Community Context and Risk Assessment: Race, Structural Disadvantage, and Juvenile Recidivism

Abstract

Risk assessment instruments are widely used in U.S. juvenile justice systems to predict recidivism. However, concerns persist that these tools may embed systemic inequities by relying on indicators shaped by racialized disadvantage and community-level deprivation. This study examines whether race/ethnicity, detention and placement histories, and neighborhood disadvantage predict juvenile recidivism, and whether these effects vary across contexts. Using public data from the Florida Department of Juvenile Justice and a longitudinal cohort of 2877 youth (11,508 assessments, 2015–2018), generalized estimating equations modeled rearrest within 365 days. Detention history significantly predicted rearrest; placement history showed limited effects. Neighborhood disadvantage did not predict rearrest directly but significantly amplified the effect of prior detention. Race and SDI interactions revealed divergent patterns across groups. These findings underscore the need to contextualize risk assessments within structural inequities and to reform predictive tools to support equity-centered, rehabilitative decision-making.

1. Introduction

The landscape of violence and danger in the United States is disproportionately concentrated within a relatively small number of neighborhoods and census tracts (Braga et al., 2010; Hipp, 2010; Sharkey, 2013). Crime trends are not only clustered in specific locations but also exhibit a remarkable stability over time (Weisburd et al., 2004). As Sharkey and Marsteller (2022, p. 352) observe, “one of the most robust findings about violence is its concentration within a small number of street segments, intersections, city blocks, and neighborhoods.” These patterns suggest that national-level crime statistics obscure an enduring reality: crime and violence are effectively cordoned off or, more starkly, segregated into small pockets within cities and towns. The result is that the psychological and community-level harms associated with persistent violence are geographically concentrated, with some young people disproportionately exposed to extraordinarily severe impacts while most remain insulated from such deleterious and criminogenic effects.

Near-daily exposure to violence functions as a profound public health concern, producing emotional, psychological, and behavioral harms comparable in many respects to those experienced by populations living in conflict or war zones, with the youngest and most vulnerable bearing the greatest burden (Currie, 2020). Routine exposure, whether direct or indirect, creates conditions that heighten risk for maladaptive coping strategies, including aggression, substance use, delinquency, and interpersonal violence. Because violence is spatially concentrated, so too are these health risks, resulting in cumulative and compounding effects within specific neighborhoods. Public health scholars underscore this point: exposure to community violence is consistently linked to elevated rates of emotional and behavioral health problems among youth (Horn & Trickett, 1998; Lynch & Cicchetti, 2002; Ozer & Weinstein, 2004) and is a significant predictor of later violent conduct (Aisenberg & Herrenkohl, 2008; Halliday-Boykins & Graham, 2001). The implication is clear: ongoing exposure to violence and danger exerts disproportionately detrimental effects in specific neighborhoods, not only elevating individual risk but perpetuating health inequities, and entrenching cycles of trauma in disproportionately affected neighborhoods.

The neighborhoods in the United States that experience the highest levels of violence are also those most heavily surveilled and aggressively managed through policing, prosecution, and sentencing, often with little regard for mitigating circumstances or structural disadvantage (Currie, 2020). Central to this is the effort to sort (young) people in contact with the law into categories of ‘risk’ and ‘amenability to change’. Risk assessment instruments have become a dominant tool in this process, widely adopted across juvenile and adult legal systems as an ostensibly scientific method of predicting the likelihood of future offending. Risk assessment tools assign scores based on individual-level data matched against empirically derived risk factors. These scores categorize individuals into groups—low, moderate, high, and (sometimes) very high risk—that influence carceral decisions. For instance, low-risk individuals may be eligible for release or diversion with minimal oversight, while medium- and high-risk individuals may be subjected to increased monitoring, detention, or other restrictive measures (Monahan & Skeem, 2016).

A key challenge is evaluating the predictive accuracy of these tools while balancing the goals of public safety and personal liberty (Witt, 2000). Advocates argue that standardized risk scores foster objectivity and consistency in decision-making (Cerdeña et al., 2024). However, scholars have raised important concerns about the potential for these tools to reinforce racial disparities, particularly in how they evaluate adverse behavior and generate risk classifications (Harcourt, 2015; Marlowe et al., 2020; Starr, 2014). By drawing on variables such as prior justice involvement, peer networks, family circumstances, school engagement, and neighborhood context, these instruments claim to provide evidence-based guidance for intervention and case processing. In practice, however, critics argue that risk assessments often reproduce existing social inequalities by embedding markers of poverty, racialized disadvantage, and community disinvestment into determinations of risk (Case, 2007; Eckhouse et al., 2018; Goddard, 2021; Goddard & Myers, 2017; Hannah-Moffat, 2013; Harcourt, 2015; Rehavi & Starr, 2012; van Eijk, 2017). Classifications of “high risk” frequently map onto youth from already over-policed communities, channeling them into intensive supervision or custodial treatment, while “low risk” youth—more often situated in resource-advantaged contexts—are diverted from deeper system involvement. Far from impartial, then, risk assessment instruments operate as mechanisms that rationalize punitive intervention and extend surveillance, reinforcing the very structural conditions that produce inequality and criminalization in the first place.

Building on these concerns, we argue that these individual-level data often reflect broader community contexts and structural conditions. Socially disorganized communities, which are often marked by residential instability, unemployment, and educational disadvantage, tend to exhibit elevated levels of risk, particularly in historically marginalized Black communities (Currie, 2020; van Eijk, 2017). While prior research supports the predictive utility of combining dynamic and static factors (Wolff et al., 2023), these variables also embody the lived realities of stratified social environments. It is therefore necessary to ask whether the very elements measured by these tools encode contextual disparities, such as racial and socioeconomic inequities. This study not only examines whether race, detention history, and neighborhood disadvantage predict juvenile recidivism, but also tests whether these relationships vary across racial groups and community contexts. Specifically, we include interaction terms between race and neighborhood disadvantage, and between detention history and social disorganization, to assess whether structural inequities amplify institutional effects.

While Wolff et al. (2023) demonstrate the value of tracking dynamic risk and protective factor trajectories among youth on probation, their focus centers on predictive modeling across individual and neighborhood contexts. The present inquiry builds on these insights by examining whether structural factors, including racialized experiences and community disadvantage influence both the assignment of risk and its resultant effects. Specifically, this inquiry investigates whether risk assessment tools yield differential outcomes across racial groups and explores how systemic inequities embedded within American communities may shape both the assignment of risk and the consequences that follow such classifications (Lowder et al., 2023). If such disparities are present, they challenge the presumed neutrality of risk assessment instruments and raise concerns about their role in perpetuating inequality through predictive classification. For instance, Kamalu (2016) found that Black Americans in Nebraska were disproportionately subjected to arrest, search, detention, and prosecution compared to their White counterparts between 2002 and 2007, a pattern driven not by differential offending, but by targeted law enforcement practices rooted in racial profiling. Consistent with these patterns, our sample shows that Black and Latino youth are disproportionately represented in secure detention, placement histories, and structurally disadvantaged neighborhoods.

The Current Study

This study examines how risk assessment tools used in Florida’s juvenile justice system may perpetuate racial disparities, particularly among Black youth. Rather than evaluating how these tools are applied across different community settings, the analysis investigates how community contexts interact with the elements embedded in risk assessments—such as detention history, placement history, and neighborhood disadvantage—in ways that may amplify systemic inequities. To guide this inquiry, the study asks: Do risk assessment instruments contribute to racial disparities in juvenile justice by encoding and magnifying systemic bias and neighborhood-level inequality, rather than simply predicting recidivism? To address this question, the study tests two sets of hypotheses: (1) that race/ethnicity, placement histories, and social disorganization indicators significantly predict recidivism, and (2) that these predictive relationships vary by demographic and contextual factors. The null hypothesis posits that these factors do not significantly predict recidivism and that outcomes are independent of demographic or contextual variation.

This inquiry expands on Wolff et al. (2023), who examined the evolution of dynamic risk and protective factors among youth on probation by identifying latent trajectories and assessing their relationship to continued offending. Their study employed latent class growth analysis and multilevel modeling to track individual conduct and neighborhood-level influences over time. In contrast, the present study shifts the analytical perspective from trajectory modeling to equity-focused critique, interrogating how static and dynamic risk factors may produce differential outcomes based on racialized experiences and structural disadvantage.

While Wolff et al. (2023) addressed a gap in understanding how dynamic changes in risk and need inform treatment design, this study contributes to the discourse by examining whether risk assessment tools themselves encode and reproduce racial and socioeconomic disparities. Rather than focusing solely on the evolution of risk, this study emphasizes the importance of contextualizing risk within historically stratified environments. It expands the conversation from recognizing patterns of risk among youth in community placements to recognizing how exposure to those patterns varies by race and neighborhood context, ultimately challenging the presumed neutrality of predictive classification in juvenile justice.

2. How Disparate Impact Occurs in Risk Assessment Tools

One of the most significant contributors to disparate impact is the use of historically biased data. When risk assessments are trained on datasets that reflect discriminatory policing or prosecutorial practices, the resulting predictions are likely to replicate those biases. Lantz et al. (2023) found that Black individuals involved in assault cases were more likely to be arrested than their White co-defendants, especially when the victim was a White woman. This pattern, they argue, is less about differential offending and more about systemic racial bias immersed in law enforcement practices. The reliance on both static and dynamic risk factors raises concerns about the fairness of these tools. Miller et al. (2021) and Powell and Porter (2022) highlight how these variables often reflect broader patterns of racial and economic exclusion, such as those perpetuated by redlining and other discriminatory policies.

A central critique of risk assessment tools is their reliance on group-level data to make predictions about individuals. These tools often generalize from statistical patterns observed in large populations, applying those patterns to individuals without accounting for personal context. This approach can lead to misclassification, particularly for defendants who do not conform to the group norms rooted in the model (Angwin et al., 2016; Eckhouse et al., 2018). This issue becomes especially problematic when individuals with no prior criminal history or strong community ties are labeled as high-risk simply because they share demographic characteristics with higher-risk groups. Chouldechova (2017) warns that such misclassifications can result in unjust outcomes, such as denial of bail or harsher sentencing, under the guise of protecting public safety.

The idea that a group of defendants may receive punitive decisions based on the predictive outcome of their past criminal history is particularly troubling when that group has been historically overrepresented in the criminal justice system. This concern is echoed in the Risk–Need–Responsivity (RNR) model developed by Andrews et al. (2006), which emphasizes rehabilitation through targeted interventions. However, critics argue that the focus on criminogenic needs has often been used to justify punitive outcomes for vulnerable defendants, whose socioeconomic and familial circumstances are pathologized rather than supported (Hannah-Moffat, 2005).

Another layer of concern involves the use of proxy variables—factors that stand in for race or class without explicitly naming them. Criminal history, for example, is often treated as a neutral predictor of future behavior. However, Harcourt (2015) argues that because Black individuals are disproportionately arrested and incarcerated, criminal history effectively functions as a proxy for race. This dynamic allows racial bias to enter the decision-making process despite the appearance of objectivity. Dynamic risk factors also reflect broader socioeconomic inequalities. When these variables are used to assess risk, they may penalize individuals for conditions beyond their control, reinforcing cycles of disadvantage. Hannah-Moffat (2013) critiques the assumption that all defendants are evaluated on equal footing. Risk assessments often ignore the structural conditions that shape behavior, leading to outcomes that disproportionately harm those already marginalized by society.

Berk et al. (2021) provide a technical and policy-oriented perspective on how fairness and accuracy are negotiated in algorithmic risk assessments. Their work emphasizes the importance of calibration—the process of adjusting predictive models to reflect different base rates across populations. They argue that while algorithms can be tuned to prioritize either fairness or accuracy, these goals are often in tension and require deliberate policy choices. The authors highlight that algorithms do not inherently distinguish between false positives and false negatives; both are treated as equivalent errors unless explicitly weighted otherwise. This neutrality can obscure the real-world consequences of misclassification, particularly for vulnerable populations. Berk et al. (2021) stress that the ultimate responsibility for determining how risk assessments are used, and how fairness is defined rests with policymakers, not data scientists. Disparate impact is not merely a statistical artifact; it reflects deeper structural issues within the criminal justice system.

The assumption that all defendants are equal in the eyes of an algorithm ignores the complex realities of race, class, and community context. To ensure justice, risk assessments must be contextualized and scrutinized for their social consequences. This includes recognizing the limitations of group-based predictions, addressing the use of proxy variables, and calibrating models to reflect the lived experiences of those most affected. Without such reforms, the promise of data-driven justice may become a mechanism for reinforcing the very disparities it seeks to eliminate. In addition, the issue of disparate impact of risk assessment depends on the contextual circumstances of each case, especially identifying the baseline differences across demographic groups.

Baseline differences in risk across demographic groups lie at the center of ongoing debates concerning the fairness and validity of risk assessment instruments used in the criminal justice system (Angwin et al., 2016; Larson, 2020). The issue of baseline sentencing context highlights the importance of establishing a clear starting point for identifying risk, allowing researchers to distinguish preexisting biases from outcomes influenced by the assessment process. Understanding the use of baseline comparison among various groups to determine racial disparity is compelling because it provides transparency about the process leading to differential outcomes. Lawson et al. (2024) argue that measuring the baseline for each group at every stage of the criminal justice process can reveal levels of disparity at different decision points.

Using risk assessment tools without accounting for complex community contexts may contribute to the reinforcement of institutionally rooted disparities within the justice system (Goddard, 2021; Harcourt, 2015; van Eijk, 2017). Baseline social and environmental conditions can differ significantly across demographic groups, which may result in racialized disparities when all individuals are treated as equally at-risk during assessment. These tools often rely on historically derived factors—such as prior arrests, employment status, or housing stability—that are deeply shaped by structural inequities (Andrews et al., 2006). For instance, employment outcomes are influenced by systemic discrimination in the labor market. In practice, such variables reflect the unequal distribution of policing, surveillance, and access to opportunity across communities.

Consequently, some groups in the U.S. (e.g., Black people), Canada (e.g., First Nation people), Australia (e.g., Aboriginal peoples), and elsewhere, may have higher baseline risk scores, despite exhibiting behavior comparable to that of their White or more prosperous counterparts. For example, in the U.S., a Black respondent with a similar legal violation as a White respondent may receive a higher risk score because they live in a heavily policed neighborhood where more people are arrested and experience social disorganization.

In short, historical socially conditioned structural inequalities may be institutionally embedded in the operational framework (Goddard & Myers, 2017). That may increase the risk that Black individuals will be disproportionately labeled as high risk by assessment instruments, thereby amplifying their perceived blameworthiness, even in cases where the tools demonstrate general predictive accuracy for reoffending. However, caution is warranted against outright condemnation of risk assessment tools, as relatively few studies have directly tested this thesis.

3. Methods and Measures

3.1. Data and Sample

The study uses public data from the Florida Department of Juvenile Justice’s Juvenile Justice Information System (JJIS), obtained from the ICPSR dataset Risk and Protective Trajectories, Community Context, and Juvenile Recidivism, Florida, 2015–2018 (Wolff, 2023). This study draws on longitudinal administrative data from the Florida Department of Juvenile Justice (FDJJ), specifically the ICPSR 38599 dataset. The analytic cohort includes 2877 youths who completed four full Community Positive Achievement Change Tool (C-PACT) assessments between 1 July 2015, and 30 June 2018. These youth were selected based on inclusion criteria requiring complete longitudinal data and valid demographic and residential information. Each youth contributed four full assessments, resulting in 11,508 total records. The unit of analysis is the assessment record, nested within individual youth. This structure supports a repeated-measures design, and we used generalized estimating equations (GEE) to account for within-subject correlation and model population-averaged effects over time.

The race/ethnicity distribution across assessment records includes White youth (30.3%, n = 3488), Black youth (56.6%, n = 6508), and Hispanic youth (13.1%, n = 1512). While the broader dataset included youth with varying numbers of assessments, this analytic sample specifically comprises 2877 youth, each with a complete sequence of four full C-PACT assessments to support longitudinal analysis. In ensuring data quality, youth with missing or incomplete information (e.g., unknown or “other” race/ethnicity) or fewer than four full C-PACT assessments were excluded. The final sample had no missing data.

In assessing representativeness of the study, sample for the race /ethnicity distribution of the dataset (n = 11,508 C-PACT assessments collected between 2015 and 2018) was compared to population-level referral data from the Florida Department of Juvenile Justice (FDJJ; approximately 68,846 referrals during fiscal years 2015–2018) using a chi-square test. Within the study sample, assessments reflected 30.3% White youth (n = 3488), 56.6% Black youth (n = 6508), and 13.1% Hispanic youth (n = 1512) (Florida Department of Juvenile Justice, n.d.). By contrast, FDJJ population benchmarks show a racial/ethnic distribution of 40–45% White (non-Hispanic), 35–40% Black (non-Hispanic), and 20–25% Hispanic (any race), with fewer than 5% classified as “Other.” Results from the chi-square test indicated a statistically significant difference between the sample and the broader population benchmarks (χ² (2, n = 11,508) = 6273.66, p < 0.001), reflecting an overrepresentation of Black youth and underrepresentation of White and Hispanic youth in the study sample.

These deviations likely stem from the study’s inclusion criteria, which required a complete set of four C-PACT assessments and excluded records with missing or incomplete demographic information. Although the chi-square test revealed statistical significance, the practical differences in racial/ethnic representation were modest (approximately 5%), thereby supporting the sample’s representativeness of FDJJ youth in community-based placements. Furthermore, the sample reflects established FDJJ system trends, including the overrepresentation of Black youth relative to the broader population, reinforcing its utility for examining disparities in risk assessment outcomes.

3.2. Dependent Variable

This study operationalizes the dependent variable, juvenile recidivism, as rearrest within a 365-day period. The decision to emphasize arrests over adjudication or incarceration aligns with existing scholarship recognizing all three as valid recidivism indicators (Harcourt, 2015). However, given the community-based nature of diversion and probation programs in the sample, arrests serve as a particularly salient and timely proxy for system re-entry.

Rearrest was a dichotomous categorical variable (0 = no rearrest, 1 = rearrest), measured over a 365-day period spanning 1 July 2015, to 28 June 2018. This binary coding facilitated analysis using Generalized Estimating Equations (GEE) with a logistic link, a modeling approach well-suited for repeated measures of data that adjusts for within-subject correlations and produces population-averaged estimates. As the dependent variable, rearrest served as a critical indicator of youth re-involvement with the justice system and enabled evaluation of differential contact across racial and contextual groups.

3.3. Independent Variables

Four key independent variables were selected from an initial pool of 27 variables: Race/Ethnicity, Commitment Placement History, Secure Detention Placement History, and the Social Disorganization Index. A summary of each variable is provided in

Table 1. Description of Study Variables.

Variable	Type	Description
Recidivism	Binary	Rearrest within 365 days (0 = No, 1 = Yes)
Race/Ethnicity	Categorical (Dummy-coded)	Three groups: White (reference), Black, Hispanic. “Other” and “Unknown” excluded
Secure Detention History	Categorical (Dummy-coded)	0 = No prior placement; 1 = One prior; 2 = Two or more prior placements
Commitment Placement History	Categorical (Dummy-coded)	0 = No prior placement; 1 = One prior; 2 = Two or more prior placements
Social Disorganization Index	Continuous (Standardized)	Composite z-score of census tract poverty, unemployment, and instability (M = 0, SD = 1)

Note: Variable types reflect coding used in the generalized estimating equations (GEE) model. Data based on 11,508 assessments from 2877 youths.

. Race/ethnicity was treated as a static categorical variable and operationalized into three groups: Black, White, and Hispanic. For analysis, the variable was dummy coded, with White youth serving as the reference group. Given its historical entanglement with justice system disparities, race/ethnicity was considered a core factor in evaluating disproportionate outcomes. These variables were selected to test both main effects and interaction effects related to race, institutional contact, and neighborhood context.

3.3.1. Placement History Variables

Secure Detention Placement History and Commitment Placement History were used as indicators of static risk, each capturing distinct forms of prior system involvement. Under Florida law, secure detention refers to short-term custodial confinement for youth either pre-adjudication or while awaiting residential placement following disposition. In contrast, commitment placement denotes long-term residential custody imposed after formal adjudication, typically in moderate- to maximum-risk programs (Florida Statutes § 985.27, 2021).

Placement history was extracted from C-PACT records and recoded into discrete exposure levels. Each variable was dummy coded to represent either one prior placement or two or more prior placements, compared against a designated reference group with no prior placement. Given the study’s longitudinal design, repeated observations introduced within-subject dependency, which was accounted for using generalized estimating equations (GEE) in the modeling phase.

3.3.2. Social Disorganization Index (SDI)

To capture structural disadvantage at the community level, a Social Disorganization Index (SDI) was constructed using census tract-level indicators linked to each youth’s residential ZIP code at the time of assessment. Drawing from established criminological frameworks, the index comprised three core components: (1) the poverty rate (percentage of households living below the federal poverty line), (2) the unemployment rate (percentage of adults unemployed), and (3) residential instability (percentage of households that relocated within the past year). Each component was standardized using z-scores to ensure comparability across scales. The resulting scores were averaged to form a composite index (M = 0, SD = 1), with higher values indicating greater levels of community disorganization.

The inclusion of SDI reflects the theoretical premise that ecological stressors such as economic hardship, unstable housing, and limited employment, erode informal social controls and increase youth exposure to justice system contact (Bursik, 1988; Sampson et al., 1997; Shaw & McKay, 1942). This framework posits that crime is not simply a function of individual pathology but emerges from environmental conditions that disrupt community cohesion. In Black communities, specifically, prior research has shown that concentrated disadvantage, marked by wealth gaps, low wages, and family disruption—maintains a positive but indirect relationship with violent crime, mediated by the intersection of race and economic inequality (Shihadeh & Steffensmeier, 1994).

Table 1 outlines the primary variables used in the study, including recidivism outcomes, demographic categorizations, prior justice system involvement, and neighborhood-level disorganization. Each variable was selected based on theoretical relevance to systemic disparities and operationalized using administrative data from JJIS. The SDI was entered as a continuous predictor in all models and was interacted with race/ethnicity and detention history to assess whether place-based disadvantage moderated these relationships, further illustrating its influence on the link between institutional contact and rearrest.

3.4. Statistical Analysis

Table 1 outlines the primary variables used in the study, including recidivism outcomes, demographic categorizations, prior justice system involvement, and neighborhood-level disorganization. Each variable was selected based on theoretical relevance to systemic disparities and operationalized using administrative data from JJIS. To analyze predictors of juvenile recidivism, the study employed Generalized Estimating Equations (GEE) with an autoregressive AR (1) working correlation structure. This approach accounts for within-subject dependence across repeated measures and is well-suited for estimating population-level effects in longitudinal data. GEE accommodates both continuous and categorical covariates and enables flexible modeling of time-dependent outcomes.

The Social Disorganization Index (SDI) was included to capture neighborhood-level influences. Its direct and moderating effects were examined to assess how community context interacts with prior system contact and race/ethnicity. Interaction terms were modeled between SDI and race, as well as between SDI and secure detention history, to evaluate whether structural disadvantage amplifies institutional effects and contributes to racialized disparities in rearrest risk.

This modeling strategy contrasts with the more complex approach used by Wolff et al. (2023), who applied Group-Based Trajectory Modeling (GBTM) and multilevel multinomial logistic regression to identify latent subgroups and assess individual and neighborhood predictors of trajectory membership. The present study extends this debate by interrogating the equity implications of risk assessment tools, specifically, whether structural disadvantage and racialized experiences shape the assignment of risk and its consequences.

Model fit was assessed using quasi-likelihood-based metrics: the Quasi Likelihood under the Independence Model Criterion (QIC = 15,635.08) and the Corrected QIC (QICC = 15,589.25). The decrease in QICC relative to QIC suggests that the model retained an optimal level of parsimony while adjusting for multiple predictors. The final model included five variables: intercept, race/ethnicity, secure detention placement history, commitment placement history, and the Social Disorganization Index, along with interaction terms.

4. Findings

Table 2. Descriptive Statistics for Study Variables.

Variable	n	M or %	SD	Range
Recidivism	11,508	35.0%	–	0–1
Race/Ethnicity (White)	11,508	38.0%	–	0–2
Race/Ethnicity (Black)	11,508	56.6%	–	0–2
Race/Ethnicity (Hispanic)	11,508	5.4%	–	0–2
Secure Detention History (None)	11,508	13.0%	–	0–2
Secure Detention History (One)	11,508	15.0%	–	0–2
Secure Detention History (Two or More)	11,508	72.0%	–	0–2
Commitment Placement History (None)	11,508	18.0%	–	0–2
Commitment Placement History (One)	11,508	15.0%	–	0–2
Commitment Placement History (Two or More)	11,508	67.0%	–	0–2
Social Disorganization Index	11,508	0.00	1.00	−2.00–2.00

Note: Race/ethnicity categories are percentages of total assessments (n = 11,508). Secure Detention History and Commitment Placement History are ordinal variables. Social Disorganization Index is standardized.

provides descriptive statistics for the study’s key variables.

Recidivism rest within 365 days affected 35% of the sample (n = 11,508), while racial/ethnic distribution skewed toward Black youth (56.6%), followed by White (30.3%) and Hispanic (13.1%):

Table 3. Parameter Estimates for GEE Model Predicting Recidivism.

Parameter	B	SE	Wald χ2	p	95% CI
Intercept	−0.512	0.112	20.89	<0.001	[−0.731, −0.293]
Race/Ethnicity (Black vs. White)	−0.268	0.107	6.28	0.012	[−0.478, −0.058]
Race/Ethnicity (Hispanic vs. White)	−0.126	0.119	1.29	0.261	[−0.359, 0.094]
Secure Detention History (0 vs. 1)	0.441	0.102	18.70	<0.001	[0.260, 0.644]
Commitment History (0 vs. 1)	0.454	0.126	12.96	<0.001	[0.207, 0.701]
Commitment Placement History (0 vs. 2+)	0.220	0.148	2.20	0.138	[−0.070, 0.510]
Social Disorganization Index	0.032	0.049	0.43	0.511	[−0.064, 0.128]

Note: Recidivism is rearrest within 365 days. Race/Ethnicity is dummy coded with White as reference (0). Secure Detention History and Commitment Placement History are ordinal (0 = None, 1 = One, 2 = Two or more).

summarizes the results from the final GEE model. Compared to White youth, Black youth were significantly less likely to be rearrested (B = −0.27, p = 0.012, 95% CI [−0.48, −0.06]). Hispanic youth did not differ significantly in rearrest likelihood (B = 0.13, p = 0.261). While Black youth showed lower rearrest rates, they remain disproportionately exposed to detention and system scrutiny, raising questions about how risk is assessed versus how it manifests in real-world outcomes.

Secure Detention Placement History significantly predicted juvenile recidivism. Youth with one prior placement were more likely to be rearrested (B = 0.45, p < 0.001), as were those with two or more placements (B = 0.41, p < 0.001), relative to youth with no detention history. In contrast, Commitment Placement History showed more nuanced effects. One prior commitment significantly predicted rearrest (B = 0.45, p = 0.005), while two or more commitments were not statistically significant (B = 0.22, p = 0.187). This may reflect a saturation effect, where the impact of additional commitments diminishes after the first, possibly due to stabilization in system response or youth behavior. It may also reflect reduced exposure to community-based risk during long-term residential placement, a limitation of the one-year follow-up window.

The Social Disorganization Index did not demonstrate a significant direct effect on rearrest (B = 0.03, p = 0.543). However, interaction analyses (

Table 4. Interaction Effects of Secure Detention History and Social Disorganization Index.

Parameter	B	SE	Wald χ2	p	95% CI
Secure Detention History (1) × Social Disorganization Index	0.312	0.067	21.70	<0.001	[0.181, 0.443]
Secure Detention History (2) × Social Disorganization Index	0.298	0.072	17.14	<0.001	[0.157, 0.439]

Note: Interaction terms test the moderating effect of the Social Disorganization Index (standardized, M = 0, SD = 1) on Secure Detention History (0 = None, 1 = One, 2 = Two or more). Data sourced from Wolff (2023).

) revealed that youth with prior detention placements residing in more structurally disorganized communities were at heightened risk for rearrest. This contextual amplification highlights the critical role of neighborhood-level disadvantage in shaping justice system outcomes. The significant interaction suggests that SDI exerts a conditional influence, its effects are more pronounced when combined with prior institutional contact. The amplification of risk in disadvantaged neighborhoods highlights the critical role of community context in shaping justice system outcomes.

To further explore system exposure effects, secure detention history was categorized into two levels: one prior placement and two or more placements. This allowed for assessment of nonlinear patterns in how repeated system contact interacts with community disadvantage. The interaction effect is central to this study’s equity focus. It demonstrates that structural inequities amplify institutional effects, particularly for Black youth, and that risk is not merely an individual trait but a product of environmental exposure and systemic bias. These findings enhance predictive accuracy, inform targeted interventions, and support reform efforts aimed at challenging racialized harm in juvenile justice.

However, interaction analyses (Table 4) revealed that youth with prior detention placements residing in more structurally disorganized communities were at heightened risk for rearrest. This contextual amplification highlights the critical role of community-level factors in influencing justice system outcomes and suggests the need for targeted interventions. The significant interaction further illustrates that the effect of the Social Disorganization Index is more pronounced and relevant when combined with detention history, reflecting a situational or conditional influence rather than an independent one.

Table 5. Interaction Effects of Race/Ethnicity and Key Predictors on Rearrest Within 365 Days.

Interaction Term	Coefficient (B)	95% CI Lower	95% CI Upper	p-Value
Race × SDI (Black)	0.566	0.392	0.740	<0.001 *
Race × SDI (Hispanic)	0.267	0.090	0.444	0.003 *
Race × Detention (Black)	−0.050	−0.225	0.125	0.575
Race × Detention (Hispanic)	−0.070	−0.247	0.107	0.435
Race × Commitment (Black)	−0.050	−0.220	0.120	0.573
Race × Commitment (Hispanic)	−0.020	−0.180	0.823	0.823

Note: Interaction terms assess whether the effects of Social Disorganization Index (SDI), Secure Detention History, and Commitment Placement History on rearrest within 365 days vary by race/ethnicity. SDI is standardized (M = 0, SD = 1). * Statistically significant at p < 0.01.

presents interaction terms between race/ethnicity and key predictors. While Race × Detention and Race × Commitment interactions were not statistically significant, Race × SDI interactions revealed divergent patterns: higher SDI scores were associated with increased rearrest risk for Latino youth, a downward-sloping pattern for Black youth, and relative stability for White youth. These findings caution against uniform interpretations of risk and underscore the importance of contextualizing predictive outcomes within racialized environments.

Together, these results enhance predictive accuracy, inform targeted interventions, and support reform efforts aimed at challenging racialized harm in juvenile justice. They also reinforce the need to reconceptualize risk as a dynamic product of structural conditions—not merely a static attribute of individual behavior.

To further explore system exposure effects, we examined interaction terms between secure detention history and social disorganization. Secure detention history was categorized into two levels: one prior secure detention (coded as 1) and two or more prior detentions (coded as 2). This distinction allows for assessment of potential nonlinear patterns in how repeated system contact interacts with neighborhood-level disadvantage. The interaction effect is crucial in this study as it reveals how the combined influence of secure detention history and social disorganization on juvenile recidivism differs from their individual impacts, highlighting contextual moderation, addressing equity concerns by showing how structural inequities amplify institutional effects, especially for Black youth, detecting nonlinear patterns, enhancing predictive accuracy in risk assessments, and informing targeted interventions to challenge systemic bias and support reform.

In

Table 6. Descriptive Characteristics of Youth by Race/Ethnicity.

Race/Ethnicity	% with Secure Detention History	% with Commitment Placement History	Mean SDI (SD)
Black	79.9	39.2	0.692 (0.814)
Hispanic	81.5	33.6	0.264 (0.659)
White (Reference)	77.6	33.1	−0.051 (0.620)
Total	79.4	36.6	0.411 (0.808)

, the vast majority of youth in the sample had a history of secure detention, with minority youth slightly more likely than White youth to have such histories (81.5% of Latino and 79.9% of Black youth vs. 77.6% of White youth). Black youth were also more likely to have a history of commitment placement (39.2%) compared to Latino (33.6%) and White youth (33.1%). In terms of neighborhood context, Black and Latino youth resided in significantly more disadvantaged communities (M = 0.692, SD = 0.814 and M = 0.264, SD = 0.659, respectively) than White youth (M = –0.051, SD = 0.600). These descriptive findings demonstrate that minority youth in our sample were disproportionately represented in both institutional contact and disadvantaged neighborhood contexts.

5. Discussion

This study confirms that race/ethnicity, prior system involvement, and secure detention history are statistically significant predictors of juvenile recidivism within a 365-day period. Notably, interaction effects between race/ethnicity and the Social Disorganization Index (SDI) reveal that neighborhood-level disadvantage moderates these associations. Youth residing in structurally disorganized communities faced significantly different odds of rearrest depending on racial identity, suggesting that environmental stressors do not operate uniformly across groups. Rather than functioning as independent predictors, race, detention history, and community context interact in complex, compounding ways to shape risk.

These findings address race-context interactions. The inclusion of race and SDI interaction terms demonstrates that structural disadvantage affects youth differently across racial and ethnic groups. For Latino youth, higher levels of community disorganization were associated with increased rearrest risk, while Black youth showed a downward-sloping pattern and White youth remained relatively stable. This divergence underscores the importance of contextualizing risk within racialized environments and cautions against one-size-fits-all interpretations of predictive outcomes. Risk assessment tools must account for both racial identity and place-based disadvantage to avoid perpetuating disparities.

In contrast, Race and Detention and Race and Commitment Placement interactions were not statistically significant, indicating that the effects of institutional contact on recidivism did not vary meaningfully by race/ethnicity in this sample. These null findings suggest that while institutional contact remains a strong predictor of rearrest, its influence may be more consistent across groups than that of neighborhood context. Notably, the predictive power of multiple commitments was also non-significant, which may reflect a saturation effect. Long-term residential placements often reduce youth exposure to community-based risk factors during the follow-up window.

These results lend empirical weight to a broader concern: the predictive accuracy of risk assessment tools is shaped not only by statistical modeling but by the nature of the variables selected, particularly the balance between static and dynamic risk factors. While statistical precision is often used to justify these tools’ application, such justification can obscure how predictive outcomes are operationalized in legal decision-making. For example, DeMichele et al. (2024) found that using an abbreviated criminal history in the Public Safety Assessment reduced racial disparities compared to a full lifetime record. This refinement suggests a path toward balancing public safety and equity by avoiding the use of historical justice involvement as a proxy for individual risk, especially for Black youth overrepresented in system data.

Black youth, in particular, face elevated assessed risk levels due to patterns of racialized surveillance and enforcement that inflate prior system contact (Skeem & Lowenkamp, 2016). This study contributes to the equity discourse by showing how risk assessment instruments—despite their empirical sophistication—may inadvertently reproduce racial disparities when they rely on static indicators like prior justice involvement. By examining how race, detention history, and neighborhood disadvantage interact to shape recidivism outcomes, this analysis exposes structural vulnerabilities in risk classification frameworks. These findings underscore the need for reforms that move beyond predictive accuracy to interrogate the social consequences of algorithmic decision-making, especially for youth navigating structurally disadvantaged environments.

The counterintuitive finding that Black youth were significantly less likely to be rearrested despite their disproportionate representation in detention and structurally disadvantaged neighborhoods underscores the limitations of risk assessment tools and the complexity of interpreting recidivism data. This outcome challenges assumptions embedded in predictive instruments, particularly those that rely on static indicators such as prior justice involvement, which may reflect systemic bias rather than actual behavioral risk (Tonry, 2010). Such indicators often encode racialized expectations, reinforcing presumptions of reoffending without accounting for structural disadvantage.

This study embraces such scrutiny of the predictive validity of risk assessment tools in relation to critical evaluation of the dynamics of society impacted by legal decisions. The application of risk assessment tools should involve separating individuals from the environments that condition them by reframing violence not as inherent pathology but as a consequence of structural deprivation, economic exclusion, family disruption, and concentrated disadvantage (Clark, 1965; Currie, 2020; Currie et al., 2015; Shihadeh & Steffensmeier, 1994). If the Florida risk assessment tool overestimates risk for Black youth, it may result in heightened supervision or custodial placement, which paradoxically reduces the opportunity for rearrest during the follow-up period. In this sense, lower rearrest may reflect constrained exposure rather than reduced risk.

Moreover, racial bias in policing and arrest practices further complicates the relationship between actual offending and recorded system contact (D’Alessio & Stolzenberg, 2003; Lantz et al., 2023). These distortions highlight the need for caution when using administrative outcomes to infer behavioral risk. Rather than undermining the thesis of systemic disadvantage, the observed rearrest pattern reinforces the importance of contextualizing risk within a system shaped by racialized surveillance, environmental harm, and predictive bias.

Beyond individual indicators, broader structural factors such as neighborhood disadvantage and social disorganization contribute to cumulative risk. Though often excluded from formal assessment models, these contextual variables interact with race and class, compounding systemic inequality. The significant moderating effect of social disorganization in this study underscores this dynamic highlighting the need to understand risk not just as an individual trait but as a product of environment and opportunity structures.

Ultimately, these findings support the rejection of the null hypothesis and challenge the field to reconsider what risk scores truly represent. Scholars and practitioners increasingly advocate for reforms to promote fairness and mitigate racialized harm, such as removing lifetime criminal records from assessments and prioritizing rehabilitative, dynamic factors. Risk assessments must evolve from tools of punitive forecasting to instruments of restorative decision-making, capable of identifying underlying needs such as family instability, community-level deprivation, or lack of support systems. Future research and policy must prioritize both statistical validity and racial equity, ensuring that tools intended for safety do not become instruments of exclusion.

Limitations

This study is subject to several limitations that warrant consideration. First, there is the potential for researcher bias. As the principal investigator is a Black scholar examining outcomes affecting Black youth, there exists an inherent risk that personal investment may influence the interpretation of findings. To mitigate this, analytic objectivity was maintained through adherence to empirical standards and professional distance. The observations presented are grounded in dispassionate analysis and do not reflect the author’s lived experiences.

A second limitation involves the structural design of the dataset. The use of repeated observations and clustered measurements introduces challenges related to within-subject correlation and statistical dependence. To address this, the GEE framework with an AR (1) working correlation structure was employed to adjust for dependency across repeated measures. Future research would benefit from more granular, disaggregated datasets to support deeper, more nuanced, and representative modeling. Continued empirical inquiry is necessary to refine risk assessment methodologies and to advance both analytic precision and racial equity in their application.

Admittedly, the current study may not reflect the same level of analytical rigor as Wolff et al. (2023), particularly regarding the complex task of tracking dynamic risk and protective factors across individual circumstances and community-level contexts. However, this study makes a distinct contribution by foregrounding the principles of fairness, equity, and justice that underpin the ethical administration of the criminal legal system. It expands the literature not by replicating methodological sophistication, but by interrogating how risk is operationalized and interpreted across race, class, and other structural dimensions, reaffirming the need for equal protection under the law irrespective of background or identity.

Hispanic youth were underrepresented in the analytic sample due to exclusion criteria requiring a full sequence of four C-PACT assessments. Several factors may account for this disparity. First, language barriers and cultural mismatches in service provision could have impacted continuity in assessment, particularly for youth from Spanish-speaking households or immigrant communities. Second, higher rates of residential mobility, often associated with economic instability or family reunification, may have disrupted sustained justice system engagement, limiting the opportunity to complete repeated assessments. Third, differential referral patterns or diversion practices may have resulted in lower system penetration for Hispanic youth, reducing their likelihood of reaching the necessary assessment threshold. Finally, data collection bias or missing demographic information (e.g., misclassification or lack of language support during intake) may have contributed to exclusion during preprocessing. These factors highlight the need for more inclusive sampling strategies and culturally responsive protocols to ensure equitable representation and analytic validity in future research.

Furthermore, Additionally, the one-year observation period may underestimate rearrest risk for youth in extended custody, as long-term residential placement reduces exposure to community-based risk factors. This timeframe may underestimate recidivism risk for youth placed in long-term residential custody, as extended institutional placement reduces exposure to community-based risk factors during the follow-up window. In effect, youth in custody may appear less likely to reoffend, not because of reduced risk, but due to limited opportunity for system contact while confined. Future research should consider longer follow-up periods or alternative measures of post-placement risk to better capture the effects of sustained institutional contact.

One notable limitation involves the study’s exclusive focus on youth who received community-based sanctions, such as probation or diversion, rather than those placed in residential custody or pretrial detention. While this decision aligns with the dataset’s structure and facilitates longitudinal assessment across repeated measures, it also narrows the analytical scope. Youth subject to more restrictive dispositions may exhibit different risk patterns, levels of system exposure, or contextual interactions that this study does not capture. Additionally, youth diverted to community placements often represent a subset with lower assessed risk or different demographic profiles, potentially limiting generalizability to the broader juvenile justice population. Future research should expand to include system-involved youth across varied disposition levels to better assess how risk assessment tools function across the full spectrum of juvenile justice settings.

6. Conclusions

Risk assessment practices in contemporary criminal justice are often anchored in predictive models that prioritize public safety through punitive decision-making, frequently leading to extended detention or custodial sentences for those deemed high risk. By contrast, a criminogenically informed framework emphasizes rehabilitation, directing attention toward addressing the underlying causes of reoffending through targeted interventions, including therapeutic and restorative practices (Barnes-Lee et al., 2023).

Alternative models, such as the Risk–Need–Responsivity (RNR) framework, advocate for a needs-based approach. The RNR model is designed to align intervention intensity with individual risk levels, identify criminogenic needs, and tailor strategies to the responsiveness of each youth (Andrews et al., 2011). Within this framework, risk assessment becomes a tool for support, not surveillance, especially when applied to justice-involved juveniles.

Crucially, predictive instruments should not serve as vehicles for penalizing youth based on immutable characteristics, such as race or historical justice involvement. When static risk factors are treated as neutral, they risk perpetuating inequities embedded within the criminal legal system. Rather than justifying further punishment for youth who have already served time, assessments should surface needs and guide rehabilitative solutions.

Fourth-generation tools, as outlined by Andrews et al. (2006), advance this vision by integrating both static and dynamic factors in ways that capture individual and contextual complexity. Empirical research supports this concern. Rehavi and Starr (2012) found that 80% of racial disparities emerge prior to charging decisions, and sentencing gaps persist even after controlling for offense type, with Black individuals receiving disproportionately longer sentences. Moreover, Black males aged 18–19 are incarcerated at rates 12.7 times higher than their White counterparts, revealing structural inequities inherent in risk assessment frameworks (Sreenivasan et al., 2022).

A narrow focus on past conduct, as Harcourt (2015) cautions, reduces complex human behavior to statistical probabilities and may foreclose more just, rehabilitative pathways. To advance equity, risk assessments must shift from instruments of exclusion to instruments of opportunity with tools that illuminate, rather than obscure, the humanity and potential of those they evaluate. The goal is to enhance the use of these instruments to achieve predictive accuracy while preserving equal justice and fairness in criminal justice administration.