Do foster youth face harsher juvenile justice outcomes? Reinvestigating child welfare bias in juvenile justice processing.

Do foster youth face harsher juvenile justice outcomes? Reinvestigating child welfare bias in juvenile justice processing - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Criminol Public Policy . Author manuscript; available in PMC: 2026 Apr 18. Published in final edited form as: Criminol Public Policy. 2024 Nov 1;24(2):273–306. doi: 10.1111/1745-9133.12689 Search in PMC Search in PubMed View in NLM Catalog Add to search Do foster youth face harsher juvenile justice outcomes? Reinvestigating child welfare bias in juvenile justice processing Ezra G Goldstein Ezra G Goldstein 1 School of Public Policy, Georgia Institute of Technology, Atlanta, Georgia, USA Find articles by Ezra G Goldstein 1 , Sarah A Font Sarah A Font 2 Department of Human Development and Family Studies, Department of Public Policy, Pennsylvania State University, University Park, Pennsylvania, USA Find articles by Sarah A Font 2 , Reeve S Kennedy Reeve S Kennedy 3 School of Social Work, East Carolina University, Greenville, North Carolina, USA Find articles by Reeve S Kennedy 3 , Christian M Connell Christian M Connell 2 Department of Human Development and Family Studies, Department of Public Policy, Pennsylvania State University, University Park, Pennsylvania, USA Find articles by Christian M Connell 2 , Allison E Kurpiel Allison E Kurpiel 4 Edinburgh Law School, University of Edinburgh, Edinburgh, UK Find articles by Allison E Kurpiel 4 Author information Article notes Copyright and License information 1 School of Public Policy, Georgia Institute of Technology, Atlanta, Georgia, USA 2 Department of Human Development and Family Studies, Department of Public Policy, Pennsylvania State University, University Park, Pennsylvania, USA 3 School of Social Work, East Carolina University, Greenville, North Carolina, USA 4 Edinburgh Law School, University of Edinburgh, Edinburgh, UK ✉ Correspondence: Ezra G. Goldstein, Georgia Institute of Technology, Atlanta, GA 30332, USA. [email protected] Issue date 2025 May. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. PMC Copyright notice PMCID: PMC13089788  NIHMSID: NIHMS2163827  PMID: 42006119 The publisher's version of this article is available at Criminol Public Policy Abstract Research summary: For decades, child welfare scholars and policy makers have been concerned with the strong association between foster care and juvenile justice involvement. Foster care placement may lead to differences in justice system outcomes if youth in foster care face “processing bias”—differentially harsh treatment by agents of the juvenile court. Previous research found that youth in foster care at the time of juvenile justice contact were treated more harshly by the court, resulting in higher rates of punitive case outcomes. We revisit the question of processing bias using detailed administrative data on more than 10,000 adolescents in Pennsylvania in 2015–2019 and a selection-on-observables design. We find no evidence of processing bias against youth in foster care. Compared to observationally equivalent cases, those that involve youth in foster care do not experience more punitive outcomes. If anything, our estimates suggest the opposite—youth in foster care are less likely to have a charge adjudicated, be placed under court-ordered supervision, or enter into juvenile detention. The precision of our estimates and bounding exercises allow us to rule out even modest evidence of punitive processing bias. Policy implications: This paper highlights the importance of revisiting the evidence of processing bias within juvenile justice and child welfare agencies. Given the decentralized and continuously evolving nature of these systems, local jurisdictions should investigate their own case outcomes and contexts before implementing reforms to address bias. Yet, many lack the resources for such research and federal support is essential to enhance local data analysis capabilities, promoting more tailored and effective policy reforms. Initiatives that aim to integrate data from multiple systems can better understand and address the needs of overlapping populations, ultimately improving the quality of services and outcomes. Any practitioner in the youth-serving field knows that connection between juvenile justice and child welfare. You see the histories and what has led to their juvenile justice system involvement. At the same time, traditionally, what we have seen is that juvenile justice and child welfare and education and behavioral health are operating in an individual way, not recognizing that these are the same kids, the same families who are touching these systems. —Michael Umpierre Director Center for Juvenile Justice Reform August 2021 Contact with the child welfare system—the collection of government agencies and private partnerships tasked with investigating child maltreatment and providing services to affected children and their families—is remarkably common in the United States. By age 18, more than one-third of children will have been subject to a formal investigation for alleged maltreatment, and 5% of children will have entered formal foster care ( Kim et al., 2017 ; Yi et al., 2023 , 2020 ). Foster care, the temporary placement outside of one’s familial home under the direction of the state or county child welfare authority, is the most intensive intervention provided by child welfare services. 1 Children, and particularly adolescents, placed in foster care tend to have poor average outcomes across behavioral, psychological, and academic domains, raising substantial concerns for policy makers ( Bald et al., 2022 ; Doyle, 2007 ; Gross & Baron, 2022 ; Turney & Wildeman, 2016 ). Of particular concern is the strong association between foster care placement and juvenile justice involvement. Child welfare contact has been highlighted as a key risk factor for juvenile arrest, and that risk is greatest among youth placed in foster care, especially those placed in restrictive settings or experiencing more placements or spells in foster care ( Cutuli et al., 2016 ; Smith et al., 2005 ; Shook et al., 2013 ). 2 In addition to the experiences of child abuse and neglect that typically precede foster care entry, the relationship between foster care and juvenile justice involvement is thought to reflect the trauma of family separation, loss of important and stabilizing social connections, and exposure to inadequate and unstable foster home environments that may increase delinquent behaviors ( Font & Kennedy, 2022 ). Juvenile justice involvement and findings of delinquency capture a combination of delinquent behavior and institutional responses to behavior. Differences in the justice system response can mask or exaggerate differences in the delinquent behavior of foster youth; that is, even in the absence of a direct effect on youth criminal behavior, foster care may exacerbate justice system involvement if youth in foster care face punitive processing bias—differentially harsh treatment by agents of the court. Processing bias is an important concern for youth in foster care as it may exacerbate delinquency system involvement. Juvenile justice system contact can further harm their social and economic prospects, given that juvenile records are not fully sealed in many states, and thus can be used in employment screenings and subsequent criminal proceedings. 3 Moreover, receiving a more punitive juvenile system response increases risk for arrest, conviction, and incarceration in adulthood ( Aizer & Doyle, 2015 ; Baron et al., 2023 ; Eren & Mocan, 2021 ). As a result, it is important to understand whether youth in foster care experience differential treatment by the juvenile justice system. This paper provides estimates of differential treatment of youth in foster care by the juvenile justice system. To do so, we assemble a rich administrative data set that links juvenile delinquency referrals and dispositions to prior and active child welfare cases in Pennsylvania, along with detailed histories of behavioral health diagnoses and socioeconomic backgrounds. Leveraging these data, we identify more than 1500 delinquency cases involving youth who were in foster care at the time of their involvement with the justice system. We define processing bias of youth in foster care as more punitive case outcomes rendered by delinquency court officials, conditional on observed case characteristics, compared to youth not in foster care at the time of their delinquency case. Since cases involving youth in foster care are likely to differ due to both observed and unobserved reasons, we use our linked data and a combination of inverse propensity weighting and regression adjustment to implement a selection-on-observables design. Doing so allows us to compare cases involving youth in foster care to a set of cases where the juvenile is not in foster care but has a current or prior confirmed child abuse or neglect case, and is similar along delinquency case characteristics, demographics, behavioral health histories, and their socioeconomic background. Decision-making frameworks developed in the criminology literature suggest processing of juvenile defendants focuses on several focal concerns—the culpability of the juvenile, their potential threat to society, and the court’s constrained set of resources available to reduce the likelihood of recidivism (e.g., court-ordered supervision or services and placement options) ( Bishop et al., 2010 ; Steffensmeier et al., 1998 ). More recently, this literature has also emphasized salvageability, or the extent to which an offender is likely to benefit from alternatives to incarceration or other punitive sanctions ( Galvin & Ulmer, 2022 ). It is, however, ambiguous how concurrent foster care status would affect the court’s decision. On the one hand, youth placed in foster care may face processing bias if out-of-home placement is viewed as a negative signal. Juvenile court officials may impose more severe sanctions when youth are perceived as poor candidates for rehabilitation and likely to recidivate ( Feld, 1999 ), and unstable family circumstances such as foster care can alter the interpretation of existing behavioral problems and increase perceived risk ( Rodriguez et al., 2009 ; Wildeman et al., 2017 ). On the other hand, youth concurrently in foster care may be treated with greater leniency if court officials view foster care placements as a stable alternative living arrangement or if services provided by the child welfare system (e.g., supervision, behavioral health treatment, or placement outside the home) are viewed as interchangeable with court oversight, particularly in jurisdictions where court resources are scarce. 4 In our analyses, we find no evidence that youth in foster care face punitive processing bias. Delinquency cases involving youth in foster care are not more likely to have a charge adjudicated or receive a serious disposition involving posttrial supervision or juvenile placement. If anything, our estimates suggest youth in foster care experience slightly more lenient case outcomes. Cases involving youth in foster care are 5% less likely to have any charge adjudicated and 8% less likely to receive a serious posttrial sanction, such as detention placement. Exploring other elements of the delinquency case yields similar conclusions. Youth in foster care are more than 8% less likely to receive court-ordered supervision at any point during their case and roughly 10% more likely to have any charge withdrawn. Importantly, we fail to find evidence that youth in foster care receive harsh case outcomes that may increase later-in-life criminal justice contact; cases involving youth in foster care are roughly 9% less likely to be placed in detention at any point during the case ( Aizer & Doyle, 2015 ; Baron et al., 2023 ). Since our estimates are derived from a selection-on-observables design, we implement several robustness checks and find our conclusions are not sensitive to alternative matching techniques or sample construction. We also conduct a bounding exercise proposed by Oster (2019) and show our findings are not sensitive to unobserved heterogeneity not captured by our empirical strategy. Given the assumptions of the bounding exercise, we argue that the degree of selection bias required to reverse our conclusions and find evidence of modest punitive processing bias would be substantial. Moreover, when we estimate processing bias along different dimensions of youth characteristics, we find similar patterns regardless of the juvenile’s demographics or charge type and severity. The remainder of the study proceeds as follows. In Section 1 , we provide a brief discussion of the literature on juvenile court decision making, as well as an overview of why the juvenile court may treat cases involving foster youth differentially. We also provide a summary of the juvenile justice system in Pennsylvania and describe its relationship to Pennsylvania’s child welfare system. Section 2 outlines our administrative data sources and defines our analysis sample. Section 3 presents our empirical strategy and probes its main assumptions. Section 4 presents the main findings of the paper, along with several robustness checks, and examines the heterogeneity of the main estimates. Finally, Section 5 provides a brief discussion of how our findings relate to the existing literature and concludes. 1 |. BACKGROUND 1.1 |. Juvenile justice system processing Juvenile justice systems are intended to emphasize rehabilitation more than the adult criminal justice system and seek to avoid punitive or intrusive sanctions unless necessary for public safety. Nevertheless, juvenile justice involvement and being adjudicated delinquent carries a social and legal stigma that can follow youth into adulthood. For instance, though commonly believed to be sealed or expunged, delinquency case records can be used as evidence of a pattern in subsequent criminal cases, inquired about on employment applications, and be accessed by a range of entities, including schools and police, depending on state law. 5 Moreover, punitive case outcomes themselves may directly increase risk for arrest, conviction, and incarceration in adulthood ( Aizer & Doyle, 2015 ; Baron et al., 2023 ; Eren & Mocan, 2021 ). Thus, the decision to adjudicate confers long-term consequences for youth. Earlier research asserted significant disparities in juvenile justice processing by race, ethnicity, and sex ( Leiber & Mack, 2003 ; Leiber & Peck, 2015 ; Love & Morris, 2019 ; Moore & Padavic, 2010 ; Rodriguez, 2010 ; Tracy et al., 2009 ; Zane et al., 2020 ). However, a recent meta-analysis highlights discordance in study conclusions, finding small or no differences by race or ethnicity on a number of outcomes ( Zane & Pupo, 2021 ) and that male youth are more likely to enter a juvenile placement than female youth ( Zane & Pupo, 2023 ). In addition, as overall detention rates have fallen in recent decades, racial disparities in detention have narrowed ( Zane, 2021 ). Nevertheless, given ongoing concerns about the disproportionate impact of juvenile justice processing by race and gender, our study considers variation in processing bias by these traits. The criminology literature argues the decision-making framework of the delinquency court’s processing of juvenile defendants relies on several focal concerns—the culpability of the youth, their threat to their community or society, the constrained set of available options (e.g., availability of services or placement options) for dealing with the case, and the perceived capacity of the offender to benefit from rehabilitative opportunities ( Bishop et al., 2010 ; Galvin & Ulmer, 2022 ; Steffensmeier et al., 1998 ). 6 For processing bias to occur, youth concurrently placed in foster care must be evaluated more harshly based on these focal concerns. However, the way foster care status would affect each concern is ambiguous. Beyond the delinquent act itself, the juvenile court considers factors that may affect the youth’s perceived threat to society, culpability, or salvageability, including their age, the presence of a behavioral or intellectual disability, and history of abuse ( Galvin & Ulmer, 2022 ; Stevenson, 2009 ). Along these characteristics, foster youth differ substantially; they tend to be diagnosed with behavioral disorders and intellectual disabilities at higher rates, and their histories of abuse and neglect are more severe than both the general population and youth with prior child welfare services involvement. Moreover, foster care itself is a legal status known to the court, and, as a result, agents of the juvenile court may have more information about youth in foster care. More thorough documentation requirements for youth in foster care may also mean that prior behavioral issues are better documented or more readily available, creating the impression of a more serious or intractable pattern of antisocial behavior. Additionally, if the label of foster care invokes negative stereotypes, youth in foster care may be perceived as more dangerous and less amenable to rehabilitation; for instance, behavioral issues of foster youth may be attributed to constant—rather than situational or malleable—causes, which may increase the perceived risk for recidivism and reduce optimism about rehabilitation ( Wildeman et al., 2017 ). Alternatively, youth in foster care might be perceived as posing a lesser threat or danger to others and society as a result of their living environment. The youth’s home life, especially whether sufficient supervision is provided, also affects court actors’ assessments of whether the youth is likely to re-offend ( Feld, 1999 ). Prior research suggests that judges are more likely to place youth in out-of-home delinquency settings when their families are perceived as dysfunctional or involved in criminal activity ( Rodriguez et al., 2009 )—reflecting a lack of confidence in their parents’ capacity to reduce the likelihood of recidivism, or potential suspicions about parental endorsement or involvement in illicit behavior. Among the population of youth receiving child welfare services—where parental criminal justice involvement ( Austin, 2016 ), as well as illicit behaviors such as drug use and domestic violence ( Palmer et al., 2022 ), is relatively more common—those remaining with their family of origin may be perceived as a higher risk for recidivism or noncompliance. In other words, foster care status may alleviate perceived threats to the public and concerns about noncompliance with rehabilitation services, as judges can be assured of oversight and supervision provided by foster care settings, especially those placed in residential facilities. 7 Lastly, given limited system capacity and a general preference not to involve youth in the juvenile justice system unless necessary, processing decisions may rest in part on whether ongoing supervision (either formal or informal) is needed to ensure the youth’s rehabilitative needs are met. Youth in foster care are typically provided several services relevant to addressing delinquency—supervision, mental and behavioral health treatment, parenting supports, and, in some cases, placement outside the home. In addition, children in foster care are entitled under federal and state policy to services that address health and educational issues, and often have a case plan that is subject to review by a dependency court judge and the child’s guardian ad litem (advocate in court). These services, coupled with ongoing oversight by the dependency court and child welfare caseworkers, may allow juvenile court judges, particularly where systems are already overburdened, to reduce their caseload without depriving the youth of needed services or supervision. Indeed, both the juvenile justice system and the child welfare system may be motivated to prevent children in foster care from deep-end justice system involvement and maintain the child solely within the custody of the child welfare system. As Sirois (2023) described in a qualitative study of proceedings pertaining to dual systems youth, “neither agency wishes to be blamed for potentially pushing a child further along the ‘pipeline’ to prison”: the child welfare system avoids blame by retaining responsibility, whereas the juvenile justice system avoids responsibility by declining to adjudicate and relying on the child welfare system to provide care and supervision. However, given the substantially limited capacity of the child welfare system to find appropriate foster care placements and services for older youth with severe emotional or behavioral challenges ( U.S. Government Accountability Office, 2018 ), the child welfare system may prefer to transfer responsibility for “hard to place” youth to the juvenile justice system. Altogether, it is unclear if, and in what direction, juvenile courts would differentially treat cases involving youth in foster care. However, a small empirical literature has found cases involving concurrently dependent youth tend to receive more punitive case outcomes. Studying processing bias in New York City during the late 1990s, Conger and Ross (2001) found youth in foster care were more likely to be detained pretrial than those not in care. Using data from Cook and Cuyohoga counties in the early 2010s, Herz et al. (2019) observed youth with dual system contact—including those with foster placement histories—were more likely to be detained prior to adjudication than youth involved only in the juvenile court system, and concurrent involvement in both systems further increased the likelihood of preadjudication detention. Two studies of first-time offenders in Los Angeles County, California, during the early 2000s similarly found evidence of processing bias against dependent youth ( Ryan et al., 2007 ) and youth with any child welfare history ( Tam et al., 2016 ). Specifically, Ryan et al. (2007) found that, though child welfare system involved youth did not differ in rates of case dismissal, they were less likely to receive probation and more likely to be placed in a secure, juvenile justice managed facility than youth not concurrently involved with the child welfare system. Conversely, Tam et al. (2016) found no difference in processing outcomes between justice involved only youth and child welfare involved youth without a history of placement. The authors did find child welfare involved youth with such a history were more likely to receive a placement sentence relative to probation, though rates of dismissal or corrections sentences did not differ. As a whole, these prior studies support the finding that child welfare involvement—including a mix of substantiated maltreatment, in-home services, and foster care placement—is primarily associated with a bias toward placement (either prior to or postadjudication), but not with differential rates of dismissal. Of note, these prior studies relied on comparison groups of youth without involvement in the child welfare system, varied substantially in terms of how child welfare system involvement was measured (ever vs. current, foster care specifically vs. dependency or any victimization), and focused primarily on first-time juvenile offenders. We instead focus on comparing youth in foster care at the time of their juvenile justice referral to youth with current noncustodial child welfare involvement or a history of child welfare involvement (custodial or not). We argue that our comparison is more appropriate for two reasons: (1) the most relevant counterfactual to foster care is noncustodial child welfare involvement rather than no child welfare contact, and (2) the comparison group in prior research was likely predominantly composed of youth with child welfare histories, but that history was not able to be measured and considered in the modeling strategy. For instance, population-level studies of California youth find that more than 60% of arrested youth have been referred to child welfare previously ( Eastman et al., 2019 ), with justice-involved youth typically referred multiple times for both abuse and neglect ( Eastman et al., 2024 ). Despite similar conclusions among previous studies, there are several reasons to revisit the nature and scope of processing bias. Both the foster care and juvenile justice populations vary significantly across jurisdictions, and both systems have significantly scaled down the use of residential placement services, especially in major metropolitan areas, since the study periods previously documented in the literature. 8 For instance, juvenile crime has fallen steadily since its peak in the mid-1990s ( Puzzanchera, 2022 ), and delinquency courts have shifted toward a greater emphasis on diversion and away from detention ( Monahan et al., 2015 ; Sickmund et al., 2021 ). Given the ambiguous predictions of juvenile processing frameworks and the broad changes in both the foster care and juvenile justice landscapes over time, this study contributes to the research on processing bias in a recent context that allows for the construction of a richly linked administrative data set along with a selection-on-observables design. 1.2 |. The child welfare and juvenile justice systems in Pennsylvania Pennsylvania has a county-administered child welfare system comprising two tracks: child protective services (CPS) and general protective services (GPS). CPS is an investigation track that responds to allegations that meet the state’s definition of child abuse, largely physical and sexual abuse. GPS is a family assessment track that responds to a range of child and family concerns that are not designated as child abuse. GPS cases span a wide range of concerns, including child neglect, parental substance abuse, and domestic violence. For adolescents, a substantial share of GPS cases involves child behavior problems, truancy, and parent-child conflict. 9 In 2021, ChildLine sent approximately 38,000 CPS and 161,000 GPS referrals to counties. Despite the differences in the allegations designated as CPS and GPS, both types of cases can be opened for ongoing services, form the basis of a court petition, or result in foster care placements. In many counties, they are investigated by the same caseworkers, and services are provided by the same contractors. If the court orders foster care placement, placement priorities follow federal guidelines, whereby kinship care (placement with relatives) is most preferred, followed by nonrelative family foster. Congregate care—placement in either a group home or residential facility—is least preferred. Compared to the nation in 2019, Pennsylvania had similar rates of foster care placement (United States 5.6 per 1000; Pennsylvania 5.7) and juvenile detention (United States 129 per 100,000; Pennsylvania 114). 10 Pennsylvania’s Juvenile Act (42 Pa.C.S.A. §6301) divides juvenile justice responsibilities into six divisions: juvenile courts, court administration, juvenile probation, detention, state-operated facilities, and private providers. Multiple state agencies or organizations provide guidance on the implementation of state juvenile statutes: the Juvenile Court Judges’ Commission (JCJC) advises juvenile court judges on the care of dependent or delinquent youth and maintains collection and reporting of state juvenile court statistics; the Department of Human Services oversees the state’s Office of Children, Youth, and Families, operates the state’s delinquency facilities, and licenses local or private facilities for juveniles; the Pennsylvania Commission on Crime and Delinquency provides oversight on system-wide planning, coordination, and policy analysis; and the Pennsylvania Council of Chief Juvenile Probation Officers collaborates with JCJC on probation-focused policy, planning, and training ( Pennsylvania Juvenile Court Judges Commission, 2018 ). While state-wide agencies provide leadership and guidance, cases involving juvenile offenders are primarily handled by local court jurisdictions responsible for administering the juvenile court, probation, and delinquency facilities. The juvenile court case processing system in Pennsylvania is similar to other states. After an arrest, a juvenile may either be released to a guardian or placed in detention while awaiting probation intake. During intake, a case may be dismissed, receive informal supervision, have a juvenile petition filed, or be transferred to criminal court. Diversion is a common practice in Pennsylvania. Instead of filing a petition for formal court processing, a juvenile may enter into an informal adjustment period, which typically spans no more than six months of informal supervision. Successful completion of this supervision program may result in the case’s dismissal. If a formal petition is filed, the court may suspend formal processing and place the juvenile into a preadjudicatory probation period known as a consent decree, which may end in dismissal or adjudication. Approximately half of new delinquency cases in Pennsylvania result in preadjudicatory diversion ( Pennsylvania Juvenile Court Judges Commission, 2021 ). If diversion is unsuccessful, an adjudicatory hearing will take place. Youth who are adjudicated delinquent may then be subject to formal probation supervision or placement in a delinquency-based residential facility ( Pennsylvania Council of Chief Juvenile Probation Officers, 2020 ). The majority of adjudicated delinquent cases result in probation, although approximately 5% of delinquency cases lead to placement in out-of-home delinquency facilities ( Pennsylvania Juvenile Court Judges Commission, 2021 ). 2 |. DATA Our analysis is primarily based on data from Pennsylvania’s Child Welfare Information Solution (CWIS), a statewide data system that aggregates child welfare records from each of the 67 counties. The data consist of all confirmed investigations conducted through the state’s county-administered child welfare agencies between January 2015 and December 2020. Investigations relate to allegations of abuse and neglect, as well as other family concerns, which are reported to the statewide ChildLine, a 24-h hotline. CWIS includes details of each investigation, such as the date of the reported allegation, the county that handled the report, the alleged types of maltreatment, and dates of dispositions and services. Given the data also include the location of the alleged victim, we collect census tract details where the alleged maltreatment took place from the American Community Survey to proxy for neighborhood characteristics and socioeconomic backgrounds. Attributes of children named on investigations include their race, gender, age, whether they were the primary victim listed in the report (as opposed to a sibling), and personal identifying information used to facilitate probabilistic linking between administrative records. We supplement the CWIS data system with records from three administrative sources. First, we collect out-of-home placement from the state’s Adoption and Foster Care Analysis and Reporting System. The Adoption and Foster Care Analysis and Reporting System data provide information on out-of-home services, including dates of foster care placement, the type of placement setting (e.g., nonrelative foster care, kinship care, or congregate care), the length of the full out-of-home episode, and reasons for exiting care. Second, juvenile justice involvement is identified through Pennsylvania JCJC’s data system. The juvenile court data include all referrals made to juvenile courts between 2015 and 2019 and contain key details of each juvenile court petition, such as the date on which the youth was referred to the court, the county system that handled the delinquency case, and the ultimate disposition of each charge brought against the youth. Importantly, the charges are detailed enough to observe why a juvenile was referred to the court and include key characteristics of the case brought to the court, such as whether an arrest was made, the specific charges against a youth, and the severity of each charge. The data also contain information on sanctions levied by the court, such as if the judgment included a fine, court-ordered supervision, or pre- and posttrial delinquency placement. Finally, considering the high prevalence of emotional or behavioral disorders among youth in foster care, with more than 40% having prior diagnoses ( Palmer et al., 2024 ), and the fact that a significant proportion of youth in juvenile detention meet the diagnostic criteria for behavior disorders ( Teplin et al., 2002 ), we incorporate Medicaid claims data from Pennsylvania’s Office of Medical Assistance Programs. The claims data include information such as the type of medical claim (i.e., inpatient, outpatient, or professional office), the date of the claim, and the diagnoses associated with the visit. This additional data allow us to assess whether individuals were diagnosed with any behavior disorder before age 10, specifically by examining claims associated with anxiety, adjustment, mood, or disruptive conduct disorders. We employ a probabilistic matching algorithm to establish links between the CWIS data and each supplemental administrative source. Since the data do not contain a common identifier, records are linked using a Fellegi–Sunter model. 11 Specifically, the model creates links between sources by calculating the probability that records across separate administrative sources belong to the same individual. To do so, the model searches for similarities and minimizes discrepancies across available identifying information such as name, gender, date of birth, and other administrative identifiers (e.g., social security number or the state’s Department of Public Welfare number). We then define an individual as a match if their calculated probability is above a threshold chosen to minimize the rate of false positive links. Overall, the probabilistic match performs well; for instance, 94% of children identified in foster care are matched to the child welfare records. 12 To validate the accuracy of the matching process, we randomly selected and manually examined 100 records, comparing the identifying information from administrative sources. Only two records exhibited discrepancies in more than three matching fields. 2.1 |. Analysis sample The basis of our analysis sample consists of all juvenile justice cases between 2015 and 2019 involving youth. We restrict our sample to youth who came into contact with the juvenile justice system after age 10, but before age 18. Given our focus on processing bias for youth in foster care at the time of their delinquency case, we further limit the sample to the set of youth who were the confirmed victims of child maltreatment before coming into contact with the justice system. Altogether, the analysis sample consists of 13,829 delinquency cases involving 10,931 youth. Of these cases, 1664 involve a juvenile defendant currently in foster care. 13 Figure 1 provides an overview of the characteristics of juvenile justice cases included in the analysis sample. The majority of episodes involve youth aged 15–16. While a significant proportion of cases resolve quickly, lasting only a couple of weeks, the median length of a case is approximately 260 days. In most cases, there are one or two charges brought against juvenile defendants, with the most common reason being a violent offense. The figure also highlights notable distinctions between cases involving youth in foster care and not. Specifically, cases with concurrently placed youth often involve older juvenile defendants, have a higher number of charges per case and are more likely to have the most serious charge be related to a violent offense. FIGURE 1. Open in a new tab Descriptive characteristics of delinquency cases. Note : The figure presents the distributions of delinquency case characteristics, comparing cases involving youth in foster care (Current FC) with those not in foster care (No FC). Panel (a) presents the distribution of age at the time of the juvenile’s referral. Panel (b) displays the total length of the case in days. Panels (c) and (d) present the distribution of the number of charges brought against the juvenile and the most serious charge in the case, respectively. Youth who are concurrently placed in foster care when they commit offenses come into contact with the child welfare system for various reasons and have diverse experiences within the system prior to their offense. In Figure 2 , we present their characteristics, including common reasons for child welfare contact, their typical placement settings at the time of offending, and the distribution of time between their out-of-home placement and their involvement with the juvenile justice system. Concurrently placed juvenile offenders are predominantly placed in congregate care settings, such as institutions or group homes, and they have histories within the child welfare system that encompass instances of abuse, neglect, and, most commonly, interventions related to their behavioral issues such as substance use, truancy, and disruptive episodes. While an outsized share of youth offend shortly after their placement in foster care, the median adolescent has been in foster care for nearly one year by the time of their offense. FIGURE 2. Open in a new tab Descriptive characteristics of foster care placements. Note : The figure plots the distributions of foster care (FC) placement characteristics for youth who were in foster care during their involvement with the juvenile justice system. Panel (a) presents the most prevalent and non-mutually exclusive child maltreatment allegations associated with these youth. Panel (b) shows the most common placement settings for youth at the time of their delinquency case. Finally, Panel (c) presents the distribution of time between the youth’s foster care placement date and the initiation of their case. 3 |. EMPIRICAL STRATEGY Identifying processing bias involves comparing the case outcomes of youth in foster care at the time of their contact with the justice system to those who were not in care. However, simply comparing outcomes between these two groups is not enough to produce valid estimates of processing bias since the youth in foster care are likely to differ from their counterparts in important ways that are both observable and unobservable. To obtain more accurate estimates of the impact of being in foster care during justice system contact, this paper adopts a selection-on-observables approach that combines exact matching and inverse propensity weights (IPWs). This approach enables us to form a comparison group similar to the youth in foster care along demographics and the most legally relevant attributes: initial juvenile justice charges and juvenile justice history, mental and behavioral health history, and previous child welfare system contact. The analysis sample, as described above, contains 1664 juvenile justice cases (1453 unique youth) involving a juvenile offender concurrently placed in foster care and 12,165 cases (9788 unique youth), where a juvenile was not in placement but has either past or ongoing child welfare services involvement. We generate exactly matched groups based on the race and gender of the youth, their age at the time of justice system contact, and the year and county in which the case was filed. Nearly all (1585) cases involving concurrently placed youth appear in an exactly matched group with variation—a group with at least one case with a concurrently placed youth and at least one comparison case. Of the cases where the youth was not in foster care, 8556 appear in an exactly matched group and form the basis of our control group. The full set of 10,141 observations that appear in an exactly matched group with variation comprises the matched sample, which we use to estimate the predicted probability of being concurrently placed in foster care at the time of juvenile justice contact. To do so, we estimate a logit model with the dependent variable as an indicator for being placed in foster care during the delinquency case and predictive variables, including a wide range of delinquency case and prejustice system contact characteristics. For instance, the model includes controls for the age and race of the alleged delinquent youth, their prior history of child welfare contact (including their histories of specific maltreatment allegations), characteristics of their current delinquency case and prior juvenile justice contact, and whether the youth had any diagnosed behavioral disorder before age 10. 14 To calculate an estimate of processing bias, we estimate the following model: Y i = δ CurrentF C i + X i ′ 𝛃 + ε i , (1) where Yi represents the outcome of interest, such as whether any charges were adjudicated in delinquent case i or if the case resulted in supervision or placement. CurrentFC i is an indicator variable set to 1 if the case involves a youth concurrently placed in foster care. Finally, the vector X i is the set of covariates used to estimate the predicted propensity of being concurrently placed in foster care at the time of juvenile justice contact. Standard errors are two-way clustered at the child and exactly matched group levels to account for the mechanical correlation that arises when the same youth has multiple cases and correlation in outcomes within exactly matched groups. The coefficient of interest is δ , which represents the difference in outcomes between cases where the youth was concurrently placed in foster care and observationally equivalent control cases where the youth was not in foster care, but had a previous or ongoing interaction with child welfare services. 15 To implement IPWs, control cases are weighted by p ^ 1 − p ^ , where p ^ is the predicted probability, and cases where the youth is concurrently in foster care are given a weight of one. By implementing IPWs and postmatching regression adjustment, the estimator δ ^ is said to be a doubly robust, two-step estimator of δ . Estimates of the effect of concurrent foster care placement on juvenile case outcomes will be unbiased if either the underlying matching model or the regression model is specified correctly ( Arkhangelsky & Imbens, 2022 ; Imbens & Wooldridge, 2009 ). Our selection-on-observables approach assumes that cases where a youth was in foster care are indistinguishable from cases where they were not in terms of relevant case characteristics, given the observable factors included in the exact matching procedure. The primary threat to the analysis is that there are some unobserved characteristics jointly correlated with delinquency case outcomes and concurrent foster care placement. For instance, youth in foster care may be more likely to be charged or referred to juvenile justice for relatively minor infractions while under the supervision of licensed caregivers, and those cases may be treated with greater leniency. While we cannot definitively rule out such unobserved differences, we show below that the design generates a comparison group that closely matches the characteristics of concurrently placed youth. In addition, we perform several robustness checks meant to refocus the exactly matched comparison group along case characteristics and a bounding exercise that suggests omitted variable bias is unlikely to affect the conclusions of the analysis. 3.1 |. Balance Before matching, there are notable differences between the cases involving youth in foster care and those without. As shown in the first three columns of Table 1 , cases that involve youth in foster care differ significantly in terms of demographics, delinquency case characteristics and histories, maltreatment histories, and behavioral health diagnoses before age 10. For example, in comparison to cases where the youth was not in care, cases with youth concurrently in foster care were 36% more likely to be Black, 21% more likely to be charged with a felony, and 50% more likely to have a prior violent charge and a history of physical abuse. TABLE 1. Sample balance. Unmatched Matched difference No FC Current FC Difference Panel A: Demographics White 0.65 0.54 −0.108*** 0.000 [0.48] [0.50] (0.013) (0.000) Black 0.33 0.44 0.117*** 0.000 [0.47] [0.50] (0.012) (0.000) Hispanic 0.11 0.12 0.001 0.000 [0.32] [0.32] (0.008) (0.000) Male 0.67 0.6 −0.066*** 0.000 [0.47] [0.49] (0.012) (0.000) Age at JJ referral 15.23 15.51 0.276*** 0.012 [1.80] [1.67] (0.047) (0.048) Panel B: Delinquency Most serious charge is felony 0.37 0.46 0.086*** −0.002 [0.48] [0.50] (0.013) (0.015) Most serious charge is misdemeanor 0.49 0.42 −0.070*** 0.004 [0.50] [0.49] (0.013) (0.015) Violent charge on referral 0.37 0.46 0.095*** −0.000 [0.48] [0.50] (0.013) (0.016) History of violent charge 0.09 0.14 0.055*** 0.000 [0.28] [0.35] (0.008) (0.012) History of history of drug or alcohol charge 0.03 0.02 −0.011** −0.000 [0.18] [0.14] (0.005) (0.004) History of weapon charge 0.02 0.04 0.012*** −0.001 [0.15] [0.19] (0.004) (0.004) History of adjudicated charge 0.07 0.07 0.000 −0.002 [0.25] [0.25] (0.007) (0.008) History of felony charge 0.11 0.16 0.045*** −0.010 [0.31] [0.36] (0.008) (0.012) Panel C: Child Welfare History Investigation within 1 year 0.58 0.58 −0.001 −0.009 [0.49] [0.49] (0.013) (0.015) Physical abuse 0.04 0.06 0.022*** −0.003 [0.20] [0.24] (0.005) (0.009) Child in need of services 0.59 0.65 0.054*** −0.014 [0.49] [0.48] (0.013) (0.016) Parent substance abuse 0.18 0.17 −0.006 −0.001 [0.38] [0.38] (0.010) (0.013) Lack of caregiver 0.05 0.14 0.090*** −0.003 [0.21] [0.34] (0.006) (0.012) Inadequate supervision 0.09 0.11 0.026*** −0.007 [0.28] [0.32] (0.008) (0.010) Unmet material need 0.17 0.17 0.007 −0.009 [0.37] [0.38] (0.010) (0.013) Panel D: Mental health diagnosis before age 10 Emotional disturbance (broad) 0.51 0.54 0.039*** 0.005 [0.50] [0.50] (0.013) (0.016) Adjustment disorders 0.19 0.21 0.020** −0.003 [0.40] [0.41] (0.010) (0.014) Anxiety disorders 0.04 0.04 −0.000 0.000 [0.19] [0.19] (0.005) (0.007) Disruptive disorders 0.25 0.27 0.020* −0.003 [0.44] [0.45] (0.011) (0.016) Mood disorders 0.14 0.17 0.037*** 0.002 [0.34] [0.38] (0.009) (0.014) F -statistic from joint test 0.335 Observations 12,165 1,664 13,829 10,141 Open in a new tab Note : The table presents summary statistics and sample balance. The first column consists of the 12,165 juvenile justice cases involving youth not concurrently placed in foster care (No FC) at the time of juvenile justice contact, while the second column consists of 1664 cases involving youth in foster care (Current FC) at the time of the delinquency court referral. Column 3 presents point estimates of a difference in means between the groups displayed in the first two columns. Column 4 similarly presents a difference in means between the two groups, but after implementing inverse propensity weights (IPWs) and limiting the sample to the 10,141 observations within exactly matched groups with variation. Standard deviations are displayed in square brackets, while standard errors (two-way clustered at the youth and exactly matched group levels) are shown in parentheses. The estimated propensity scores ( p ^ ) for cases with youth concurrently in foster care and those without are shown in Figure 3 . The figure illustrates that while the two distributions share a common support, they differ significantly. However, when focusing on control units in exactly matched groups and implementing IPWs, the difference in distributions becomes minimal. This is supported by the final column of Table 1 , which shows that the differences in characteristics between matched cases with youth concurrently in foster care and control units are small and precisely estimated. Furthermore, an F -test of joint significance from a regression of being concurrently placed in foster care on these characteristics indicates that the covariates are not jointly predictive of concurrent foster care placement. Taken together, the distribution of propensity scores and the negligible differences presented in the table suggest that the selection-on-observables design produces an analysis sample that is well balanced along observable characteristics. FIGURE 3. Open in a new tab Distribution of propensity scores. Note : The figure displays the distribution of propensity scores across three groups. The solid blue line displays the propensity score distribution for youth concurrently in foster care (Current FC) at the time of juvenile justice contact. The gray dashed line plots the propensity score distribution for children not in foster care (No FC) at the time of juvenile justice contact, while the dotted black line plots the distribution of the propensity scores after implementing inverse propensity weights. 4 |. RESULTS Table 2 presents the main results. Each column of the table presents estimates from separate regressions. The first column presents a difference in means between cases involving youth currently in foster care at the time of juvenile justice contact and those with prior child welfare system contact but were not concurrently in out-of-home care. These simple differences contrast those of the prevailing literature; cases involving concurrently placed youth are somewhat less likely to be adjudicated (Panel A) or result in a serious disposition involving posttrial supervision or delinquency placement (Panel B). TABLE 2. Main estimates of processing bias. (1) (2) (3) (4) Panel A: Any adjudicated charge Current FC −0.024 −0.021* −0.024 −0.021* (0.015) (0.013) (0.015) (0.013) Observations 13,829 13,829 10,141 10,141 R 2 0.000 0.003 0.179 0.404 Control mean 0.360 0.391 0.374 0.374 Percent effect −5.278 −12.788 −6.417 −5.615 Panel B: Serious disposition Current FC −0.023* −0.060*** −0.035** −0.032** (0.012) (0.015) (0.015) (0.013) Observations 13,829 13,829 10,141 10,141 R 2 0.000 0.004 0.180 0.396 Control mean 0.365 0.402 0.383 0.383 Percent effect −6.301 −14.925 −9.138 −8.355 IPWs Y Y Y Groups with variation Y Y Controls Y Open in a new tab Note : The table presents the main estimates of processing bias for two key outcomes: having any charge adjudicated and having a punitive disposition (court-ordered supervision or juvenile delinquency placement), in panels A and B, respectively. The first row of each panel shows estimates of δ from Equation 1 . Standard errors are two-way clustered at the youth and exactly matched group levels and are shown in parentheses. Below the point estimates, the table displays the amount of observations in each model, the R 2 of the model, the control mean—the average value of the dependent variable among youth not in foster care at the beginning of their juvenile justice case—and the estimate as a percent of the control mean. The first column presents a difference in means between cases where the youth was concurrently placed in foster care (Current FC) and cases where the youth was not in out-of-home care. Column 2 implements inverse probability weighting—weighing control units by p ^ 1 − p ^ . Column 3 restricts the sample to cases within an exactly matched group with variation. Finally, the last column, our preferred specification, implements postmatching regression adjustment by including the set of covariates used to compute propensity scores. Abbreviation: IPWs, inverse propensity weights. Columns 2 through 4 further restrict the comparison: Column 2 applies the inverse probability weights calculated above; Column 3 shows the differences in means while restricting the sample to exactly matched groups with variation in treatment; Column 4, our preferred specification, additionally incorporates postmatching regression adjustment. Each of the estimates presented in the table share a similar conclusion: contrary to prior studies of processing bias, there is no evidence that cases involving youth in foster care at the time of juvenile justice contact are more likely to experience more punitive case outcomes. In fact, the estimates suggest the opposite; cases involving concurrently placed youth were 2 percentage points less likely to have any charge adjudicated (5.6%) and 3 percentage points less likely to have a serious disposition (8.3%). Table 3 probes other typical case outcomes for evidence of processing bias. Columns 1 and 2 reprint the estimates of the effects of having any adjudicated charge and having a serious disposition from Column 4 of Table 2 , respectively. The remainder of the table presents results from separate regressions that implement the same specification to examine differences in other important case outcomes involving concurrently placed youth. Estimates from those regressions are consistent with the results thus far; specifically, cases involving youth in foster care were more likely to have a charge withdrawn, less likely to have been placed in pretrial juvenile detention, and less likely to have been placed under court-ordered supervision at any point during the case. We present p -values from one-sided tests that youth placed in foster care face more punitive case outcomes and reject the null in all cases. TABLE 3. Estimates of case outcomes. (1) (2) (3) (4) (5) (6) Any adjudicated charge Serious disposition Any charge withdrawn Any court supervision Pretrial detention Postadjudication detention Current FC −0.021* −0.033** 0.029** −0.032** −0.021* −0.019 (0.013) (0.013) (0.014) (0.013) (0.012) (0.012) Observations 10,142 10,142 10,142 10,142 10,142 10,142 R 2 0.406 0.396 0.308 0.377 0.327 0.326 Control mean 0.375 0.384 0.268 0.360 0.224 0.229 Percent effect −5.600 −8.594 10.821 −8.889 −9.375 −8.297 One-sided test 0.050 0.007 0.015 0.008 0.037 0.053 Open in a new tab Note : The table presents estimates of processing bias for youth concurrently placed in foster care (Current FC) during their delinquency case across several case outcomes. The first two columns display estimates of processing bias in having any charge adjudicated and having any serious disposition in the case, respectively, while implementing inverse propensity weights (IPWs) and postmatching regression adjustment. Columns 3 and 4 estimate processing bias in having any charge withdrawn or being placed under court-ordered supervision at any point of the delinquency case. Lastly, the table presents estimates of differences in juvenile detention, either before an adjudication decision (Column 5) or after the youth is adjudicated (Column 6). Standard errors are two-way clustered at the youth and exactly matched group levels and are shown in parentheses. We examine the sensitivity of the baseline estimates to various sample restrictions in Table A2 . As highlighted above, we construct the analysis sample by examining cases that involve youth with any prior history of child welfare system contact. Since foster care is a temporary arrangement, one may be concerned that youth concurrently in foster care at the time of juvenile justice contact have more recent experiences with child welfare than those with any prior history and that recent contact may influence the court’s decision making in unobserved ways. To address this concern, the first column limits the sample to cases where the youth had prior child welfare system contact within one year of their court referral date. A related concern is that, given the close relationship between the juvenile justice and child welfare systems, there may be a mechanical hand-off between the two systems (i.e., youth may come into contact with juvenile justice due to their contact with child welfare and vice versa). This concern is compounded by Panel (d) of Figure 1 , which shows a disproportionate amount of youth placed in foster care are placed within one month of juvenile justice contact. Column 2 addresses this issue by omitting juvenile cases that begin within a 90-day window of out-of-home placement. Finally, since large urban areas represent an out-sized share of the data, Columns 3 and 4 omit Philadelphia and Allegheny (home to Pittsburgh) counties, respectively. Across all specifications, the estimates provide the same conclusions as the baseline. If anything, the estimates are larger in magnitude, though not statistically different from our main findings. Finally, Table A3 probes the sensitivity of the estimates to alternative matching techniques. In the first column, we instead implement a kernel matching procedure with an Epanechnikov function. Columns 2 and 3 instead use the estimated propensity scores to apply nearest neighbor matching for one and three neighbors, respectively. Finally, in Column 4, we construct more stringent match groups that additionally match on the characteristics of the individual juvenile case. These case characteristics include if the case involved any violent, drug-related, or weapon-related charge, if the most serious charge was a felony, and if the juvenile had a history of felony charges or any history of adjudicated charges. While the estimates are smaller in magnitude than the baseline estimates, the differences are not statistically distinguishable. 4.1 |. Bounding Although our model of foster care propensity includes a large set of predictive controls, the assumptions of our empirical approach mean we cannot definitively rule out that unobserved characteristics of youth concurrently placed in foster care may bias our estimates away from finding evidence of processing bias. For instance, while we observe the charges listed on the court referral, we do not observe written descriptions or aspects of the case that may warrant differential treatment by the court and be correlated with placement out of the home. Such correlation may be particularly concerning if youth in foster care are more likely to be charged or referred to juvenile justice for relatively minor infractions due to higher levels of surveillance. For example, a physical altercation with a peer in a group home may be referred to juvenile justice whereas a fight between siblings within one’s familial home would not. In this case, even if youth in foster care are brought to the attention of juvenile justice for the same type and level of offense (e.g., simple assault), potential outcomes of their cases may differ due to unobserved differences in the severity of their offenses, rather than due to differential treatment based on foster care status. To test how sensitive the estimates may be to unobserved heterogeneity, we perform two complementary diagnostics proposed by Oster (2019) . Following earlier work by Altonji et al. (2005) , the Oster (2019) exercises require a proportional selection assumption—that the selection on unobservables is proportional to observables. The first estimates a bias-adjusted effect, which adjusts the baseline estimate given some assumed degree of selection on unobservables. The bias-adjusted effect allows us to ask how much would the baseline estimates change for some assumed amount of unobserved selection bias. As a result, Oster (2019) interprets the bias-adjusted estimate as a bound on the true effect of interest. The second diagnostic instead examines how severe the degree of selection on unobservables (relative to observables) would need to be to reduce the effect to zero. If a substantial degree of selection on unobservables is required to push the effect to zero, then the effect can be considered robust to omitted variable bias. The results of these diagnostics are shown in Table 4 for having any adjudicated charge in Panel A and having a serious disposition in Panel B. The first column shows our baseline estimate, while the second column reports the bias-adjusted effect. The final two columns show the amount of proportional selection needed for the effect to equal zero. In the third column, we follow Oster (2019) and assume that the maximum R 2 from a hypothetical regression of the outcome on the observable and unobservable characteristics would be equal to one and a third times the R 2 from the baseline regression. In the final column, we take a more conservative approach suggested by Altonji et al. (2005) and instead assume a maximum R 2 of 1. TABLE 4. Sensitivity analysis. Proportional Degree of selection Baseline Bias-adjusted effect Oster Altonji Panel A: Any adjudicated charge Current FC −0.021 −0.018 7.398 1.51 Panel B: Serious disposition Current FC −0.032 −0.03 15.82 3.126 Open in a new tab Note : The table presents the results of the sensitivity checks proposed in Oster (2019) . Panels A and B report the results for any adjudicated charge and having any serious disposition, respectively. The first column reports the baseline estimate of the effect of being in out-of-home care at the time of the juvenile justice referral as in the final column of Table 2 . Column 2 displays the bias-adjusted effect, assuming the amount of selection on unobservables is equal in proportion to the amount of selection on observables. Following Oster (2019) , we assume that the maximum R-squared from a hypothetical model, which includes both observed and unobserved factors is 1.3 times the R 2 from the preferred baseline model. The final column instead follows Altonji et al. (2005) and assumes the R 2 from a hypothetical model, which includes both observed and unobserved factors is 1. The estimates of the effects of concurrent foster care placement do not appear to be driven by omitted variable bias. The effects on both the presence of any adjudicated charge and the seriousness of disposition are tightly bounded, ranging from −2.1 to −1.8 percentage points and −3.2 to −3 percentage points, respectively. Moreover, the degree of selection on unobservables required to reduce the estimates to zero is substantial. To reduce the adjudication and serious disposition estimates to zero, the exercise suggests that unobservables would need to be roughly 7 and 15 times as informative as observables, respectively. Though it is not obvious how relatively important observed versus unobserved variation can be in explaining away estimates, prior work offers a rule of thumb to formalize a test of the degree of proportional selection. Both Oster (2019) and Altonji et al. (2005) suggest that values larger than 1 imply selection on unobservables would need to be larger than the amount of selection on observables, and observed effects are less likely to be driven by omitted variable bias. Beyond improving confidence in the baseline estimates, the results of these bounding exercises provide further evidence that processing bias is unlikely to be important in our context. Since differences between the baseline estimates and the bias-adjusted effects are small, we can rule out the potential for unobserved differences that may mask evidence of processing bias. Moreover, given that the proportional degree of selection exercise estimates the amount of selection on unobservables required for the effect to be zero, the degree of selection would need to be even larger to reverse the direction of the baseline estimates and provide evidence of processing bias. 4.2 |. Heterogeneity While our baseline estimates imply cases involving youth concurrently in foster care are less likely to result in formal adjudication or court-ordered placement and supervision, one may worry that the average effect masks important subgroup heterogeneity. As indicated, prior research on juvenile justice processing observed disparities by race, ethnicity, and gender, particularly at times in the process where there is more discretion ( Leiber & Mack, 2003 ; Love & Morris, 2019 ; Rodriguez, 2010 ; Zane et al., 2020 ). These effects tend to be most pronounced for Black and Hispanic youth ( Rodriguez, 2010 ; Zane et al., 2020 ). Nonetheless, there is minimal research that has examined heterogeneity in how youth involved with child welfare are processed through the juvenile justice system ( Tam et al., 2016 ). This may be particularly important for the population of cases involving youth concurrently placed in foster care given their demographic composition and differing reasons for coming into contact with the juvenile court system, and prior research asserting that child welfare involvement contributes to the overrepresentation of Black children in juvenile justice systems ( Ryan et al., 2007 ). Figure 4 explores heterogeneity across demographic characteristics of delinquent youth. Specifically, we examine subgroups of the analysis sample by gender, race, and if the youth had a previous behavioral health diagnosis. The figure shows that we fail to identify a subgroup of concurrently placed youth that experienced more punitive juvenile justice outcomes. If anything, the estimates suggest that boys concurrently placed in foster care and cases involving Black youth in foster care were less likely to have a case result in an adjudicated charge or a serious disposition relative to our baseline estimates, although the magnitude of the estimates is similar. FIGURE 4. Open in a new tab Heterogeneity by youth characteristics. Note : The figure plots estimates of processing bias across subgroups of youth characteristics along with 95% confidence intervals. Left-hand estimates display estimates of processing bias for youth in foster care on having any charge adjudicated, while right-hand estimates do the same for having a serious disposition. Each separate estimate is derived from separate subgroups described below the estimate, and standard errors are two-way clustered at the youth and exactly matched group levels. We also examine whether the baseline estimates of processing bias varied by characteristics of the delinquency case including whether the case included a violent charge, the severity of the case, whether the youth had prior contact with the juvenile justice system, and whether the case involved court-ordered probation. Figure 5 presents estimates of differences in case outcomes for youth concurrently placed in foster care by these subgroups of case characteristics. Regardless of the characteristics of the case, the overall effect appears consistent in direction and magnitude. FIGURE 5. Open in a new tab Heterogeneity by case characteristics. Note : The figure plots estimates of processing bias across subgroups of the youth’s case characteristics along with 95% confidence intervals. Left-hand estimates display estimates of processing bias for youth in foster care on having any charge adjudicated, while right-hand estimates do the same for having a serious disposition. Each separate estimate is derived from separate subgroups described below the estimate, and standard errors are two-way clustered at the youth and exactly matched group levels. Finally, as discussed above, the resources of the juvenile court system are highly constrained, and court officials face the difficult decision of allocating scarce resources for monitoring delinquent youth. As a result, juvenile court officers may view concurrent foster placement as an alternative to expending court resources and rely on the child welfare system for services. Second, and related, is the incapacitation effects of foster placement, which may prevent delinquent youth from violating court-ordered conditions put in place to divert cases from adjudication. 16 In this case, the court may similarly offer preadjudication probation to delinquent youth both in and out of foster care, but increased monitoring from foster parents and child welfare case managers results in greater compliance. Consequently, the lower likelihood of adjudication for youth in foster care may stem from their stronger adherence to court-ordered conditions rather than differential treatment by the court itself. To test this hypothesis, we examine the differences in case outcomes for those offered preadjudication probation in the final column of Figure 5 . However, we find similar estimates to the baseline, suggesting that differences in adherence to court-ordered diversion cannot explain why youth in foster care tend to experience greater leniency in their cases. 5 |. DISCUSSION For decades, media reporting and research linking foster care with increased antisocial behavior culminated in a narrative that youth in foster care were unsupported, unstable, and unsafe—hardly a context for rehabilitation. This narrative was not altogether wrong: the foster care system of the 1990s and early 2000s brought in far more children than it could safely and effectively serve and kept them far too long, failing many ( Font & Gershoff, 2020 ). In recent years, with a smaller and shorter term foster care system, the conclusion is more nuanced: foster care can meaningfully improve children’s life chances ( Bald et al., 2022 ; Baron & Gross, 2022 ; Gross & Baron, 2022 ), and provides more monitoring, mental health services, and other supports than children receive if they remain in the home after a confirmed child welfare case ( Burns et al., 2004 ; Harman et al., 2000 ). Delinquency is more common, and more severe, among adolescents who are abused, neglected, or live within families where illicit activity occurs ( Font & Kennedy, 2022 ), such that the overlap between child welfare and juvenile justice systems is pronounced. The idea of a processing bias—wherein delinquent youth receive harsher or more sustained juvenile justice sanctions on the basis of also having potentially abusive or neglectful family environments—could affect all child welfare systems involved youth, regardless of foster care status. Moreover, foster care, as an explicit legal status, may confer more stigma within the juvenile court than remaining in the home following maltreatment, reflecting a deep skepticism about the quality and capacity of the foster care system. Additionally, it may reflect a reluctance among foster caregivers to provide home-based care for youth with an active probation case, necessitating greater reliance on pre- or postadjudication placements. However, our findings suggest that processing bias was not present when comparing cases involving youth concurrently placed in foster care to similar cases with active child welfare involvement. If anything, our estimates imply youth in foster care experience modestly more lenient case outcomes. What might explain the finding that cases involving youth in foster care tend to experience more lenient case outcomes? First, we reiterate that, across specifications, the coefficients are of modest size and we cannot reject that youth in foster care receive case outcomes similar to the counterfactual set of youth who were not concurrently placed. Nevertheless, the direction and magnitude of coefficients are remarkably consistent with the main findings—providing suggestive evidence of modestly more lenient outcomes for concurrently placed youth. Moreover, leniency for youth in foster care is seen even in the bivariate models; prior to adjusting for covariates, we see no evidence of punitive processing bias. The lack of baseline difference is somewhat surprising: foster care is an intervention of last resort, reserved for children who have experienced significant harm or face imminent safety risks ( Font & Gershoff, 2020 ), and studies consistently find that unadjusted comparisons misattribute or overattribute harms (negative outcomes) to foster care that instead reflect selection bias ( Berger et al., 2015 , 2009 ; Gross & Baron, 2022 ). These findings, on their surface, run counter to previous findings of processing bias. Yet, prior research asserted processing bias largely in the use of detention—either preadjudication ( Herz et al., 2019 ) or postadjudication relative to probation ( Ryan et al., 2007 ; Tam et al., 2016 )—rather than in adjudication, itself, versus dismissal. Our findings, which suggest leniency in adjudication and case withdrawal among youth concurrently placed in foster care, may reflect that rates of adjudication and placement were far higher in prior studies, whereas a majority of cases in our sample were diverted from adjudication and placement, consistent with lower rates as reported in recent studies from other jurisdictions ( Evangelist et al., 2017 ; Thomas et al., 2022 ; Zane, 2021 ). As states have become far more selective about which delinquency cases proceed to adjudication and which youth require secure placement, it is plausible that prior biases have lessened. Juvenile courts may draw on their knowledge of the county child welfare system and the specifics of a youth’s foster care setting when determining the role of the juvenile justice system that is needed to ensure youths’ compliance with conditions and avoidance of recidivism. This individualized approach—which requires consideration of nonlegal factors and thus judicial discretion—may result in dispositional outcomes that are inconsistent with what would be predicted based on legal factors alone ( Painter-Davis & Ulmer, 2020 ). Our main conclusion—similar but modestly more lenient treatment by the juvenile justice system for youth in foster care than youth remaining in their homes—suggests an understanding that foster care services are insufficient to meaningfully replace juvenile justice services. But, in some cases, the increased resources, oversight, and services provided in foster care versus in-home settings may allow youth to avoid higher level engagement with the juvenile justice system despite their delinquent offending. This scenario may signal a form of “institutional offloading” from the juvenile justice system to the child welfare system ( Sirois, 2023 ), which, though stigmatized, is perceived as less punitive for youth ( Asherman-Jusino, 1995 ). We highlight, however, that fewer than a third of our full sample experienced an adjudication or serious disposition in their first juvenile justice referral, consistent with high usage of diversion and deferment options. Yet, a large share of our sample experienced multiple juvenile justice referrals, suggesting that services provided through diversion, even coupled with child welfare services, may be insufficient. This is not surprising: both the child welfare system and the juvenile justice system face considerable constraints in the care and rehabilitation of youth including declining placement options, persistent workforce shortages, and few qualified behavioral health providers available to treat the youth in their care. 17 As states seek to avoid intervention by stigmatizing systems, policy makers are tasked with ensuring that alternatives are available to meet the complex needs of a population that continues to face poor life prospects. 5.1 |. Limitations We note two major limitations with regard to this study. First, our data stem from a single state and span a period that precedes the COVID-19 pandemic. As discussed further below, we cannot be confident of generalizability, given the diversity of state and local child welfare and juvenile justice systems and the vast changes each system has undertaken in the past few decades. However, we note that this limitation similarly applies to all previous research on processing bias, perhaps to a greater extent given that prior research stems from a single county or city. In contrast, our data reflect Pennsylvania’s 67 counties, throughout which the service populations, privatization and service array, local quality assurance mechanisms, and funding for both juvenile justice and child welfare systems vary immensely. Second, our selection-on-observables design cannot definitively rule out that unobserved characteristics of youth concurrently placed in foster care may bias our estimates. Though our approach relies on a large set of predictive controls meant to narrow the scope of omitted variable bias, we do not rely on quasi-random variation in concurrent foster care placement that would allow stronger inference. Nonetheless, we implement a bounding technique to assuage concerns and provide a suggestive calculation of how large omitted variable bias would have to be to drive our estimates toward zero. The exercise suggests that unobserved variation would need to be many times more informative than our rich set of observable characteristics. We also note that our bounding exercise assumes the excluded, unobserved variation attenuates our estimates. Though we cannot unequivocally rule it out, neither research nor practice literature asserts any omitted variable that would upwardly bias our findings (i.e., a variable that is positively associated with punitive juvenile justice outcomes but negatively associated with foster care placement). Instead, consistent with prior research indicating that youth in foster care are, on average, higher risk than their nonplaced peers, we find that youth in foster care were more likely to have current and prior violent and felony charges, have an emotional disturbance diagnosis, and to be involved with child welfare due to their own behavioral problems—all differences one might expect to be associated with more punitive delinquency case outcomes. As a result, the omitted variation in this context would mean our estimates are biased upward, and youth concurrently placed in foster care could be even less likely to have a case adjudicated or face a serious disposition. 5.2 |. Policy implications We view the central empirical contribution of this paper as highlighting that processing bias may not be widespread. Prior research found evidence that youth involved in child welfare services (i.e., an open or ongoing case, including a mix of substantiated allegations, in-home services, or foster care placement) at the time of juvenile justice contact faced some instances of punitive processing bias—including preadjudication detention ( Conger & Ross, 2001 ; Herz et al., 2019 ) or postadjudication placement ( Ryan et al., 2007 ; Tam et al., 2016 )—resulting in higher rates of court-ordered detention, especially for Black youth. However, in recent decades, juvenile courts have reduced the use of detention and placed greater emphasis on diversion ( Monahan et al., 2015 ; Sickmund et al., 2021 ), and increased attention is paid to delinquent youth with histories of child welfare system involvement. 18 In child welfare systems, foster care has similarly become far less common and of shorter duration. Importantly, it was never clear that broad generalizations could be made from studies in a few major urban centers (New York City [ Conger & Ross, 2001 ], Los Angeles [ Ryan et al., 2007 ; Tam et al., 2016 ], or Cook and Cuyohoga counties, including Chicago and Cleveland, respectively [ Herz et al., 2019 ]), communities characterized by entrenched racial and economic segregation, with overwhelmingly nonwhite foster care and juvenile justice populations. Although our findings—of no punitive processing bias in a recent statewide sample—were consistent across urban and rural counties, racial groups, gender, and other individual and contextual factors, we cannot be confident that the findings of this setting generalize well to other institutional contexts. Instead, we highlight the critical importance of revisiting conventional wisdom about system impacts and bias, particularly when systems are heavily decentralized (i.e., state and local variation in practice) and are subject to continuous policy reforms that affect the size and characteristics of the service populations and the nature of system involvement itself. 19 Put differently, just as juvenile justice and child welfare system practices and service populations vary across jurisdictions and throughout time, the prevalence of processing bias is likely to vary as well. Prior to implementing policies aimed at addressing and reducing concerns of bias, local jurisdictions should conduct a thorough examination of their own case outcomes. This examination should focus specifically on identifying evidence of processing bias that may affect youth with concurrent child welfare cases, ensuring that any reforms are tailored to the circumstances, practices, and population served by the local jurisdiction. Yet many states and jurisdictions lack the resources, staffing, or research expertise to conduct rigorous internal research and thus rely on evidence from published works in other settings. Given that the development of appropriate policy reforms must reflect local conditions and needs, given the diversity in institutional practices, more support is needed to bolster local data examination and policy evaluation. Recent federal investments and regulations seek to increase the capacity of state child welfare data systems to provide timely, localized information about populations, processes, and outcomes. 20 Federal support in bolstering states’ capacity to use their own data for quality improvement, evaluation, and resource allocation may be more effective, however, if it facilitates the integration of data from multiple systems that serve overlapping populations. 21 ACKNOWLEDGMENTS We would like to extend our gratitude to Jason Baron, Lindsey Bullinger, Christian Connel, and Max Gross as well as seminar participants at APPAM 2023 for providing helpful comments. We also note special thanks to Alex Winters for helping to link and maintain our administrative data. The results presented and derived in this study do not represent the opinions or views of the Pennsylvania Department of Human Services, Juvenile Court Judges’ Commission, or Office of Medical Assistance Programs. This research was funded with help from the National Institute of Child Health and Human Development R01 HD095946 and P50HD089922. Funding information Eunice Kennedy Shriver National Institute of Child Health and Human Development, Grant/Award Numbers: P50HD089922, R01HD095946 Biographies Ezra G. Goldstein is an assistant professor of public policy at the Georgia Institute of Technology. Sarah A. Font is an associate professor of sociology and public policy at the Pennsylvania State University. Reeve S. Kennedy is an assistant professor of social work at East Carolina University. Christian M. Connell is the Ken Young Family Professor for Healty Children and the director of the Child Maltreatment Solutions Network at the Pennsylvania State University. Allison E. Kurpiel is a research fellow in criminology at the University of Edinburgh. APPENDIX A TABLE A1. Model for estimating the propensity score. Current FC Demographics Age at JJ referral 0.795 *** (0.266) Age at JJ referral sq. −0.025 *** (0.009) Black 0.434 *** (0.080) Hispanic 0.606 *** (0.111) Male −0.345 *** (0.066) Child welfare (CW) contact Any recent CW referral 0.002 (0.067) History of any physical abuse 1.758 ** (0.697) History of any sex abuse 0.144 (0.444) History of allegation of child health or behavioral conflict −0.126 (0.243) History of parent mental illness or disability −0.109 (0.221) History of parent substance use −0.237 (0.149) History of domestic violence 0.073 (0.240) History of abandonment or lack of caregiver −0.173 (0.238) History of inadequate supervision −0.110 (0.184) History of unmet material needs −0.220 (0.136) History of inappropriate discipline −0.128 (0.136) History of confirmed any physical abuse −1.522 ** (0.708) History of confirmed any sex abuse 0.014 (0.459) History of confirmed allegations child health or behavioral conflict 0.257 (0.242) History of confirmed parent mental illness or disability 0.308 (0.243) History of confirmed parent substance use 0.415 ** (0.165) History of confirmed domestic violence −0.302 (0.290) History of confirmed abandonment or lack of caregiver 1.625 *** (0.259) History of confirmed inadequate supervision 0.311 (0.208) History of confirmed unmet material needs 0.264 * (0.155) History of confirmed inappropriate discipline 0.445 *** (0.164) Total past CW referrals 0.194 *** (0.030) History of foster care placement 0.501 *** (0.075) Juvenile justice contact Any violent charge 0.057 (0.076) Any drug or alcohol charge −0.370 *** (0.090) Any weapon charge −0.333 *** (0.098) Any financial crime charge −0.312 *** (0.080) Any sex crime charge 0.473 *** (0.116) Most serious charge is misdemeanor = 1 0.228 ** (0.109) Most serious charge is felony = 1 0.445 *** (0.126) History of any violent charge 0.177 (0.109) History of any drug or alcohol charge −0.490 ** (0.212) History of any weapon charge 0.177 (0.179) History of any weapon charge −0.015 (0.135) Has prior adjudication −0.036 (0.131) Has prior felony charge 0.022 (0.114) Total past JJ referrals 0.023 (0.036) Behavioral health history Emotional disturbance before age 10 −0.012 (0.090) Adjustment disorder before age 10 −0.142 (0.113) Anxiety disorder before age 10 0.033 (0.183) Psychotic disorder before age 10 −0.292 (0.233) Disruptive disorder before age 10 −0.451 *** (0.092) Mood disorder before age 10 0.251 *** (0.093) Any prior emotional disturbance 0.060 (0.105) Any prior adjustment disorder 0.275 *** (0.093) Any prior anxiety disorder 0.050 (0.102) Any prior psychotic disorder 0.301 ** (0.126) Any prior disruptive disorder 0.522 *** (0.083) Any prior substance abuse disorder 0.250 *** (0.094) Observations 10,142 Open in a new tab Note : The table presents the estimates of a logit model of concurrent foster care (Current FC) placement on a rich set of youth characteristics. The model additionally includes a set of dummy variables for each county in which the juvenile was referred and the referral year in which the referral was made. We omit these coefficients for ease of exposition. * p < 0.1, ** p < 0.05, *** p < 0.01. TABLE A2. Robustness to alternative sample selection. (1) (2) (3) (4) Recent CW 90-day window Drop Philadelphia Drop Allegheny Panel A: Any adjudicated charge Current FC −0.049 *** −0.021 −0.033 ** −0.022 (0.017) (0.015) (0.015) (0.015) Observations 5,645 9,619 8,259 8,803 R 2 0.458 0.401 0.460 0.409 Control mean 0.387 0.374 0.387 0.378 Percent effect −12.661 −5.615 −8.527 −5.820 Panel B: Serious disposition Current FC −0.044 ** −0.041 *** −0.052 *** −0.035 ** (0.018) (0.016) (0.015) (0.015) Observations 5,645 9,619 8,259 8,803 R 2 0.447 0.394 0.456 0.401 Control mean 0.393 0.383 0.396 0.386 Percent effect −11.196 −10.705 −13.131 −9.067 Open in a new tab Note : The table shows the robustness of the main estimates to alternative sample construction decisions. Column 1 presents estimates if we instead restrict the control units to only those with prior child welfare (CW) services contact within one year of the delinquency case. In the second column, we omit delinquency cases that may stem from contact with child welfare services and drop delinquency cases that begin within 90 days of the child welfare referral. Finally, Columns 3 and 4 separately drop observations from the two largest counties in the sample, Philadelphia and Allegheny, respectively. * p < 0.1, ** p < 0.05, *** p < 0.01. TABLE A3. Robustness to alternative matching. (1) (2) (3) (4) Kernel matching One nearest neighbor Three nearest neighbors Alternative exact matching Panel A: Any adjudicated charge Current FC −0.020 −0.020 −0.011 −0.007 (0.013) (0.015) (0.013) (0.016) Observations 9,429 2,896 4,455 5,473 R 2 0.414 0.461 0.415 0.448 Panel B: Serious disposition Current FC −0.023 * −0.024 −0.020 −0.014 (0.013) (0.016) (0.014) (0.016) Observations 9,429 2,896 4,455 5,473 R 2 0.400 0.452 0.402 0.447 Open in a new tab Note : The table shows the robustness of the main estimates to alternative matching specifications. The first column presents estimates if we instead implement kernel matching with an Epanechnikov kernel function. In the second and third columns, we instead use our predicted propensity scores to implement nearest-neighbor matching. Estimates using one and three nearest neighbors are shown in Columns 2 and 3, respectively. Finally, in Column 4, we instead construct more strict, exactly matched groups that additionally match the characteristics of the individual juvenile case. These case characteristics include if the case included any violent, drug-related, or weapon-related charge, if the most serious charge was a felony, and if the juvenile had a history of felony charges or a history of adjudicated charges. * p < 0.1, ** p < 0.05, *** p < 0.01. Footnotes CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. 1 When necessary, children are placed most often with kin or unrelated foster families. Most children removed from their homes due to an allegation of abuse or neglect are reunified with their families. 2 For example, prior studies estimate that, depending on the nature of child welfare system involvement, 7% to 24% of youth in foster care become involved with the juvenile justice system ( Cutuli et al., 2016 ), and justice system contact approaches 50% among transition-age youth in foster care ( Courtney et al., 2004 ) 3 Juvenile records, though commonly sealed or expunged, can be used as evidence of a pattern in subsequent criminal trials, inquired about on employment applications, and be accessed by a range of entities, including schools and police, depending on the state; see, for example, Coleman, A. (2020), Expunging juvenile records: Misconceptions, collateral consequences, and emerging practices . U.S. Department of Justice Office of Juvenile Justice and Delinquency Prevention. https://ojjdp.ojp.gov/ . 4 Prior research suggests that judges are more likely to place youth in detention settings when their families are perceived as dysfunctional or involved in criminal activity ( Rodriguez et al., 2009 )—reflecting a lack of confidence in their parents’ capacity to reduce the likelihood of recidivism, or potential suspicions about parental endorsement or involvement in illicit behavior. Thus, among children involved with child welfare services—where parental criminal justice involvement ( Austin, 2016 ), as well as illicit behaviors such as drug use and domestic violence ( Palmer et al., 2022 ), is relatively more common—those remaining with their family of origin may be perceived as a higher risk for recidivism or noncompliance. 5 See, for example, Coleman, A. (2020), Expunging juvenile records: Misconceptions, collateral consequences, and emerging practices . U.S. Department of Justice. https://ojjdp.ojp.gov/ . 6 In Pennsylvania, the juvenile court also makes use of assessment tools such Youth Level of Services and the Pennsylvania Detention Risk Assessment Instrument to aid the court’s decision making and minimize reoffending. Unfortunately, our administrative data do not contain the Pennsylvania Detention Risk Assessment Instrument, and the Youth Level of Services assessment is missing in over two-thirds of cases. As a result, our analysis does not include assessment tools. 7 Children living in foster care have state-screened and approved caregivers and are routinely visited by caseworkers and other service providers. 8 Juvenile justice outcomes vary substantially by jurisdiction, both between and within state lines. For instance, in 2019, the rate of juvenile offenders in residential placements per 100,000 by state ranged from 20 to 330 ( Sickmund et al., 2021 ). For a brief overview of the reduction in the size of the foster care population, see Administration for Children Youth and Families. (2013). Recent demographic trends in foster care . Department of Health & Human Services. https://www.acf.hhs.gov/ . 9 Cases that pertain to child behavioral health appear under different names and structures across states but, as in Pennsylvania, often do not fall under CPS. Analogous designations include “youth in conflict” in Colorado ( https://cdhs.colorado.gov/ ) and “juvenile in need of protective services” in Wisconsin ( https://docs.legis.wisconsin.gov/ ). 10 Authors’ calculations derived from the Kids Count Data Center hosted by the Annie E. Casey Foundation. See, for example, Kids Count Data Center . The Annie E. Casey Foundation. https://datacenter.aecf.org/ . 11 Without a common identifier, creating a cross-system data set is generally not possible. Probabilistic matching algorithms have generated incredible opportunities for researchers seeking to answer questions that require information set across disparate administrative systems. For instance, The Wisconsin Administrative Data Core housed by the Institute for Research on Poverty at the University of Wisconsin uses similar probabilistic matching techniques to combine administrative records across several state systems ( Brown et al., 2020 ). Studies using probabilistically linked data have been instrumental in answering frontier research questions such as providing estimates of parental contact with the criminal justice system ( Finlay et al., 2023 ), documenting the rates of child welfare system contact from birth to adulthood ( Putnam-Hornstein et al., 2021 ), and understanding the consequences of pretrial juvenile detention ( Baron et al., 2023 ). 12 We drop all records associated with individuals who cannot be uniquely matched to the CWIS administrative data. We note the potential for bias due to (un)successful linking is likely negligible since the match rate is consistent across demographic characteristics such as race and gender. 13 Specifically, we define a juvenile defendant currently in foster care if their delinquency case begins between the beginning of an out-of-home placement episode and its end date. 14 For the full list of predictive covariates included in the propensity model and their coefficients; see Table A1 . 15 The outcomes of interest are all binary dependent variables. Each model is estimated separately via ordinary least squares, which we prefer since it permits the use of fixed effects, makes comparisons across specifications simple, provides the best linear approximation of the conditional mean, and allows a straightforward interpretation. However, nonlinear estimation choices such as a probit model provide similar conclusions. 16 In Pennsylvania, the juvenile justice system provides alternatives to formal court processing or adjudication for delinquent youth. One option is an informal adjustment, which serves as an alternative to filing a formal juvenile petition. In this arrangement, the juvenile is required to fulfill certain court conditions, such as community service, payment of fees, and probation, in order to avoid formal court processing. If the juvenile successfully completes the informal adjustment, the case is dismissed. Another alternative is a consent decree, where the court suspends the delinquency case and imposes up to one year of probation. If the juvenile fulfills the conditions outlined in the consent decree, their case is closed, avoiding an adjudication order. 17 For coverage on declining placement options see, for instance, Pennsylvania JCJC. (2021). Pennsylvania secure detention analysis: Impact of facility closures on accessibility of services . https://pccyfs.org/ ; and also Melamed, S. (2022). Here’s how Philly kids ended up sleeping in a DHS conference room for weeks on end . The Philadelphia Inquirer. https://www.inquirer.com/ . For coverage on persistent workforce shortages in Pennsylvania’s child welfare system see, for instance, Person, E. (2023). A staffing shortage in PA juvenile justice system is creating a public safety crisis . WHP Harrisburg. https://local21news.com/ . 18 Several juvenile court jurisdictions have implemented varied programs to deter youth with child welfare histories from further justice system involvement. Programs such as the Systems Integration Initiative and the George-town Center for Juvenile Justice Reform’s Crossover Youth Practice Model seek to improve coordination between local juvenile justice, child welfare, and other systems to enhance cross-system response and services. Such programs have become widespread. For instance, since launching in 2010, the Crossover Youth Practice Model has been implemented in 23 states. In addition, juvenile justice agencies have significantly reduced the use of detention through similar initiatives. For instance, the Anne E. Casey Foundation’s Juvenile Detention Alternatives Initiative encourages alternatives to juvenile detention, such as at-home detention or community supervision, and has been implemented in 40 states. 19 For example, long-accepted conventional wisdom about the negative effects of foster care on justice systems involvement ( Doyle, 2008 ) have been challenged by more contemporary data ( Baron & Gross, 2022 ). 20 See, for example, Federal guidance for child welfare IT systems . U.S. Department of Health & Human Services. https://www.acf.hhs.gov/ . 21 One such promising model is the CHILDREN initiative, a partnership between the U.S. DHHS Assistant Secretary for Planning and Evaluation and Mathematica that focuses on states’ use of child welfare and Medicaid data. See, for example, CHILDREN initiative , Mathematica. https://www.mathematica.org/ . REFERENCES Aizer A, & Doyle JJ Jr (2015). Juvenile incarceration, human capital, and future crime: Evidence from randomly assigned judges. Quarterly Journal of Economics, 130(2), 759–803. [ Google Scholar ] Altonji J, Elder T, & Taber C (2005). Selection on observed and unobserved variables: Assessing the effectiveness of Catholic schools. Journal of Political Economy, 113(1), 151–184. [ Google Scholar ] Arkhangelsky D, & Imbens GW (2022). Doubly robust identification for causal panel data models. Econometrics Journal, 25(3), 649–674. [ Google Scholar ] Asherman-Jusino J (1995). The right of children in the juvenile justice system to inclusion in the federally mandated child welfare services system juvenile detention symposium: III. Critiques of the law and practice affecting juvenile detention in the district of columbia. District of Columbia Law Review, 3(2), 311–354. [ Google Scholar ] Austin A (2016). Is prior parental criminal justice involvement associated with child maltreatment? A systematic review. Children and Youth Services Review, 68, 146–153. [ Google Scholar ] Bald A, Chyn E, Hastings J, & Machelett M (2022). The causal impact of removing children from abusive and neglectful homes. Journal of Political Economy, 130(7), 1919–1962. [ Google Scholar ] Baron EJ, & Gross M (2022). Is there a foster care-to-prison pipeline? Evidence from quasi-randomly assigned investigators (NBER Working Paper Series 29922). National Bureau of Economic Research. [ Google Scholar ] Baron EJ, Jacob B, & Ryan J (2023). Pretrial juvenile detention. Journal of Public Economics, 217, 104798. [ Google Scholar ] Berger LM, Bruch SK, Johnson EI, James S, & Rubin D (2009). Estimating the “impact” of out-of-home placement on child well-being: Approaching the problem of selection bias. Child Development, 80(6), 1856–1876. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Berger LM, Cancian M, Han E, Noyes J, & Rios-Salas V (2015). Children’s academic achievement and foster care. Pediatrics, 135(1), 109–116. [ Google Scholar ] Bishop DM, Leiber M, & Johnson J (2010). Contexts of decision making in the juvenile justice system: An organizational approach to understanding minority overrepresentation. Youth Violence and Juvenile Justice, 8(3), 213–233. [ Google Scholar ] Brown PR, Thornton K, Ross D, Smith J, & Wimer L (2020). Technical report on lessons learned in the development of the institute for research on poverty’s wisconsin administrative data core (Technical Report). University of Wisconsin-Madison Institute for Research on Poverty. [ Google Scholar ] Burns B, Phillips S, Wagner H, Barth R, Kolko D, Campbell Y, & Landsverk J (2004). Mental health need and access to mental health services by youths involved with child welfare: A national survey. Journal of the American Academy of Child and Adolescent Psychiatry, 43, 960–970. [ DOI ] [ PubMed ] [ Google Scholar ] Conger D, & Ross T (2001). Reducing the foster care bias in juvenile detention decisions: The impact of project confirm (Technical Report). The Vera Institute. [ Google Scholar ] Courtney ME, Terao S, & Bost N (2004). Midwest evaluation of the adult functioning of former foster youth: Conditions of youth preparing to leave state care (Technical Report). Chapin Hall Center for Children at The University Of Chicago. [ Google Scholar ] Cutuli JJ, Goerge RM, Coulton C, Schretzman M, Crampton D, Charvat BJ, Lalich N, Raithel J, Gacitua C, & Lee EL (2016). From foster care to juvenile justice: Exploring characteristics of youth in three cities. Children and Youth Services Review, 67, 84–94. [ Google Scholar ] Doyle JJ (2007). Child protection and child outcomes: Measuring the effects of foster care. American Economic Review, 97(5), 1583–1610. [ DOI ] [ PubMed ] [ Google Scholar ] Doyle JJ Jr (2008). Child protection and adult crime: Using investigator assignment to estimate causal effects of foster care. Journal of Political Economy, 116(4), 746–770. [ Google Scholar ] Eastman AL, Foust R, Prindle J, Palmer L, Erlich J, Giannella E, & Putnam-Hornstein E (2019). A descriptive analysis of the child protection histories of youth and young adults arrested in California. Child maltreatment, 24(3), 324–329. [ DOI ] [ PubMed ] [ Google Scholar ] Eastman AL, Herz DC, Palmer L, & McCroskey J (2024). Comparing maltreatment experiences for young people with child protection or dual system involvement. Child and Adolescent Social Work Journal, 41, 747–753. [ Google Scholar ] Eren O, & Mocan N (2021). Juvenile punishment, high school graduation, and adult crime: Evidence from idiosyncratic judge harshness. Review of Economics and Statistics, 103(1), 34–47. [ Google Scholar ] Evangelist M, Ryan JP, Victor BG, Moore A, & Perron BE (2017). Disparities at adjudication in the juvenile justice system: An Examination of race, gender, and age. Social Work Research, 41(4), 199–212. [ Google Scholar ] Feld BC (1999). Bad kids: Race and the transformation of the juvenile court. Oxford University Press. [ Google Scholar ] Finlay K, Mueller-Smith M, & Street B (2023). Children’s indirect exposure to the us justice system: Evidence from longitudinal links between survey and administrative data. Quarterly Journal of Economics, 138(4), 2181–2224. [ Google Scholar ] Font SA, & Gershoff ET (2020). Foster care and “best interests of the child”: Integrating research, policy, and practice. Springer Advances in Child and Family Policy and Practice. Springer. [ Google Scholar ] Font SA, & Kennedy R (2022). The centrality of child maltreatment to criminology. Annual Review of Criminology, 5(1), 371–396. 10.1146/annurev-criminol-030920-120220 [ DOI ] [ Google Scholar ] Galvin MA, & Ulmer JT (2022). Expanding our understanding of focal concerns: Alternative sentences, race, and “salvageability”. Justice Quarterly, 39(6), 1332–1353. 10.1080/07418825.2021.1954234 [ DOI ] [ Google Scholar ] Gross M, & Baron EJ (2022). Temporary stays and persistent gains: The causal effects of foster care. American Economic Journal: Applied Economics, 14(2), 170–199. [ Google Scholar ] Harman JS, Childs GE, & Kelleher KJ (2000). Mental health care utilization and expenditures by children in foster care. Archives of Pediatrics & Adolescent Medicine, 154(11), 1114–1117. [ DOI ] [ PubMed ] [ Google Scholar ] Herz DC, Dierkhising CB, Raithel J, Schretzman M, Guiltinan S, Goerge RM, Cho Y, Coulton C, & Abbott S (2019). Dual system youth and their pathways: A comparison of incidence, characteristics and system experiences using linked administrative data. Journal of Youth and Adolescence, 48(12), 2432–2450. [ DOI ] [ PubMed ] [ Google Scholar ] Imbens GW, & Wooldridge JM (2009). Recent developments in the econometrics of program evaluation. Journal of Economic Literature, 47(1), 5–86. [ Google Scholar ] Kim H, Wildeman C, Jonson-Reid M, & Drake B (2017). Lifetime prevalence of investigating child maltreatment among us children. American Journal of Public Health, 107(2), 274–280. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Leiber MJ, & Mack KY (2003). The individual and joint effects of race, gender, and family status on juvenile justice decision-making. Journal of Research in Crime and Delinquency, 40(1), 34–70. [ Google Scholar ] Leiber MJ, & Peck JH (2015). Race, gender, crime severity, and decision making in the juvenile justice system. Crime & Delinquency, 61(6), 771–797. [ Google Scholar ] Love TP, & Morris EW (2019). Opportunities diverted: Intake diversion and institutionalized racial disadvantage in the juvenile justice system. Race and Social Problems, 11(1), 33–44. [ Google Scholar ] Monahan K, Steinberg L, & Piquero AR (2015). Juvenile justice policy and practice: A developmental perspective. Crime and Justice, 44(1), 577–619. [ Google Scholar ] Moore LD, & Padavic I (2010). Racial and ethnic disparities in girls’ sentencing in the juvenile justice system. Feminist Criminology, 5(3), 263–285. [ Google Scholar ] Oster E (2019). Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics, 37(2), 187–204. [ Google Scholar ] Painter-Davis N, & Ulmer JT (2020). Discretion and disparity under sentencing guidelines revisited: The inter-relationship between structured sentencing alternatives and guideline decision-making. Journal of Research in Crime and Delinquency, 57(3), 263–293. [ Google Scholar ] Palmer L, Font S, Eastman AL, Guo L, & Putnam-Hornstein E (2022). What does child protective services investigate as neglect? A population-based study. Child Maltreatment, 29, 96–105. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Palmer L, Font S, Herd T, Prindle J, & Putnam-Hornstein E (2024). Rates of emotional disturbance among children in foster care: Comparing federal child welfare data and medicaid records in two states. Child Maltreatment, 29, 8–13. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Pennsylvania Council of Chief Juvenile Probation Officers. (2020). A family guide to the Pennsylvania juvenile justice system (Technical Report). Pennsylvania Juvenile Court Judges Commission. (2018). Pennsylvania delinquency benchbook (Technical Report). Pennsylvania Juvenile Court Judges Commission. (2021). Juvenile court annual report (Technical Report). Putnam-Hornstein E, Ahn E, Prindle J, Magruder J, Webster D, & Wildeman C (2021). Cumulative rates of child protection involvement and terminations of parental rights in a California birth cohort, 1999–2017. American Journal of Public Health, 111(6), 1157–1163. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Puzzanchera C (2022). Trends in youth arrests for violent crimes (Technical Report). Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention. [ Google Scholar ] Rodriguez N (2010). The cumulative effect of race and ethnicity in juvenile court outcomes and why preadjudication detention matters. Journal of Research in Crime and Delinquency, 47(3), 391–413. [ Google Scholar ] Rodriguez N, Smith H, & Zatz MS (2009). “Youth is enmeshed in a highly dysfunctional family system”: Exploring the relationship among dysfunctional families, parental incarceration, and juvenile court decision making. Criminology, 47(1), 177–208. [ Google Scholar ] Ryan JP, Herz D, Hernandez PM, & Marshall JM (2007). Maltreatment and delinquency: Investigating child welfare bias in juvenile justice processing. Children and Youth Services Review, 29(8), 1035–1050. [ Google Scholar ] Shook JJ, Goodkind S, Herring D, Pohlig RT, Kolivoski K, & Kim KH (2013). How different are their experiences and outcomes? Comparing aged out and other child welfare involved youth. Children and Youth Services Review, 35(1), 11–18. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Sickmund M, Sladky A, Puzzanchera C, & Kang W (2021). Easy access to the census of juveniles in residential placement: 1997–2019 (Technical Report). Department of Justice Office of Juvenile Justice and Delinquency Prevention. [ Google Scholar ] Sirois C (2023). Contested by the state: Institutional offloading in the case of crossover youth. American Sociological Review, 88(2), 350–377. [ Google Scholar ] Smith CA, Ireland TO, & Thornberry TP (2005). Adolescent maltreatment and its impact on young adult antisocial behavior. Child Abuse & Neglect, 29(10), 1099–1119. [ DOI ] [ PubMed ] [ Google Scholar ] Steffensmeier D, Ulmer J, & Kramer J (1998). The interaction of race, gender, and age in criminal sentencing: The punishment cost of being young, black, and male. Criminology, 36(4), 763–798. [ Google Scholar ] Stevenson MC (2009). Perceptions of juvenile offenders who were abused as children. Journal of Aggression, Maltreatment & Trauma, 18(4), 331–349. [ Google Scholar ] Tam CC, Abrams LS, Freisthler B, & Ryan JP (2016). Juvenile justice sentencing: Do gender and child welfare involvement matter?. Children and Youth Services Review, 64, 60–65. [ Google Scholar ] Teplin LA, Abram KM, McClelland GM, Dulcan MK, & Mericle AA (2002). Psychiatric disorders in youth in juvenile detention. Archives of general psychiatry, 59(12), 1133–1143. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Thomas C, Wolff KT, & Baglivio MT (2022). Understanding juvenile pre-adjudicatory detention and front-end juvenile case processing: The moderating role of race. Journal of Criminal Justice, 81, 101916. [ Google Scholar ] Tracy PE, Kempf-Leonard K, & Abramoske-James S (2009). Gender differences in delinquency and juvenile justice processing: Evidence from National Data. Crime & Delinquency, 55(2), 171–215. [ Google Scholar ] Turney K, & Wildeman C (2016). Mental and physical health of children in foster care. Pediatrics, 138(5). 10.1542/peds.2016-1118 [ DOI ] [ Google Scholar ] U.S. Government Accountability Office. (2018). Additional actions could help HHS better support states’ use of private providers to recruit and retain foster families (Report to Congressional Requesters 18–376). Washington, DC. [ Google Scholar ] Wildeman C, Scardamalia K, Walsh EG, O’Brien RL, & Brew B (2017). Paternal incarceration and teachers’ expectations of students. Socius, 3, 2378023117726610. [ Google Scholar ] Yi Y, Edwards F, Emanuel N, Lee H, Leventhal JM, Waldfogel J, & Wildeman C (2023). State-level variation in the cumulative prevalence of child welfare system contact, 2015–2019. Children and Youth Services Review, 147, 106832. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Yi Y, Edwards FR, & Wildeman C (2020). Cumulative prevalence of confirmed maltreatment and foster care placement for us children by race/ethnicity, 2011–2016. American Journal of Public Health, 110(5), 704–709. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Zane SN (2021). Have racial and ethnic disparities in juvenile justice declined over time? An empirical assessment of the DMC mandate. Youth Violence and Juvenile Justice, 19(2), 163–185. [ Google Scholar ] Zane SN, Mears DP, & Welsh BC (2020). How universal is disproportionate minority contact? An examination of racial and ethnic disparities in juvenile justice processing across four states. Justice Quarterly, 37(5), 817–841. 10.1080/07418825.2020.1766544 [ DOI ] [ Google Scholar ] Zane SN, & Pupo JA (2021). Disproportionate minority contact in the juvenile justice system: A systematic review and meta-analysis. Justice Quarterly, 38(7), 1293–1318. [ Google Scholar ] Zane SN, & Pupo JA (2023). What predicts out-of-home placement in juvenile court dispositions? A systematic review and meta-analysis. Journal of Youth and Adolescence, 52(1), 229–244. [ DOI ] [ PubMed ] [ Google Scholar ] ACTIONS View on publisher site PDF (1.2 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Global meta-analysis of drought associated rice rhizosphere microbiome data and soil application of selected bacteria reveals potential of Pantoea dispersa for drought stress alleviation in rice plants.#1001263
• High-magnification inpatient cases under DIP payment reform: prevalence, cost burden, and risk factors in an NHC-administered hospital in China.#1001264
• Superior-head automated external defibrillator placement during single-rescuer cardiopulmonary resuscitation: a randomized crossover simulation study and retrospective multicenter cohort.#1001265
• Spatial-temporal epidemiological characteristics of thyroid cancer mortality in China cancer registration areas: analysis based on multiple models.#1001266