Is Foster Care Placement in the U.S. More Likely for Black Children Than for Equally at-Risk White Children?

This study asks whether Black children in the United States are more likely than equally at-risk White children to be placed in foster care. Using data from over 23 million child maltreatment investigations in 45 states between 2008 and 2020, the researchers compare placement rates while accounting for each child’s risk of future maltreatment. They find that Black children are placed in foster care at significantly higher rates—about 1.2 percentage points more, or 21% higher—than White children with the same risk levels. The racial gap has narrowed over time but remains concentrated among cases with potential for future maltreatment.

The Policy Scientist's Perspective

This article tackles one of the most enduring and morally charged issues in social policy: whether U.S. foster care decisions continue a historical pattern of unjust family separation. Across time and place—from Native American boarding schools to contemporary Russian child removals from Ukraine—states have used the rhetoric of protection to justify coercive child placements. The question of whether such biases persist today is thus both empirically and historically significant. This article extends recent causal inference innovations to an extraordinary national dataset encompassing over 23 million child protection investigations across 45 states. The authors employ a nonparametric bounding approach that offers a credible test for “unwarranted” racial disparity while avoiding the overreach of traditional regression models. The data are exceptionally broad and internally consistent, enhancing confidence in both the findings and their generalizability to other advanced welfare systems. The study’s results—showing that Black children are still placed in foster care at significantly higher rates than equally at-risk White children—underscore the persistence of structural inequities within state authority over families.

Full Citation and Link to Article

Baron, E. J., Doyle, J. J., Emanuel, N., & Hull, P. (2025). Unwarranted racial disparity in U.S. foster care placement. Journal of Policy Analysis and Management, 1–22. https://doi.org/10.1002/pam.70030

Central Research Question

This article investigates whether racial disparities in U.S. foster care placement reflect differences in children’s underlying risk of maltreatment or evidence of unwarranted bias. Specifically, it asks whether Black children are more likely than equally at-risk White children to be placed into foster care. The study distinguishes between warranted and unwarranted disparities by comparing placement rates for children with identical potential for subsequent maltreatment if left at home. This research question directly targets a central policy concern: whether the observed racial gap in child protective service (CPS) interventions represents rational responses to risk or systemic discrimination embedded in child welfare decision-making.

The authors also examine how these disparities vary geographically across states, how they have changed over time, and whether they are concentrated among particular types of cases—specifically, those with higher or lower risk of subsequent maltreatment. By doing so, the paper not only measures racial disparity but also seeks to clarify its policy relevance: whether differences in placement rates imply over-intervention for Black families, under-intervention for White families, or both.

Previous Literature

Prior research consistently shows that Black children in the United States are disproportionately represented in the foster care system. Approximately 10 percent of Black children experience foster care placement before adulthood, compared to about 5 percent of White children. This gap has persisted for decades and varies widely across states. Earlier studies, including those by Wildeman and Emanuel (2014) and Yi et al. (2023), documented these disparities but left open the question of whether they stem from discrimination or from genuine differences in family risk factors.

Existing empirical work often relied on multivariate regression analyses controlling for observable socioeconomic and family characteristics such as poverty, parental substance abuse, or prior CPS involvement. However, the authors note that such models are limited by two types of bias. First, omitted variable bias arises because key determinants of child risk—like the potential for future maltreatment—are unobservable. Second, included variable bias occurs when researchers control for factors that may themselves mediate discriminatory treatment, such as poverty or neighborhood effects. Controlling for these variables can inadvertently “explain away” discrimination rather than measure it.

The study builds directly on recent advances in causal inference and bounding techniques, particularly the framework introduced by Arnold, Dobbie, and Hull (2022) to measure discrimination in bail decisions and the state-level work by Baron et al. (2024a) on racial disparities in Michigan’s foster care system. These prior studies pioneered nonparametric bounding approaches to infer discrimination under minimal assumptions, recognizing that in many policy domains, counterfactual outcomes are unobservable. The current paper extends this methodological approach to national data, offering the first set of nationwide estimates of “unwarranted disparity” (UD) in foster care placements.

Data

The study uses administrative data from the National Child Abuse and Neglect Data System (NCANDS) Child Files, maintained by the National Data Archive on Child Abuse and Neglect. The dataset spans 2008–2021 and includes approximately 23 million maltreatment investigations from 45 states, of which roughly 8 million involve Black children. The coverage is comprehensive and includes nearly all U.S. states except New York, which does not report foster care placements. For a subset of states, data availability varies by year, but the main analyses are restricted to a balanced panel from 2008 to 2020.

The NCANDS files include child-level information on race, gender, age, type of alleged maltreatment, prior CPS history, and whether the investigation resulted in foster care placement. Each case also contains unique identifiers allowing the researchers to track repeat investigations and determine whether children were re-investigated for maltreatment within six months—a key variable used to proxy “subsequent maltreatment potential.” The analysis focuses on Black and White children under age 17 to ensure sufficient follow-up time for reinvestigation.

The authors supplement these data with state-level variables that may correlate with foster care decisions, such as average income, inequality (Gini coefficient), single-parent rates, social worker racial composition, and welfare generosity (e.g., TANF benefits and Medicaid eligibility). These contextual variables provide insight into how broader socioeconomic and institutional factors may correlate with racial disparities.

The NCANDS dataset is widely recognized for its breadth but also for variability in data quality across states. The authors address this by excluding problematic variables—such as those inconsistently reported—and by focusing on consistently defined outcomes like foster care placement and subsequent investigation. Given its national scope, the data provide unusually high external validity and allow generalization to other jurisdictions with comparable CPS structures.

Methods

The main methodological challenge is that the true risk of maltreatment for children placed in foster care cannot be observed: once a child is removed from the home, it is impossible to know what would have happened had the child remained. The authors overcome this missing counterfactual using a nonparametric bounding approach inspired by Charles Manski’s (1990) work. They construct upper and lower bounds on each child’s likelihood of subsequent maltreatment if left at home and then estimate racial disparities within these bounds.

The upper bound assumes the extreme case that all children placed in foster care would have experienced subsequent maltreatment had they remained at home. The lower bound assumes they would have faced the same risk as children of the same race who were not placed in care. Using these two assumptions, the authors estimate a bounded range of possible “unwarranted disparities” (UDs) in placement rates between Black and White children with equivalent maltreatment potential.

They then summarize these bounds into a single national measure of UD using the midpoint and uniform average across the feasible range, following the robust decision rule proposed by Song (2014). This aggregation produces interpretable and conservative estimates that reflect the uncertainty inherent in the counterfactual.

The analysis also disaggregates results by state, year, and maltreatment potential (cases with or without future risk). To validate the approach, the authors compare their bounded estimates for Michigan with quasi-experimental results from Baron et al. (2024a), finding close alignment. They further test robustness by substituting alternate proxies for maltreatment—such as subsequent substantiated investigations or subsequent foster placements—and by checking consistency across mandated and non-mandated reporters.

Findings/Size Effects

The national analysis finds that racial disparities in foster care placement persist even after accounting for children’s underlying risk. Across the United States, Black children are significantly more likely to be placed in foster care than White children with identical maltreatment potential. The average unwarranted disparity (UD) is approximately 1.22 percentage points—equivalent to a 21 percent higher placement rate for Black children relative to White children (5.8 percent baseline). This disparity is statistically significant even under the most conservative assumptions; the lower bound remains above zero, indicating bias robust to alternative interpretations.

Unwarranted disparity is not evenly distributed across cases. It is concentrated among children who exhibit potential for subsequent maltreatment. For these higher-risk cases, the UD is roughly 3.6 percentage points—five times larger than among lower-risk cases (0.69 percentage points). This pattern suggests that disparities are most pronounced when CPS investigators face the most discretion, reinforcing concerns about subjectivity in high-stakes decisions.

Geographically, 37 of 45 states exhibit positive UD estimates, with particularly large disparities in states with predominantly White caseworker populations. States with higher shares of Black social workers or larger Black populations tend to show smaller UDs. These correlations are descriptive but consistent with a “racial concordance” effect documented in earlier Michigan research: investigators are less likely to remove children of their own race.

Over time, UD has declined substantially—from 1.9 percentage points in 2008 to 0.76 by 2020, a 60 percent reduction. The decline is driven primarily by reduced placement rates among Black children, while placement rates for White children remain stable. This suggests that CPS reforms and greater awareness of racial inequities have narrowed, but not eliminated, bias.

The authors also find that reporting bias—differential likelihood of being investigated again—does not drive the results. UD trends are similar for mandated and non-mandated reporters, and robustness checks using substantiated cases yield consistent results.

Conclusion

This study provides the strongest national evidence to date that Black children in the United States face higher odds of foster care placement than equally at-risk White children. By using nonparametric bounds rather than parametric regression, the analysis avoids reliance on unverifiable functional form assumptions and illustrates how causal inference principles can be applied even in non-experimental data. Although the method cannot fully identify the causal mechanism—whether investigator bias, institutional incentives, or community-level differences—the evidence strongly suggests that structural inequities persist in the administration of child welfare.

The findings carry significant implications for child protection policy. If disparities are concentrated among children with genuine maltreatment potential, the issue may not simply be “over-removal” of Black children but also “under-removal” of White children exposed to equivalent risks. This reframes the debate from one of uniform bias to one of inconsistent thresholds of intervention.

The study also informs broader debates about algorithmic decision-making in social policy. By quantifying the uncertainty and bias inherent in human judgment, it supports the case for decision-support systems that can improve consistency and fairness. The results are timely given recent federal and state efforts to address systemic racism in child welfare and to reform CPS practices following historical abuses against marginalized families, including Native American child removals and racially disparate interventions.

Overall, the paper represents a major empirical contribution. It combines national-scale administrative data with a transparent, assumption-light methodology, producing estimates that are both credible and policy-relevant. While not a randomized control trial, it approaches causal identification more closely than conventional observational studies. The data’s scope, the rigor of the method, and the prominence of the authors make this one of the most influential studies on racial disparity in child welfare policy in recent years.