 |
0.05 |
<.001 |
0.56 |
–0.21 |
0.08 |
.025 |
0.81 |
–0.97 |
0.07 |
<.001 |
0.69 |
|
Gender of youngest child (ref: male) |
0.20 |
0.13 |
.053 |
1.22 |
0.21 |
0.13 |
.058 |
1.23 |
–0.004 |
0.08 |
.958 |
1.00 |
|
Age group of youngest child (ref: infant + toddler) |
|
|
|
|
|
|
|
|
|
|
|
|
|
Pre-schooler |
–0.83 |
0.09 |
<.001 |
0.44 |
–0.17 |
0.18 |
.416 |
0.84 |
–0.66 |
0.13 |
<.001 |
0.52 |
|
School age |
–1.26 |
0.06 |
<.001 |
0.28 |
–0.42 |
0.14 |
.055 |
0.66 |
–0.85 |
0.15 |
<.001 |
0.43 |
|
Adolescents |
–1.23 |
0.07 |
<.001 |
0.29 |
–0.64 |
0.13 |
.012 |
0.53 |
–0.59 |
0.18 |
.001 |
0.55 |
|
Parenting stress |
–0.18 |
0.02 |
<.001 |
0.83 |
–0.07 |
0.02 |
<.001 |
0.93 |
–0.12 |
0.01 |
<.001 |
0.89 |
|
Family monthly income |
0.00 |
0.00 |
.069 |
1.00 |
0.00 |
0.00 |
.264 |
1.00 |
0.00 |
0.00 |
.378 |
1.00 |
|
Partner violence or control |
–0.26 |
0.03 |
<.001 |
0.77 |
–0.15 |
0.03 |
<.001 |
0.86 |
–0.11 |
0.03 |
<.001 |
0.90 |
|
Family relationship |
0.04 |
0.01 |
.001 |
1.04 |
0.06 |
0.01 |
<.001 |
1.06 |
–0.01 |
0.01 |
.268 |
0.99 |
|
Community-contextual factors |
|
|
|
|
|
|
|
|
|
|
|
|
|
Residential area (ref: urban) |
|
|
|
|
|
|
|
|
|
|
|
|
|
Suburban |
0.71 |
0.23 |
<.001 |
2.04 |
0.36 |
0.17 |
.002 |
1.43 |
0.36 |
0.09 |
<.001 |
1.43 |
|
Rural |
0.44 |
0.30 |
.020 |
1.55 |
–0.41 |
0.14 |
.053 |
0.66 |
0.86 |
0.18 |
<.001 |
2.35 |
|
Social support |
–0.02 |
0.01 |
.076 |
0.98 |
–0.01 |
0.01 |
.353 |
0.99 |
–0.01 |
0.01 |
.314 |
0.99 |
REFERENCES
- 1. Shim WS. New public-centered child protection system in Korea. Clin Exp Pediatr. 2023;66(5):179-181. https://doi.org/10.3345/cep.2023.00150
- 2. Ma J, Han Y, Kang HR. Physical punishment, physical abuse, and child behavior problems in South Korea. Child Abuse Negl. 2022;123:105385. https://doi.org/10.1016/j.chiabu.2021.105385
- 3. Ministry of Health and Welfare. Child Abuse & Neglect Korea 2023. Ministry of Health and Welfare; 2024.
- 4. Lee JY, Ryu JH. The relationship between latent classes of parent’s adverse childhood experiences and child maltreatment. J Korean Counc Child Rights. 2021;25(1):107-130. https://doi.org/10.21459/kccr.2021.25.1.107
- 5. Racine N, Deneault AA, Thiemann R, Turgeon J, Zhu J, Cooke J, et al. Intergenerational transmission of parent adverse childhood experiences to child outcomes: a systematic review and meta-analysis. Child Abuse Negl. 2025;168(Pt 1):106479. https://doi.org/10.1016/j.chiabu.2023.106479
- 6. Berzenski SR, Yates TM. Revisiting a multidimensional model of intergenerational transmission of child maltreatment. In: Krugman RD, Korbin JE, editors. Handbook of child maltreatment. 2nd ed. Springer; 2022. p. 351-370.
- 7. Langevin R, Abou Chabake S, Beaudette S. Intergenerational cycles of maltreatment: an updated scoping review of psychosocial risk and protective factors. Trauma Violence Abuse. 2026;27(2):490-505. https://doi.org/10.1177/15248380251316908
- 8. Marshall C, Langevin R, Cabecinha-Alati S. Victim-to-victim intergenerational cycles of child maltreatment: a systematic scoping review of theoretical frameworks. Int J Child Adolesc Resil. 2022;9(1):95-115. https://doi.org/10.54488/ijcar.2022.283
- 9. Madigan S, Turgeon J, Racine N, Zhu J, Alink LR, Ereyi-Osas W, et al. Patterns of continuity and discontinuity of childhood maltreatment across generations: a meta-analysis. Dev Psychopathol. 2025 May 19 [Epub]. https://doi.org/10.1017/S0954579425000239
- 10. Russotti J, Warmingham J, Swerbenski H, Handley ED, Li Z, Cicchetti D. Intergenerational continuity of multidimensional patterns of child maltreatment exposure: a person-centered approach. Dev Psychopathol. 2025 Jul 3 [Epub]. https://doi.org/10.1017/S0954579425100217
- 11. Knisely MR, Draucker CB. Using a person-oriented approach in nursing research. West J Nurs Res. 2016;38(4):508-520. https://doi.org/10.1177/0193945915602856
- 12. Buffarini R, Hammerton G, Coll CV, Cruz S, da Silveira MF, Murray J. Maternal adverse childhood experiences (ACEs) and their associations with intimate partner violence and child maltreatment: Results from a Brazilian birth cohort. Prev Med. 2022;155:106928. https://doi.org/10.1016/j.ypmed.2021.106928
- 13. Rivera PM, Fincham F. Forgiveness as a mediator of the intergenerational transmission of violence. J Interpers Violence. 2015;30(6):895-910. https://doi.org/10.1177/0886260514539765
- 14. Swenson CC, Schaeffer CM. A multisystemic approach to the prevention and treatment of child abuse and neglect. Int J Child Maltreat. 2018;1(1):97-120. https://doi.org/10.1007/s42448-018-0002-2
- 15. Belsky J. Child maltreatment: an ecological integration. Am Psychol. 1980;35(4):320-335. https://doi.org/10.1037//0003-066x.35.4.320
- 16. Ryu JH, Lee SJ, Jeon JA, Park SK, Yeo YJ, Lee JY, et al. Comprehensive Survey on Korean Children. Korea Institute for Health and Social Affairs; 2019.
- 17. World Health Organization. Adverse Childhood Experiences International Questionnaire (ACE-IQ) [Internet]. World Health Organization; 2018 [cited 2022 Dec 15]. Available from: https://www.who.int/publications/m/item/adverse-childhood-experiences-international-questionnaire-(ace-iq)
- 18. Chang X, Jiang X, Mkandarwire T, Shen M. Associations between adverse childhood experiences and health outcomes in adults aged 18-59 years. PLoS One. 2019;14(2):e0211850. https://doi.org/10.1371/journal.pone.0211850
- 19. Lee J, Choi E. Effects of parental experiences of childhood abuse and parenting stress on offspring abuse and depression: the buffering role of children’s social support. Health Soc Welf Rev. 2022;42(1):125-145. https://doi.org/10.15709/hswr.2022.42.1.125
- 20. Straus MA, Hamby SL, Finkelhor D, Moore DW, Runyan D. Identification of child maltreatment with the Parent-Child Conflict Tactics Scales: development and psychometric data for a national sample of American parents. Child Abuse Negl. 1998;22(4):249-270. https://doi.org/10.1016/s0145-2134(97)00174-9
- 21. Kohout FJ, Berkman LF, Evans DA, Cornoni-Huntley J. Two shorter forms of the CES-D (Center for Epidemiological Studies Depression) depression symptoms index. J Aging Health. 1993;5(2):179-193. https://doi.org/10.1177/089826439300500202
- 22. Rosenberg M. Conceiving the self. Basic Books; 1979.
- 23. Kong J. Effect of caring for an abusive parent on mental health: the mediating role of self-esteem. Gerontologist. 2018;58(3):456-466. https://doi.org/10.1093/geront/gnx053
- 24. Abidin RR. Parenting Stress Index-Short Form. Pediatric Psychology Press; 1990. p 118.
- 25. Straus MA, Hamby SL, Boney-McCoy SU, Sugarman DB. The revised Conflict Tactics Scales (CTS2) development and preliminary psychometric data. J Fam Issues. 1996;17(3):283-316. https://doi.org/10.1177/019251396017003001
- 26. Lubben J, Blozik E, Gillmann G, Iliffe S, von Renteln Kruse W, Beck JC, et al. Performance of an abbreviated version of the Lubben Social Network Scale among three European community-dwelling older adult populations. Gerontologist. 2006;46(4):503-513. https://doi.org/10.1093/geront/46.4.503
- 27. Lee KW, Kim SY, Chung W, Hwang GS, Hwang YW, Hwang IH. The validity and reliability of Korean version of Lubben Social Network Scale. Korean J Fam Med. 2009;30(5):352-358. https://doi.org/10.4082/kjfm.2009.30.5.352
- 28. Zhou H, Wang X, Li M. Parental stress, coercive and encouraging parenting among Chinese one-child, two-child and three-child families. Fam Relatsh Soc. 2024;13(4):528-548. https://doi.org/10.1332/20467435Y2024D000000018
- 29. World Health Organization. Child maltreatment [Internet]. World Health Organization; 2024 [cited 2026 Jan 22]. Available from: https://www.who.int/news-room/fact-sheets/detail/child-maltreatment
- 30. Beatriz ED, Salhi C, Griffith JL, Molnar BE. Urbanicity matters in self-reported child maltreatment prevalence: Findings from a nationally representative study. Child Abuse Negl. 2018;79:371-383. https://doi.org/10.1016/j.chiabu.2018.02.028
