Racial/Ethnic Differences in Child Protective Services Reporting, Substantiation and Placement, With Comparison to Non-CPS Risks and Outcomes: 2005–2019

We used National Child Abuse and Neglect Data System and Census data to examine Black–White and Hispanic–White disparities in reporting, substantiation, and out-of-home placement both descriptively from 2005–2019 and in multivariate models from 2007–2017. We also tracked contemporaneous social risk (e.g., child poverty) and child harm (e.g., infant mortality) disparities using non-child protective services (CPS) sources and compared them to CPS reporting rate disparities. Black–White CPS reporting disparities were lower than found in non-CPS risk and harm benchmarks. Consistent with the Hispanic paradox, Hispanic–White CPS reporting disparities were lower than risk disparities but similar to harm disparities. Descriptive and multivariate analyses of data from the past several years indicated that Black children were less likely to be substantiated or placed into out-of-home care following a report than White children. Hispanic children were slightly more likely to be substantiated or placed in out-of-home care than White children overall, but this difference disappeared in multivariate models. Available data provide no evidence that Black children were overreported relative to observed risks and harms reflected in non-CPS data. Reducing reporting rates among Black children will require addressing broader conditions associated with maltreatment.


Introduction
Child protective services (CPS) referrals represent a suspicion of child abuse or neglect and are investigated if they meet statutory guidelines.A CPS investigation is a widely recognized indicator of potential victimization, and researchers have generally found that investigation alone-regardless of substantiation or intervention-is predictive of future victimization and adverse social and developmental outcomes (Drake et al., 2003;Hussey et al., 2005;Kohl et al., 2009).It is well established that Black children are investigated by CPS at higher rates than White children (Edwards et al., 2021;Kim et al., 2017;Yi et al., 2020).Yet because there is no population-level census of victimization or maltreatment exposure independent from CPS contact, competing perspectives persist as to whether disproportionate representation in CPS investigations derives from differential risk of child maltreatment or differential treatment independent of risk.In support of the latter explanation, some scholars have argued that Black children are not only more likely to be referred for maltreatment, but also more likely to have their investigations substantiated and be placed in out-of-home care following an investigation (e.g., Cénat et al., 2021;Columbia Journal of Race and Law, 2020;Dettlaff, Weber, et al., 2021;Palusci & Botash, 2021;Tajima et al., 2022).
Racial bias (by which we mean unfair or unequal treatment among otherwise similar children that may derive from either individual prejudices or inequitable policies or practices) is difficult to measure and assess directly.It is, however, possible to contextualize racial disproportionality in CPS relative to disproportionality in other non-CPS indicators of child risk and harm.The baseline or "expected" rate of CPS involvement for a particular group should be equal to the rate at which children in that population experience child abuse, neglect, or imminent risk thereof.However, because that base rate is unobservable, we considered the population rates of indicators that are known to track closely with risk of child abuse and neglect, along with indicators that demonstrate other forms of empirically observable harm experienced by a child.In the current study, we posed a straightforward empirical question: Do racial disproportionalities in CPS contact (i.e., referral, investigation, substantiation, or placement in foster care) exceed disproportionalities in independently documented indicators of social risk (e.g., children in poverty, children in single-parent families) and child harm (e.g., infant mortality, homicide injury) that are not subject to concerns of systemic measurement bias?If CPS contact is a high outlier-meaning that it ascribes risk to Black children in substantial excess of the social risk and physical harm indicators-that would suggest discriminatory practices by the CPS system.In contrast, if patterns of CPS contact are consistent with external metrics of risk or harm, it implies CPS contact likely reflects persistent societal inequities.

Background and Framework
Several studies suggested that Black children are reported to CPS at similar or slightly lower rates than White children (Barth et al., 2020;Drake et al., 2011;Kim & Drake, 2018;Maloney et al., 2017;Putnam-Hornstein et al., 2013) when adjusting for indicators related to income and family context.The current study explored these questions using an intentionally consequentialist lens.Put simply, we wanted to understand the movement of children of different races and ethnicities into the CPS system, relative to other benchmarks.To be clear, a finding that CPS contact is not an outlier with respect to race-specific signaling of risk or harm is not a justification for inequities or inaction.Rather, it would point our attention to the necessity of addressing conditions that most commonly lead to CPS involvement-multigenerational poverty, unequal access to substance abuse treatment programs, underresourced schools, poorly coordinated mental health supports-which cannot be resolved by any reform strategy that focuses solely on the actions of CPS or the behavior of mandatory reporters.
Importantly, the current study further differentiated the risk or suspicion of harm that leads to a CPS referral and investigation from decisions to substantiate or place a child in foster care.Several studies have found that once a decision has been made to investigate a referral of alleged maltreatment, Black children are no more likely to be substantiated or placed than White children, and may be less so (e.g., Barth et al., 2020Barth et al., , 2021;;Casanueva et al., 2009;Cheng & Lo, 2013;Fix & Nair, 2020;Maloney et al., 2017;Putnam-Hornstein et al., 2013).The most recent National Incidence Study of Child Abuse and Neglect (Sedlak et al., 2010) and rates of child maltreatment fatalities (U.S.Department of Health and Human Services [DHHS], 2021, 2022), also indicated elevated rates of child abuse and neglect for Black children.Still, studies have found that Black race is associated with increased odds of substantiation or placement (Dettlaff et al., 2011;Maguire-Jack et al., 2020).These studies, however, have been misrepresented or overgeneralized in the literature.For example, Maguire-Jack et al. ( 2020) reported a bivariate (unadjusted) difference in Black-White substantiation rates of 11% and found that the difference dropped to 3% after very modest covariate adjustments.They acknowledged that this small remaining difference could be explained by unobserved differences in risk factors, given that the only maltreatmentrelated adjustment in the study was for allegation type, which (due to large and poorly defined categories) is weakly related to risk or harm (Jackson et al., 2019).Meanwhile, Dettlaff et al. (2011; a single-state study using data from nearly 20 years ago) found no disparity in substantiation in a model adjusting for income and other sociodemographic factors.Only when they added a caseworker risk assessment score to the model did the coefficient for Black race become larger and statistically significant.
Nevertheless, the perspective that anti-Black bias pervades CPS has been made confidently in recent articles, commentaries, conferences, and the grey literature by political movements dedicated to abolishing CPS, respected child advocacy organizations, leaders of eminent professional organizations, and U.S. government publications (Annie E. Casey Foundation et al., 2022;Child Welfare Information Gateway, 2021;Columbia Journal of Race and Law, 2020;Dettlaff, Boyd, et al., 2021;Dettlaff, Weber, et al., 2021;Fitzgerald, 2022;Palusci & Botash, 2021;Tajima et al., 2022;United Nations, 2022).Encouraged by testimony based on this view (White, 2022), the United Nations recently decided to assess whether the U.S. CPS system is among America's racist institutions (Fitzgerald, 2022).To the extent that racial disparities in substantiation and placement persist among investigated reports, this would not per se indicate conclusive evidence of bias, but it would clearly motivate and justify additional empirical inquiry, particularly with implicated states or localities.Understanding where disparities emergein initial referrals to CPS versus subsequent decisions made by CPS-and whether they are in substantial excess of known indicators of social risk and physical harm to children-has profound consequences for policy and training and deserves rigorous empirical evaluation.

Objective
This paper used universal national data to address two questions: (1) Are Black-White and Hispanic-White disparities in CPS reporting lower than, similar to, or higher than those observed in non-CPS measures of social risk (e.g., poverty, low education) and child harm measures (e.g., mortality, very low birthweight)?(2) Once referred and screened in for investigation, do Black or Hispanic children experience substantiation and removal into foster care at lower, similar, or higher rates than White children, in either unadjusted or adjusted estimates?We examined these questions longitudinally on a yearly basis from 2007-2017, with a few measures expanded to 2005-2019.
We focused on Black, Hispanic, and White populations for several reasons.First, much of the current debate has centered on these populations.Second, these three groups have large populations, allowing for stable estimates over time, even of rare events such as maltreatment fatality reports.Finally, some less populous groups are subject to censoring, which limits available data from public sources.As an example, it is quite common for income data on Native Hawaiian or Pacific Islander populations to be censored in the American Community Survey due to small populations and fears that aggregate data could be traced back to individuals.Native American populations pose even more substantial challenges, because some are served by parallel child welfare systems under tribal jurisdiction and it is not clear that those children are represented in the National Child Abuse and Neglect Data System (NCANDS) data.Although some of these difficulties can be addressed using restrictive sampling and analysis, the complexity of the issues addressed in this study led us to focus on the three largest racial and ethnic groups with the best available data.

Data
We obtained CPS data from the NCANDS Child File.Data from 2005-2019 were used for descriptive bivariate analyses.We excluded 2020 data due to fluctuations in reporting during the COVID-19 pandemic.For multivariate analyses, we used a subset of the data (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017) for which countylevel household income data were available from the American Community Survey 5-year estimates.Due to issues with data quality or missingness, we omitted data from the following states and years: Pennsylvania (all years), Oregon (2005-2011), North Dakota (2005-2009), Michigan (2005Michigan ( -2007)), and Maryland (2005Maryland ( -2006)).Additionally, due to missing data in foster care files, we excluded North Carolina and New York (all years), Illinois (2017), Georgia (2005( -2011( ), Alabama (2005( -2008)), and Arkansas (2005).When calculating rates for years when NCANDS data were not present in some states, we adjusted the national population counts by omitting those states.When duplicate records existed for the same child event (<1 in 1000 records), the earlier record was deleted because those records were generally less complete.

Child File Variables
Children in the report dataset were coded as substantiated if any maltreatment disposition variables were coded as 1 (substantiated) or 2 (indicated or reason to suspect).This is consistent with the definition of "victim" status used in national publications (DHHS, 2021, p. 20).A child race and ethnicity variable was derived by aggregating the child race and child ethnicity indicator variables.Children with a Hispanic designation were coded as Hispanic, regardless of racial status.Otherwise, non-Hispanic Black (no other race indicated) children were coded as Black, and non-Hispanic White (no other race indicated) children were coded as White.A reporter type variable was coded as (a) professional if the report source was listed in the Child File as education, legal or law enforcement, medical, social service, mental health, child daycare provider, or foster care provider; (b) nonprofessional if the source was listed as parent, other relative, friend, neighbor, or alleged victim; or (c) unclassified for other report source categories (i.e., other, anonymous, unknown, or missing).We used the maltreatment type variables in NCANDS to code maltreatment as physical, neglect (neglect or deprivation of necessities or medical neglect), sexual, or psychological (psychological or emotional) if any of these occurred as the sole form of maltreatment.Children with more than one type were classified as multiple.Children with none of the maltreatment types or with missing data were classified as other.The various subtypes we combined into the other category were not interpretable at a national level, because these subtypes are used very differently in different states.As an extreme example, Missouri classifies almost all unsubstantiated cases as other.For those reasons, our other category is best thought of as none of the above.Our age and sex variables were taken from data on child age at report and child sex.Less than 1% of records had missing final values using these categorizations or ages outside the 0-17 years range, and these were dropped.Fatality records (<2000 records per year of more than 3,000,000) were also omitted from the main data file, because they lacked geographic identifiers due to confidentiality concerns (DHHS, 2019).

Census Variables
Census data were used to create our contextual income variable, which represented same-race household income for Black, White, and Hispanic residents of each county during the year.For example, St Louis County, Missouri, from 2015-2019, had a full-county median household income based on all residents of $67,420, but a same-race income of $43,801 when only Black households were considered and a same-race income of $77,989 when only White households were considered (U.S. Census Bureau, 2020).Instead of using full-county income metrics for all races, we used samerace income.This approach has been used in the past (Kim & Drake, 2018;Wulczyn et al., 2013) and demonstrated to provide superior model fit and better convergence with individual-level study findings than the traditional fullcounty income metric (Jones et al., 2022).County income data by race are not available yearly, requiring the use of the 5-year estimates data from the American Community Survey obtained through Social Explorer (n.d.).These data were applied to the center year of each 5-year interval (e.g., our 2017 income data was based on 2015-2019 estimates).Income data for the 2016-2020 period were not used due to the sampling timeframe including post-COVID-19 data.

County, State, and Year Coding
Approximately 1 in 6 children in the NCANDS Child File do not have a county identifier but do have a state identifier.This is due to confidentiality concerns, because any child in a county with less than 1000 reports in any given year is coded as "000" rather than with a county code.These federal codes identify geographic regions in the United States (e.g., "189" for St Louis County).These censored counties can change from year to year, as report totals move above or below 1000.We accommodated this problem in two ways.First, we created pseudo-counties consistent with the methodology adopted in prior work (e.g., Kim & Drake, 2018).Censored counties were combined into a single pseudo-county in each state.The income for this pseudo-county was derived from the populationweighted census income values for all censored counties (for more details, see Kim & Drake, 2018).Second, we assigned children to the year their report was submitted to NCANDS, not the year the report occurred.These yearly dates differed about 14% of the time, with virtually all differences being that the reporting year was the federal year prior to the submission year.Of course, this worked both ways, with a given year gaining reports from the prior year and losing reports to the next year.
We used submission year instead of reporting year to ensure the conformity of county membership for all children each year, because county censoring was consistent with submission year.If the reporting date were used instead of submission year, children in the same reporting year (but different submission years) could be either in their actual county or in a pseudo-county (if the report was submitted the following year and the report total dropped under 1000), thus undesirably producing children in the same county with different county income values.Genuinely missing county identifiers also exist (coded as "999"), and these records (<0.4%) were excluded.In addition, the income variable was censored in the census for a very small number of Black, Hispanic, and White children (<0.02%) due to small populations during the 2007-2017 timeframe in which income was used for multivariate analyses, and those records were also removed from the data.A separate fatality dataset was used only to generate yearly national CPS fatality counts from the NCANDS Child File to be used in Figure 1, noting cases involving deaths.Fatality data lack county or state identifiers and were coded for race and ethnicity and by year only.

External Risk and Harm Indicators
We developed two sets of external referents.These fall broadly into social risk and child harm categories.For these data, we used non-Hispanic White, non-Hispanic Black, and Hispanic racial and ethnic categories for the child harm variables, but social risk variables were restricted to Black rather than non-Hispanic Black categorization due to data availability constraints.Social risk variables included children in poverty, children in single-parent families, teen birth rate, and adults without a high school degree.Children in poverty, children in single-parent families, and teen birth rate data were obtained from the Kids Count Data Center (2020a, 2020b, 2020c), whereas the percentage of people aged 25 or older without a high school degree was obtained from the U.S. Census Bureau (2022).Population counts used to calculate rates of CPS reporting were obtained from the Kids Count Data Center (2021).
Child harm variables included very low birthweight, very preterm births, infant mortality, homicide injury, and unintentional death.Data for very low birthweight (weight <1.5 kg) and very preterm (gestational age <32 weeks) births were from annual National Vital Statistical Records publications (Martin et al., 2009(Martin et al., , 2017(Martin et al., , 2021)).Infant mortality data were from the Centers for Disease Control and Prevention (CDC) Wide-Ranging Online Data for Epidemiological Research database, which uses linked records of live births and infant deaths, specifying year of death and race and ethnicity (CDC, n.d., 2022).Age-adjusted homicide injury (homicide only, no legal intervention data included) and unintentional death (not including adverse effects or transportation injuries) data by race and ethnicity for children aged 0-14 years were from the CDC Web-Based Injury Statistics Query and Reporting System (Centers for Disease Control and Prevention, 2020).

Analysis
Descriptive bivariate data are presented in graphic form by race and ethnicity as yearly disparity ratios (DRs).These are simply the rate of Black or Hispanic events divided by the rate of White events-for example, a rate of 210 in 1000 Black children and 100 in 1000 White children in a given year would yield a DR of 210 divided by 100, or 2.1 (e.g., Figure 1).Among CPS reports, simple percentages of Black, Hispanic, and White children substantiated or placed in care following a report are shown in Figure 2. We also present yearly multivariate estimated odds ratios (ORs) of substantiation and placement disparities by year (Figure 3, Tables 1-4) while controlling for child age, sex, maltreatment type, report source, and same-race county income.To enable comparisons, we calculated bivariate ORs using only children of each race and ethnicity experiencing or not experiencing a given outcome, along with multivariate ORs that accounted for other variables.We report both estimates to determine if the multivariate analyses provided fundamentally different results than simpler calculations, because both are often used in the literature.All data elements in the figures are mirrored in tabular numeric form in the Supplemental Materials.
We employed multilevel logistic regression modeling to estimate the odds ratios of Black or Hispanic children, in comparison to White children, being substantiated or entering foster care.To enable direct comparisons, each analysis included only either Black and White children or Hispanic and White Children, yielding four models for each year.We controlled for same-race county median household income, child age, child sex, maltreatment type, and reporter type.Models included county-and state-level random effects (intercepts) to address the nesting of children in counties and states.We used the GLIMMIX procedure in SAS (version 9.4) to fit the models, which can be found in Tables 1-4.

CPS Reports
We present four graphs (four panels in Figure 1) to address our first question.Figure 1a reports Black-White risk DRs for CPS reports, children in poverty, children in single-parent families, teen birth rate, and adults without a high school degree.All DRs were greater than 1.0 and ranged from 2.0 to about 3.0, showing consistently higher risk for Black children.The CPS report DR was less than 2.0 and lower than all non-CPS social risk DRs. Figure 1b again shows the Black-White DRs for CPS reports, alongside DRs for the external measures of child harm: White DRs for very low birthweight, very preterm births, infant mortality, homicide injury and unintentional death, and CPS fatal reports.Figure 1a and b are remarkably consistent.With the exception of accidental death DRs (which are similar to CPS report DRs), all non-CPS harm DRs were higher than CPS report DRs.This is even true of CPS fatality reports, where the DR was slightly greater than 2.5, compared to the CPS report DR of less than 2.0.
Hispanic-White risk DRs (Figure 1c) show more variability than Black-White DRs, with DRs ranging from slightly more than 1.5 for single-parent households to more than 4 for not completing high school.DRs for other risk measures ranged from about 2.2 to 3.3.The Hispanic-White CPS report DR was much lower than the Black-White CPS report DR at about 1.0.Hispanic-White harm DRs (Figure 1d) were fundamentally different from the three prior graphs.In general, child harm indicator DRs clustered around 1.0, consistent with the Hispanic-White CPS reporting DR.

Substantiation and Foster Care
Addressing our second question, pertaining to bias in the CPS system, we present the descriptive bivariate percentages of Black and White children in the NCANDS Child File who were substantiated or indicated (Figure 2a) or who were removed into foster care (Figure 2b), along with similar tables for Hispanics (Figure 2c and d).The core finding is that none     of these rates were very different-the (very compressed) Yaxis in all three graphs shows that the differences were not great.Small differences do exist, however.Prior to about 2011 or 2012, Black children were slightly more likely to be substantiated or removed than White children.But from that time forward, the trend reversed: Black children were slightly less likely to be substantiated or removed.The picture for Hispanic children is quite different, with Hispanic children in the bivariate analyses showing slightly higher (1%-3%) rates of substantiation and placement in foster care than White children in most years.All these relationships change when adjusting for multivariate controls (i.e., child age, sex, maltreatment type, report source, and same-race county income) while clustering at the state and county levels.The multivariate models on substantiations and placements for Black and White children are presented in Tables 1 and 2. Estimates for Hispanic and White children are presented in Tables 3 and 4. Figure 3 visually shows both bivariate and adjusted Black-White and Hispanic-White ORs.For Black-White disparities, both unadjusted substantiation and placement ORs were slightly greater than 1.0 early in the timeframe examined but less than 1.0 in more recent years.The multivariate Black-White substantiation ORs never exceeded 1.0 and since 2011, they have been notably lower.The multivariate Black-White placement ORs were slightly higher than 1.0 in 2007-2009, but as with substantiation, they show a similar declining trend as of 2010 and were below 0.8 in recent years.These findings indicate that holding demographic factors constant, Black children had somewhat lower odds of substantiation and placement in foster care than White children in the most recent pre-COVID-19 data.
Again, Hispanic children look quite different.Bivariate ORs for Hispanic children were almost always above 1.0, most recently about 1.1, showing slightly higher odds of being both substantiated and placed in foster care in the raw data.When multivariate controls were employed, this difference disappeared, with Hispanic and White children having essentially similar odds of being substantiated or placed in care (about 1.0) in recent years.

Discussion
Overall, these data suggest two primary conclusions.First, Black-White CPS report rate disparities are in alignment with or smaller than Black-White disparities in external indicators of social risk and child harm.Depending on how abuse and neglect are defined, it is certainly possible that all children are overreported relative to external indicators of risk and harm.But we found no evidence to suggest that overreporting was concentrated among Black children.We also found no evidence that once investigated, Black children were disproportionately substantiated or placed in foster care.Black children, however, faced much higher rates of exposure to social risk and fared worse in non-CPS indicators of child harm such as infant mortality, homicide, and unintentional childhood death.Second, we found continued evidence for the Hispanic paradox in CPS reporting compared to observed risk exposure.The slightly greater raw rates of substantiation and placement among Hispanic children were explained when other controls are employed.We expand on these findings in the sections that follow.
Regarding our first research question, which addressed the degree to which racial disparities in CPS reporting are higher than, lower than, or similar to those in non-CPS risk and harm indicators, the results varied.That said, Black-White disparities in all indicators exceeded 1.0, meaning that Black children were consistently exposed to more risks and experienced harm at greater rates than White children.Additionally, Black-White disparities in CPS reporting were clearly and consistently lower than what was observed for the non-CPS risk and harm indicators.To the extent that CPS is an outlier in signaling racial disparities in risk, it operates in the opposite direction than would be expected if disparities were driven by CPS reporting of Black-but not White-children in excess of social risk or harm.This does not rule out systemic overreporting in general, nor does it indicate that CPS should respond to all reports.Rather, it suggests that if there is systemic overreporting, it is not specific to Black children and thus, unlikely to be driven by racial animus.Instead, contemporary data suggest that disparities in CPS reports align with differential exposure to risk and barriers to resources (Lloyd et al., 2021;Parolin, 2021).
Results specific to Hispanic-White risk, harm, and CPS report disparities suggest that although Hispanic children face far greater exposure to social risks than White children (e.g., poverty), they experience harm and CPS reporting at about the same rates as White children.Although we could not control for nativity due to data limitations, these findings are consistent with the Hispanic paradox or healthy immigrant effect.These terms describe a consistent pattern wherein relatively recently immigrated Hispanic children and mothers, despite having much higher risk factors increasing their likelihood of poorer well-being indicators, tend to experience well-being indicators similar to those of Whites.This phenomenon is well documented in the child maltreatment (Davidson et al., 2019;Kim & Drake, 2018;Maguire-Jack et al., 2015;Millett, 2016) and medical (Montoya-Williams et al., 2021) literature, if not universally accepted (Tarraf et al., 2020).Because of limitations inherent to child welfare administrative data, few studies describing the Hispanic paradox have disaggregated immigrant populations by nativity, national origin, or socioeconomic position (Acevedo-Garcia & Bates, 2008;Johnson-Motoyama et al., 2015;Pulver et al., 2020).Although National Incidence Study of Child Abuse and Neglect findings suggest little difference in child maltreatment incidence among Hispanic or Latino children compared to White children (Sedlak et al., 2010), aggregate data may mask considerable variation in state-and county-level estimates of disparity or disproportionality (e.g., Johnson-Motoyama, 2014).
Our second research question involved disparities in substantiation and placement following investigation, which could indicate differential treatment or bias in the CPS system.Earlier studies have found different rates of substantiation and placement among those reported and investigated; however, the more recent data used here show a DR near 1.0 across all years and less than 1.0 in more recent years, in both adjusted and unadjusted estimates (see Figure 2).The cause of these changes cannot be ascertained from the data we used for this study.For Hispanic children, slightly greater unadjusted rates of substantiation and foster care placement were noted, but these differences did not appear in multivariate models.Racism is held by many to be pervasive in CPS.We do not doubt that instances of racism are present in all large systems.Our data, however, do not show clear evidence of accumulating anti-Black outcomes in the form of increased substantiation and removal in the CPS system.
It is indisputable that, despite progress in certain areas, the United States has not overcome the legacy of slavery, segregation, and Jim Crow.This legacy lingers most clearly in the patterns of segregation that emerge in many of our major metro areas (Osypuk et al., 2009), which relegate Black children to failing schools in profoundly disadvantaged and often dangerous neighborhoods.In some areas, there is near-zero overlap in the neighborhoods of White and Black children (Manduca & Sampson, 2019).To assert that these patterns, and the poverty and chronic stress they perpetuate, would have no impact on behavioral and psychosocial functioning among the individuals and families in those neighborhoods is to reject decades of scientific consensus on human development (Hyde et al., 2020).Indeed, this history and its unresolved legacy is essential to understanding why Hispanic children face similar individual socioeconomic disadvantage but appear to have significantly lower rates of CPS involvement than Black children (Drake et al., 2011).

Limitations
This study carries strengths and limitations.Core study strengths include the use of national data, the ability to examine long-term trends spanning more than a decade, and estimates that are recent and therefore, relevant to current policy discussions.To our knowledge, this is the only study of child maltreatment reporting, substantiation, and placement with this temporal and geographic span.The presentation of both descriptive and multivariate models is another advantage, allowing assessment of the differences and similarities between raw report data (which are commonly reported and discussed in policy making) and more sophisticated estimates.The limitations of this study are those inherent in the NCANDS Child File, with any errors or omissions necessarily affecting our findings.A second limitation is the inconsistent availability of specific race and ethnicity data for economic controls at the county level and for some risk and harm indicators.Additionally, we did not factor in any possible influences of differential or alternative response (vs.traditional investigation) pathways in our analysis.Some preliminary data have suggested small differences by race in how these path assignments are made (Choi et al., 2021).Factors beyond those considered could have important effects on children and families.For example, once taken to foster care, Black children spend 25% longer in care than White children (Committee on Ways and Means, 2016, Chapter 11, Table 11-11).Differences such as these could have broader impacts across the system that we have not considered.Additionally, available data simply do not support national examination of less populous racial categories.Most maltreatment research has focused on Black and White populations (Johnson-Motoyama et al., 2021), with Hispanic children understudied relative to their numbers and population growth.Other populations, including Native American, Alaska Native, Asian, Native Hawaiian, Pacific Islander, and multiracial and multiethnic populations have received even less attention, partly because their smaller populations make creating stable estimates, or employing suitable contextual variables (such as geographically derived contextual income) problematic (Jones et al., 2022;Kim & Drake, 2018).Finally, we could not disentangle subgroups of broader racial and ethnic categories by country of origin or nativity because that information is not available in the NCANDS data files available to researchers.

Implications
Our findings have implications for the current policy debate about whether CPS and the foster care system should be abolished in the United States.Proponents of abolishing this system routinely cite racial disparities in reporting and subsequent decision-making as justification for their policy preferences, arguing that Black children's contact with CPS is grossly in excess of their exposure to risk or harm.In the current analysis, we found that the racial disparities in CPS contact do not exceed, and are generally smaller than, racial disparities in external measures of social risk and harm.This suggests the need for broader efforts that respond to wellestablished factors, such as residential segregation and intergenerational poverty, that continue to adversely affect the safety, health, and opportunities of Black children and families.Abolition proponents believe that abolishing child protection would-in and of itself-help Black children, premised on the assumption that CPS is not only racially discriminatory but also provides no essential protective service.It is possible that a narrow focus on reducing Black children's CPS involvement without addressing the pronounced inequities documented by the external indicators will result in disproportionate and systematic unresponsiveness to abuse and neglect experienced by Black children.This has been raised as a possibility but is generally absent in the policy debate (Barth et al., 2021;Smith et al., 2021).
Of course, it is essential that CPS improve the quality of services and outcomes for the children brought to its attention, which requires addressing legitimate criticisms about the sensitivity and respectfulness with which CPS engages families of varying racial, ethnic, and cultural backgrounds (Reddy et al., 2022;Simon et al., 2022).We also believe it is essential to acknowledge the limited scope and resources of CPS and, similar to the discussion in medicine (Metzl & Roberts, 2014), encourage greater societal investment in the full range of issues facing Black children and families if we wish to see the desired reductions in exposure to risk and harm.
Some may argue that because all harm indicators included here (e.g., infant mortality, very low birthweight) are influenced by racism, they are not objective and cannot be used as benchmarks.Our argument is different; we argue that these indicators are external to CPS and thus, cannot be caused by racially biased CPS decision-making.Thus, if mandatory reporters or CPS caseworkers systematically targeted Black children with overreporting or overintervention, disparities in CPS contact should be observably larger than disparities in external indicators of risk and harm (regardless of the ultimate cause of those disparities).Our findings do not point to levels of CPS contact that are systematically higher for Black children, given elevated levels of risk and harm.Importantly, we wish to draw attention to the idea that CPS is just one among many indicators suggesting that more support is needed for Black children and families-few would dispute that the unresolved legacy of racism is a root cause of inequalities experienced by Black children.There also appears to be general agreement that CPS cannot be solely responsible for fixing the harms that spill from historical and sustained racial inequality.
We suggest three courses of action in moving forward.First, because the disparity in Black-White CPS reporting is far lower than disparities in other indicators of social risk and child harm, we should acknowledge the current realities of racial inequity and focus attention on large drivers of inequity.Policy efforts based on incorrect or insufficiently influential drivers will at least not alter the needs of Black children and families, and they may make matters worse.That said, racial inequity is real and pervasive in our society and must be addressed far more aggressively.Efforts to reduce disparities in income, access to services, and other issues bearing on children and families must be redoubled.These efforts are key to advancing a just society.Second, although our findings capture nationally representative trends, we must acknowledge that racial or ethnic inequities due to bias may exist in any locality.Such biases demand concrete and tailored responses.Such responses can only be provided when biases can be more precisely described.Indeed, the approach of this study could be replicated in states and counties to identify whether and at what stage (reporting, intervention) disparities may exist at the local level and to target reform strategies accordingly.Legislative initiatives or commissions can assess these concerns and take remedial action to ensure effective and fair distribution of services and decision-making.Third, there is clearly room to consider restructuring child and family policy generally (Berger et al., 2018) to include a focus on providing preventive services, including material assistance, to families.Numerous studies have shed light on the promise of reducing maltreatment and CPS involvement using existing mechanisms to address economic need (Maguire-Jack et al., 2021) and bolder approaches to reduce poverty (Berger et al., 2018).Earlier targeting of preventive interventions could increase diversion from CPS and prevent maltreatment (e.g., Branco et al., 2021;Briar-Lawson et al., 2021;Goodman et al., 2021;Simon et al., 2021), an area that deserves further research.Pending advances in effective prevention, we should continue to bolster the quality, consistency, and effectiveness of CPS responses to children and families subject to a CPS report.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Figure 1 .
Figure 1.Black-White and Hispanic-White disparity ratios for CPS reports, risks, and well-being.

Figure 2 .
Figure 2. Black versus White and Hispanic versus White raw substantiation and foster care trends.

Figure 3 .
Figure 3. Black-White and Hispanic-White descriptive and multivariate odds ratios.

Table 1 .
Odds Ratios from Multilevel Logistic Regression on Child Maltreatment Report Substantiation for Black and White Children.
Note.Reference categories for abuse type and reporter were neglect and nonprofessional, respectively.Fit statistics (generalized chi-square by degrees of freedom) close to 1 indicate adequately modeled dispersion.*p≤ .05.

Table 2 .
Odds Ratios from Multilevel Logistic Regression on Foster Care Placement for Black and White Children.
Note.Reference categories for abuse type and reporter were neglect and nonprofessional, respectively.Fit statistics (generalized chi-square by degrees of freedom) close to 1 indicate adequately modeled dispersion.*p≤ .05.

Table 3 .
Odds Ratios from Multilevel Logistic Regression on Child Maltreatment Report Substantiation for Hispanic and White Children.Reference categories for abuse type and reporter were neglect and nonprofessional, respectively.Fit statistics (generalized chi-square by degrees of freedom) close to 1 indicate adequately modeled dispersion.

Table 4 .
Odds Ratios from Multilevel Logistic Regression on Foster Care Placement for Hispanic and White Children.Reference categories for abuse type and reporter were neglect and nonprofessional, respectively.Fit statistics (generalized chi-square by degrees of freedom) close to 1 indicate adequately modeled dispersion.