Abstract
Child maltreatment and child welfare system contact are both associated with an elevated risk of adverse outcomes in childhood, adolescence, and adulthood. Yet, data on variation in system contact are available for only a handful of countries, limiting knowledge about the societal correlates of system contact. As reported in this research note, we identified, collected, and harmonized administrative data on child welfare agency investigations, confirmed maltreatment, and placements into out-of-home care for 44 countries in the Global North. We analyzed 15 sociodemographic factors commonly associated with child maltreatment and child welfare system contact. Results support three core conclusions. First, data are much more available on late-stage system contact (e.g., foster care caseloads) than for early-stage system contact (e.g., investigations). Second, whereas early-stage contact tended to be on the rise in most countries, late-stage contact was stable or declining. Cross-national variation in these trends was generally less substantial than cross-national variation in levels of child welfare system contact, indicating relatively stable cross-national differences. Third, cross-national variation in out-of-home care largely reflected, but was not reducible to, regional and sociocultural variation: we find little evidence for universal drivers of foster care caseloads across the Global North.
Introduction
Child maltreatment and child welfare system (CWS) or child protective services (CPS) involvement1 account for significant health and economic burdens in the Global North (Conti et al. 2021; Doyle and Aizer 2018; Fang et al. 2012; Fang et al. 2015; Gilbert et al. 2009; Habetha et al. 2012). Compared with those who have not experienced maltreatment or been exposed to CWS/CPS, children who have are often substantially more likely to have poor health and well-being across the life course, including elevated risks of mortality, poor educational outcomes, teenage childbirth, contact with the criminal justice system, and poor physical and mental health outcomes (Berger and Waldfogel 2004; Currie and Tekin 2012; Currie and Widom 2010; Doyle 2007, 2008; Fallesen 2021; Gruhn and Compas 2020; Norman et al. 2012).
Limited existing evidence indicates that the effects of exposure to maltreatment and to CWS/CPS are relatively consistent across national and subnational contexts. But dramatically less is known about how much exposure to maltreatment and system contact varies across national and historical contexts (Degli Esposti et al. 2019; Gilbert et al. 2012; Roehrkasse 2021; Stoltenborgh et al. 2015). This lack of evidence, driven partially by contextual differences in the definition of child maltreatment and partially by differences in how CWS/CPS responds to allegations of maltreatment, is pressing (Jud et al. 2013). First, because maltreatment and CWS/CPS exposure are important risk factors for children's poor health and behavioral outcomes, our lack of knowledge about their comparative prevalence is a core gap in comparative research on population well-being. Second, because CWS/CPS is a key institution through which societies exercise both social support and social control, a lack of comparative knowledge about system contact obscures potentially important differences in how modern welfare states approach family dynamics and population health. Indeed, CWS/CPS contact may have important implications for how welfare states vary in their manifestations of welfare capitalism and in their deployment of the “hands of the state” (Esping-Andersen 1990; Morgan and Orloff 2017).
In this research note, we extend work by Gilbert and colleagues (Gilbert et al. 2012) by providing the most extensive information to date on how rates of contact with CWS/CPS—specifically, maltreatment investigation, confirmed maltreatment, and out-of-home placement—vary across countries in the Global North. We focus on these three levels of CWS/CPS contact because each of these three CWS/CPS contact stages is a consequential intervention into family life. Instead of attempting to measure actual maltreatment, we emphasize that these measures reflect varying levels and types of CWS/CPS contact across nations: the relationship between system contact and underlying child maltreatment depends on differing definitions of maltreatment, differential responses to maltreatment, and potential differences in measurement error. In comparatively analyzing CWS/CPS contact on a novel scale, we expand our knowledge about how common state interventions in response to child maltreatment are and, as a result, how significant a role each might play in shaping population health and disparities therein. At the same time, we introduce the new publicly available ROCKWOOL–Duke Global Child Welfare Database (RDGCWD) as a future resource for the research community (Roehrkasse et al. 2022; Roehrkasse et al. 2023).2
Outcomes of Interest, Data Collection, and Methods
Outcomes of Interest
Our outcomes were three types of contact with the child welfare system or child protective services, measured by country and year. First, investigation was defined as an official inquiry by CWS/CPS personnel resulting from an allegation of maltreatment. Investigations were counted multiple times if investigations involved multiple children. Countries varied in whether they reported investigations started or completed in a 12-month period. Second, confirmed maltreatment was defined as an investigation resulting in the substantiation of alleged maltreatment. In the United States, indications of maltreatment were treated as confirmed maltreatment. Confirmed maltreatment in Canada and Germany included the substantiation of risk for future maltreatment.
Third, out-of-home care was defined as living without parents under the auspices of a CWS/CPS agency, including foster care, kinship care, and various institutional care settings. We excluded children residing in criminal justice institutions. However, because child behavioral issues cannot be easily separated from child maltreatment, we included children in out-of-home care administered by CWS/CPS, even if the primary reason for placement was child behavioral issues. Excluding children residing in criminal justice institutions means that estimates are likely downward biased in countries with higher levels of punitiveness toward children, such as the United States, particularly in the early 2000s. The number of minor children in U.S. prisons declined from 3,892 in 2000 to 653 in 2019 (Carson 2020); over the same period, the number of children in institutionalized group quarters decreased from 158,118 to 130,166 (Manson et al. 2023). Thus, the decline in children being held in prisons accounts for only a small share (11.6%) of the total decline in children being held in all types of institutionalized group quarters.
For investigations and confirmed maltreatment, we calculated the annual incidence (investigations or investigations confirming maltreatment per 12-month period per 1,000 children) and the annual prevalence (distinct children experiencing investigation or confirmed maltreatment per 12-month period per 1,000 children). Given the limited availability of prevalence data, the main analysis focuses on incidence, but our findings are consistent with an analysis of prevalence (Figures S1 and S2; all figures and tables designated with an “S” appear in the online appendix). We calculated the annual incidence and prevalence of placement into out-of-home care (entrances into care per year per 1,000 children and distinct children entering care per year per 1,000 children). Because of the limited availability of incidence data, the main analysis focuses on the prevalence of entrance into out-of-home care, but we also report calculations of the annual incidence of entrance into out-of-home care (Figure S3). Finally, we measured the point prevalence of children residing in out-of-home care (per 1,000 children), which is our most widely available measure. We utilize this measure for a more in-depth analysis of cross-national variation.
Data Collection
Data collection occurred between August 2021 and May 2022. We sought data from 64 countries in the Global North for 2000–2020 (Table S1). For simplicity, we refer to Greenland and the constituent countries of the United Kingdom (England, Northern Ireland, Scotland, and Wales) as countries. Our principal data sources were tabulated data files and statistical reports published by national statistical agencies and national child welfare agencies, accessed through agency websites or national databases (Table S2). If data were not publicly available, we accessed data directly from relevant agencies. If no data could be accessed through national agencies, we used data from UNICEF's TransMonEE database and the Nordic Health and Welfare Statistics database. Where multiple data sources were available, we used them to cross-validate; when sources of data were discrepant, we determined the correct figure through additional research or removed data when a reliable figure could not be established. Agency personnel and subject-area experts provided support in accessing, translating, and interpreting data.
We obtained data from 44 countries, including 104 country-year observations of CWS/CPS investigation rates, 194 country-year observations of CWS/CPS maltreatment confirmation rates, 233 country-year observations of children's out-of-home care entry rates, and 402 country-year observations of children's out-of-home care residence rates.3Figure 1 presents information on data availability across country-years and our measures. The figure indicates substantial variation in data availability across countries and measures, with much more data on late-stage contact than on early-stage contact.
We did not impute missing values because doing so would involve historical extrapolation, and multiple imputations would complicate the use of RDGCWD. Documentation for RDGCWD details the implications of missing data for potential data users (Roehrkasse et al. 2022).
Methods
Our analysis proceeded in two stages. In the first stage, we described (1) the country-years for which data are available on each outcome and (2) differences between countries in levels and trends of CWS/CPS contact.
In the second stage, we used three methods to explore the classification of countries in the Global North according to their prevalence of CWS/CPS contact. We focused this part of our analysis on the outcome for which data were more widely available: the number of children in out-of-home care per 1,000 children in the population. First, we grouped countries into geographic and sociocultural groups using United Nations geographic regions and a group for Anglophone countries of the Global North. Second, we used hierarchical clustering to identify groups on the basis of the similarity between each country's trends of rates over time, assessed via dynamic time warping. Third, we used the same methods to cluster countries according to the similarity of the pairwise correlations between out-of-home rates and key socioeconomic, demographic, and population health indicators (see Table S3). Note that the hierarchical clustering analyses omitted countries with five or fewer (of 15 total) covariates or two or fewer years of data.
Results
CPS/CWS Contact in the Global North: Levels and Trends
Maltreatment Investigation
Reliable data on CWS/CPS investigations for alleged maltreatment were available across 12 countries. Incidence rates varied widely between countries. However, within most countries, the levels of investigation increased over the period (Figure 2, panel a; see Table S4 for point estimates). In recent years, the United States had the highest investigation rates, peaking at 58.37 investigations per 1,000 children in 2017. England had comparably high rates. In contrast, Singapore had extremely low rates, ranging from 0.16 to 1.51 investigations per 1,000 children between 2002 and 2020.
Confirmed Maltreatment
Data on incidents of maltreatment confirmed by CWS/CPS were available for 19 countries (Figure 2, panel b; also see Table S5). Cross-national variation was considerably lower for confirmed maltreatment rates than for investigation rates (panel a), with most countries having confirmed maltreatment rates between 2.5 and 6.0 per 1,000. New Zealand displayed the highest confirmed maltreatment rates, peaking at more than 20 per 1,000 in 2013. Canada also exhibited similarly high rates but had only two data points, both occurring in the 2000s. Australia, the United States, and Wales had consistently higher rates than all other countries throughout the years. Countries such as Bulgaria and Tajikistan exhibited extremely low rates of confirmed maltreatment, ranging from 0.05 to 0.57. South Korea had a similarly low confirmed maltreatment rate until it began increasing around 2012, converging with most other countries in recent years.
Overall, rates of confirmed maltreatment had steady, meaningful increases over the twenty-first century with a few exceptions. In 2013–2015, New Zealand experienced a sharp decline in its confirmed maltreatment rate that coincided with the decline in its investigation rate. Australia experienced a similar dramatic drop in confirmed maltreatment in 2018 that also coincided with a drop in the investigation rate, although both rates returned to prior levels in 2019–2020. Confirmed maltreatment rates also decreased in the United States between 2005 and 2010 but remained stable thereafter.
Out-of-Home Care
Of the three points of CWS/CPS contact analyzed in this study, reliable data on out-of-home care were most widely available. Across 20 countries, annual rates of children entering out-of-home care varied significantly (Figure 2, panel c; see Table S6), although nearly half of the countries observed in 2020 had out-of-home care entry rates between 2 and 3 per 1,000 children. In recent years, Sweden had the highest out-of-home care entry rate, at 4.05 per 1,000 children in 2020.4 Albania had the lowest recorded out-of-home care entry rate, at 0.22 per 1,000 children in 2020.
Most countries exhibited stable out-of-home care entry rates. The exceptions were Sweden and Germany, which experienced dramatic peaks in entrances in the mid-2010s. The limited data on the migrant status of children entering care suggest that a substantial proportion of these increases were attributable to large, temporary increases in the number of unaccompanied children who entered these countries and were placed into care (Figures S3 and S5). Russia, Latvia, Lithuania, and Estonia experienced significant declines from high care entry rates, although the timing of those declines varied across these countries. A larger number of countries experienced more modest declines in entry rates between 2015 and 2020.
Data on children residing in out-of-home care on a specific date was the most widely available measure, with data on 37 countries (panel d of Figure 2, and Table S7). Although data were available for only two years, children's out-of-home care residence rates in Greenland were grossly outlying,5 at 67.88 per 1,000 children in 2006 and 53.91 per 1,000 children in 2019. In these years, out-of-home care rates in Greenland were more than 2.5 and 3.3 times greater, respectively, than the next highest (Russia and Latvia, respectively). Most countries observed in 2020 had care rates between 5 and 10 per 1,000 children. After Greenland, Russia exhibited the highest out-of-home care rates, peaking at 28.42 in 2008; however, data for Russia were unavailable after 2015. Comparing extrema in 2020, Latvia had the highest out-of-home care rates (16.45 per 1,000 children), and Singapore had the lowest rate (1.27 per 1,000 children)—a nearly 13-fold difference. Similarly to trends in out-of-home care entry (Figure 2, panel c), the point prevalence of out-of-home care residence in many countries in recent years either ceased to grow or experienced modest declines.
Worlds of Child Welfare? Clustering Countries According to Out-of-Home Care
For countries with available data, we used data on children residing in out-of-home care for additional analyses to understand more broadly the patterns and predictors of CWS/CPS contact over time.
Region and Language
Cross-national variation in out-of-home care rates partly reflected variation in care rates across geographic and cultural regions. Observed countries in Asia (Armenia, Azerbaijan, Georgia, Kazakhstan, Singapore, South Korea, and Uzbekistan) exhibited many of the lowest shares of children in care (Figure 3). Although geographically diverse, Anglophone countries (Australia, England, Ireland, Northern Ireland, the United States, Scotland, and Wales) displayed very similar levels and trends in out-of-home care.
Children's out-of-home care rates varied more widely in Europe. Eastern European countries (Belarus, Czech Republic, Hungary, Poland, Moldova, Russia, and Slovakia) generally had much higher rates than Southern European countries (Albania, Spain, and Montenegro). The share of children in out-of-home care also varied widely among observed Northern European countries: relative to other Northern European countries, Latvia and Lithuania had among the highest rates of residence in care, and Estonia, Iceland, and Sweden had consistently lower rates. Classification by region and language therefore provides a suggestive but not definitive classification of countries in the Global North in terms of CWS/CPS contact.
Time Series
Utilizing hierarchical clustering with dynamic time warping, we grouped countries by the measured similarity between each country's time series (i.e., countries with similar patterns of rates of children in care over time).6 We focused on three emergent clusters in which the first cluster (in green; Figure 4) mostly includes countries where the rate appeared to peak and then generally decreased in recent years: most visibly, Russia, Latvia, Lithuania, Norway, Armenia, and Belarus (Figure 3). The second cluster (in blue; Figure 4) includes countries that appeared to have stable rates in the early 2000s but steadily increasing rates in the last few years: most visibly Hungary, Northern Ireland, and Wales (Figure 3). The last cluster (in red; Figure 4) is similar to the second but includes countries with a more pronounced increase in rates, especially from 2000 to 2015, and continually increasing rates in recent years: most visibly, the Czech Republic, Finland, France, and Scotland (Figure 3). Time series clustering, while illuminating similarities in trends not evident in the comparison of levels, invited further investigations into why geographically and linguistically dissimilar countries cluster closely.
External Predictors of CWS/CPS Contact
We clustered countries with sufficient data (see the Methods section) according to the similarity of the pairwise correlations between various measures and the rate of children in care, identifying two emergent clusters. Cluster 1 (Figure 5, panel a) includes countries that have highly negative correlations with the proportion of the population aged 55 or older and strong positive correlations with the proportions of younger populations (0–19, 20–39, and 40–54). These countries include Norway, the United States, Denmark, Lithuania, Estonia, and South Korea. By contrast, Cluster 2 (Figure 5, panel a) countries have highly positive correlations with the proportion of the population aged 55 or older and strong negative correlations with the proportions of younger populations. These countries include Finland, Northern Ireland, France, Sweden, and Singapore.
Regarding the relationship between children's out-of-home care rates and population health and socioeconomic indicators, Cluster 1 includes countries with largely strong positive correlations with the infant mortality rate and the pregnancy rate and strong negative correlations with the immigration rate. These countries include Ireland, Poland, and Latvia. Cluster 2 displays the opposite patterns. Countries such as Finland, Spain, and England have strong negative correlations with infant mortality and teen pregnancy rates and strong positive correlations with immigration.
Less consistency is evident in correlations with the percentage living in poverty, the Gini index, and measures of substance abuse disorders: these correlations vary from very negative to very positive. As the dendrogram in Figure 5 (panel b) reveals, the clusters could be split into smaller subgroups (e.g., branches pruned at lower heights), which would create more internal consistency. The total correlation (i.e., the average within-country correlation; panel a) demonstrates that the association of demographic and socioeconomic measures with the rate of children in care varies greatly across countries, with most averaged correlations ranging from .1 to –.1. The total correlations are most positive for measures of entering care (total average correlation of .54) and the maltreatment rate (.33), the proportion of the population living in poverty (.29), and the proportion of the population with a substance abuse disorder (.27). The strongest negative correlation is with the proportion of the population aged 0–19 (–.20).
The global correlations are between-country correlations of the rate of children in care and external measures averaged over time and indicate how external measures relate to children in care globally. The strong, positively correlated measures are the care entry rate (.61) and the infant mortality rate (.58), the teen pregnancy rate (.28), the proportion of the population with a substance use disorder (.35), and the substance abuse disorder death rate (.55). The strongest negative correlation is with the proportion of the population aged 0–19 (–.31).
Clustering according to correlations between out-of-home care and various sociodemographic predictors reveals a latent structure to CWS/CPS contact that differs from but complements classification according to region and language.
Discussion
Our results demonstrate significant cross-national variation in child welfare system (CWS) and child protective services (CPS) contact across the Global North. However, they also show that rates of investigation, confirmed maltreatment, and out-of-home care are fairly stable in most cases and that many trends in system contact are shared across many countries. Furthermore, countries within geographic regions and countries sharing social and legal institutions exhibit similarities in system contact. Hierarchical clustering based on correlations between various demographic, health, and socioeconomic measures and the rate of children in care reveals two main groups of countries with contrasting correlational patterns, highlighting the varying associations of these factors with out-of-home care rates across countries. Our analysis does not reveal a single, definitive classification of countries in the Global North according to CWS/CPS contact and its relationship to other social factors. However, it demonstrates helpful ways to think about such a classification and supports the conclusion that further large-scale comparative research on CWS/CPS contact is feasible and likely to be fruitful.
Although our data do not permit us to directly evaluate statistical differences in CWS/CPS systems across countries, several comparisons offer suggestive evidence about qualitative variation. First, rates of investigation, confirmed maltreatment, and out-of-home care are not always consistent within countries: some countries display relatively high rates of some measures and relatively low rates of others, suggesting that countries vary significantly in their transition probabilities from investigation to substantiation to out-of-home placement. Second, the incidence of care entry and the point prevalence of care also diverge in several countries, indicating that countries vary meaningfully in the duration that children spend in out-of-home care—another core area meriting future research using these data. These two points of evidence suggest that countries in the Global North might differ meaningfully in their handling of child maltreatment cases, conditional on children's initial contact with CWS/CPS.
Note that the mechanisms through which CWS/CPS contact occurs are highly variable across countries. Within countries, CPS contact is highly unequally distributed by race, ethnicity, and class (Fallesen et al. 2014; Kim et al. 2017; Magruder and Shaw 2008; Pierce et al. 2022; Rebbe et al. 2022; Rouland and Vaithianathan 2018; Sabol et al. 2004; Wildeman and Emanuel 2014; Wildeman et al. 2014). Accordingly, observed cross-national differences in system contact might be driven meaningfully or even predominantly by exceptionally high contact rates among marginalized groups. Structural racism plays a central role in shaping child welfare system contact in countries such as the United States (Edwards et al. 2021), but it likely plays a smaller role in more racially homogeneous nations. Additionally, the role of police in shaping child welfare contact likely varies across nations. For example, in 2022, legal and law enforcement referrals accounted for 21% of referrals in the United States and 24% of cases in Australia (Australia Institute of Health and Welfare 2024; U.S. Department of Health and Human Services 2024), but they likely represent a much smaller share of cases in less heavily policed societies. For all these reasons, measuring cross-national variation in the mechanisms that shape CWS/CPS contact is a pressing priority for future research.
The cross-national variation in CWS/CPS contact observed in our study raises key questions about the extent to which it results from differences across countries in their underlying incidence of child abuse and neglect or in their child welfare definitions, policies, and procedures. Countries define maltreatment differently. For instance, Sweden banned all forms of corporal punishment in 1979 (Durrant 1999), and many countries have since followed suit, but it remains legal in certain contexts in countries such as the United States. Even within the United States, maltreatment definitions vary by state. Although corporal punishment is officially seen as maltreatment in 96% of U.S. states, 79% of states have exemptions for “physical discipline considered reasonable” that “does not cause bodily harm” (Lee and Weigensberg 2022). Different cultural and legal norms regarding disciplining children could result in differing levels of CWS/CPS contact.
Further research is needed to understand the relative importance of behavioral and institutional factors, including institutions beyond those traditionally examined in the CWS/CPS space. This research could build on work by Sutton (2000) and others who have examined the interplay of behavioral and institutional factors in criminal legal systems. Just as Sutton (2000) found that prison growth in different nations was influenced not only by crime rates but also by welfare spending and political dynamics, future CWS/CPS research should consider a broad range of potential determinants beyond child maltreatment itself. Moreover, as Wildeman (2016) demonstrated in the context of incarceration and population health, the societal consequences of CWS/CPS involvement might vary across countries. Harmonized data on CWS/CPS contact are now broadly available through the RDGCWD—introduced in this research note—providing valuable resources for understanding the various national factors that contribute to both levels and changes in levels of system contact and the subsequent population-level consequences of system contact.
Given that CWS/CPS exposure is consequential for both individuals and society, it is paramount that countries secure and make available data on children's CWS/CPS contact. We sought data from 64 countries and obtained data on at least one measure from 44 countries. Our inability to access data might be systematically related to levels of child maltreatment or CWS/CPS contact. Although it is possible to compare several countries across indicators, in many countries in the Global North, there is a dearth of publicly available knowledge about children's CWS/CPS contact, particularly for earlier stages of contact.
Acknowledgments
The study was funded by grants from the ROCKWOOL Foundation (grant 1241) (to Christopher Wildeman, Alexander F. Roehrkasse, Liza Becker, and Peter Fallesen) and FORTE—the Swedish Research Council for Health, Working Life and Welfare (grant 2016-07099) (to Peter Fallesen). The authors have no conflicts of interest to disclose. A previous version of this note was presented at the 2023 annual meeting of the Population Association of America.
Notes
We use CWS/CPS throughout because CWS and CPS are common acronyms in different countries.
The database is freely available through the National Data Archive on Child Abuse and Neglect. Because the data are aggregated and do not pose disclosure risks for individual children exposed to CWS/CPS, they are available without IRB approval. The code for replicating the analysis is available on the Open Science Framework (https://osf.io/wfqpb/).
With one exception, we limited data collection to administrative data on children aged 0–17. For Canada, administrative survey data were available only for children aged 0–15. Comparing data from Canada with peer countries, such as the United States, indicates that any bias resulting from differing age ranges is likely small (Figure S4).
Data for Sweden in 2014–2017 were not comparable because of administrative changes.
Note the break in the y-axis in Figure 2, panel d.
Bulgaria and Kazakhstan were excluded in this clustering because of a lack of data (see the Methods section).