Abstract
This study documents the extent to which prosecutors and judges in New Jersey deviate from new case review protocols post-bail reform. We draw on a random sample of hearings observed in two courts (N = 372) to (a) map prosecutorial requests to depart from risk-based recommendations for pretrial monitoring and (b) explore patterns in judicial review signaling more limited case processing. Findings show that in 16% of cases, prosecutors deviated from guidelines and most of these motions were adopted by judges to increase supervision. Deviations from process by judges were more frequent, with 39% of cases observed linked to one or more markers of reduced review. Multivariate results suggest that case-level factors influenced the patterning of these departures in different ways, with older defendants and most Latinx defendants receiving a more limited assessment of their cases. We discuss findings in connection to literature on corrections reform and smart decarceration initiatives.
Cash bail is changing. In the past 10 years, states across the country have modified key protocols related to monetary conditions and other rules governing pretrial release and case review (National Conference of State Legislatures, 2018). Although there is heterogeneity in strategies for implementation, these reforms have been prompted by similar concerns about disparities in process and punishment (Koepke & Robinson, 2018). As noted by research and policy, pretrial incarceration is a critical component in the reorganization of justice systems, given its cumulative effects where persons trapped into cycles of poverty and detention are more likely to receive harsher decisions (Cohen & Reaves, 2007; Gupta et al., 2016).
There are “bright plots” of evidence pointing to the success of new initiatives targeting cash bail practices and other preadjudication decisions (New Jersey Judiciary Court System [NJJCS], 2019; Stevenson, 2018). However, the empirical foundation describing emerging practices is surprisingly thin. In particular, studies of the implementation of new models remain scarce (Stevenson, 2018, p. 336), with adaptation to updated mandates remaining “sorely underexamined” (Koepke & Robinson, 2018, p. 1805). With some exceptions (DeMichele et al., 2018; Eckhouse et al., 2019), little is known about the dynamics of pretrial hearings across local actors and jurisdictions including the evolving influence of case-level legal factors—statutorily relevant information such as the seriousness of the charges—and extralegal factors—variables such as a defendant’s race or gender—that are excluded from legal decision-making. These “blind spots” are not specific to reforms of pretrial systems. Process remains a critical gap in the implementation of evidence-based policies (Viglione, 2017, p. 1357) and smart decarceration programs (Pettus-Davis & Epperson, 2015), despite studies noting that line-staff or managers condition the transformative potential of new models (Lynch, 1998; Rengifo et al., 2017; Rudes, 2012). The shortage of courtroom-based data and theory is acute in the case of prosecutors and defense attorneys relative to judges (Laskorunsky, 2018), with little known about how new models shape their exchanges and behaviors (Yule & Schumann, 2019).
Addressing these issues is critical to map the mechanisms that may account for the relative success of pretrial reform. A more explicit focus on courtroom actors and interactions also enables the examination of the path-dependent routines that blend change with established practices (Beckett, 2018; Verma, 2015). This specification is important, given that the adoption of new policies is patterned by strategies that are “more complex and contradictory than much of the theoretical literature has assumed” (Hannah-Moffat et al., 2009, p. 393). This includes, for instance, changes in case-level cues used to assign danger or blame, and their integration with broader forms of institutional bias (Gonzalez Van Cleve, 2016; Omori & Lautenschlager, 2018).
This study contributes to this literature by exploring how case processing in the courts has been reshaped by a reorganization of pretrial protocols. We answer two interrelated questions: First, how do judges and prosecutors adapt their actions and exchanges to new guidelines for open case review and pretrial monitoring decisions? Second, to what extent are these responses shaped by defendant-level legal or extralegal factors? To address these questions, we document departures in a random sample of supervised release hearings observed in two New Jersey courts. In particular, we map prosecutorial requests to modify monitoring levels recommended by a new risk assessment process and examine their subsequent adoption. In addition, we measure the range of engagement by judges through case review protocols during hearings, such as whether they openly communicated risk scores and monitoring conditions to defendants. These two forms of circumvention reflect specific practices tied to the new local decision-making framework and to the functional roles of prosecutors and judges. Whereas the former act by formulating supervision requests, the latter act by adjudicating motions and framing case processing. More generally, our approach seeks to better specify how exchanges among courtroom actors not only “jointly produced discretionary decisions” (Ulmer, 2019, p. 485) but also anchor new regimes of “risk governance” (Hannah-Moffat, 2018, p. 7).
Background
The Evolution of Bail Practices in New Jersey
Prior to 2017, pretrial decisions in New Jersey seldom took the form of unconditional release: One study estimated the release on recognizance (ROR) rate at 29% drawing on statewide records corresponding to a week of felony cases (Kazemian et al., 2013). This figure was significantly lower in a more recent study based on observations of selected jurisdictions (8%) (Rengifo et al., 2020). Estimates from these studies also indicate that bonds tended to be set at remarkably high amounts—as much as US$7,500, median average—with one third of defendants remaining detained due to their inability to pay bail (VanNostrand, 2013). Critically, Black and Latinx defendants were more likely to be denied unconditional release and more likely to receive higher value cash bails relative to Whites (Kazemian et al., 2013; Mauer & King, 2007).
A new Joint Committee on Criminal Justice was tasked with developing recommendations in response to this “broken” system (Friedman, 2014). Their work provided the foundation of the Criminal Justice Reform Act that virtually abolished cash bail in 2017 by creating a presumption of release for most offenses. This statute also replaced the advisory offense-based schedule governing pretrial decisions with a risk-based decision-making framework that tied new recommendations of pretrial monitoring to the Public Safety Assessment (PSA) developed by Arnold Ventures (2019). This realignment also revised time frames for first appearances and created new roles for pretrial units. The new detention hearings reserved for cases motioned by the prosecutor for incarceration were set to involve an extended process of judicial review including witness interviews and a preliminary examination of evidence. Pretrial release hearings for all other cases remained structured similarly to pre-reform standards with sessions consisting of short case reviews managed by a judge.
More than 80,000 defendants were processed in the first 2 years of the new system. About 7,000 of these defendants were released on recognizance, and another 40,000 were referred by the prosecution for detention consideration. In 2018, nearly 34,000 cases were reviewed for pretrial release, with about 3% receiving monetary conditions and another 35% receiving pretrial monitoring levels (PMLs) of 3 or 3+ (Grant, 2018; NJJCS, 2019). Thus, consistent with expectations, cash bail has been effectively eliminated, and the most restrictive decisions (detention or enhanced monitoring) are no longer a dominant system feature. Crucial to the initiative was to show that the expanded use of conditional releases did not increase failures to appear or new crimes by defendants awaiting trial. This has not been the case: About 90% of defendants made their court appearances in 2017, a figure consistent with appearance rates under the more restrictive, pre-reform regime (Grant, 2018). Jail crowding also eased considerably, with statewide populations decreasing by about 2,000 between 2016 and 2018 (Grant, 2018).
Despite these encouraging results, early assessments also pointed to a host of persisting challenges. While the new law prompted the temporary release of many individuals who would otherwise remain incarcerated due to their inability to pay bail, it did not result in a significant expansion of unconditional releases. Some racial gaps narrowed, such as length of pretrial detention and time between complaint issuance and release; however, other disparities remained, including the overall racial composition of jail populations (Grant, 2018). Most critically, the ambitious system-level design of the reform is yet to be contrasted with a microlevel account of its implementation (the reform’s “bright plots” and “blind spots”). For example, it is unclear the extent to which the new central role given to risk assessment is being considered in the practice of judicial review and whether attorneys and judges tend to agree with its recommendations regarding monitoring levels. These gaps are salient to pretrial release hearings as motions and review protocols are not tracked systematically, reinforcing the idea that both process and decisions tied to these specific sessions are little more than a mechanic process with no substantive meaning. Below we explore how prior research conceptualizes case processing as a routine shaped by action and reaction to new policies.
Court Process and Decision-Making
A long tradition of scholarship describes the practice of process in the courtroom in terms of markers of judicial engagement or activity (Thibaut & Walker, 1978). The deployment of these and other micro-behaviors denoting interest or interventionism has been tied to various judge-level attributes (Eisenstein & Jacob, 1977; for other perspectives, see Blanck, 1996; Ottone & Scott-Hayward, 2018). Our study builds on this literature by considering how differences in functions of judges and prosecutors ground their case review practices in open court and relative adherence with the revised pretrial guidelines. To support this perspective, we integrate studies on judicial behaviors with contributions that describe dynamics of policy change and adoption.
Departures in Formal Decisions
A conventional strategy to examine judicial activity is to document the correspondence between schedules and decisions. For instance, research on the circumvention of sentencing guidelines shows how judges’ variation in conformity to recommendations is sensitive to factors such as the defendants’ sex offense status or evidence of mental health issues (Engen et al., 2003; Wormith et al., 2012). Studies have also noted that departures are often triggered by prosecutor requests (Kaiser & Spohn, 2018), prompting researchers to examine decisions and other judicial behaviors as collective processes (Kutateladze et al., 2014; Phillips, 2004; Ulmer, 2019).
There is some evidence of judicial circumvention in pretrial adjudication, in part because pretrial guidelines tend to be less codified relative to sentencing guidelines and their application is moderated more explicitly by the local workgroup (Jones & Goldkamp, 1991). In some cases, release recommendations by pretrial agencies tend to be dismissed by magistrates in favor of requests by attorneys—particularly prosecutors (Suffet, 1966, but see Nagel, 1983). More recent studies also indicate that judges show uneven conformity to recommendations tied to risk assessments. Hilton and Simmons (2001), for example, find that scores seldom ground detention decisions for mentally ill individuals. Instead, magistrates use more familiar sources seen as more credible like clinicians’ judgments and testimonies. Another study finds that judges in Virginia adopt assessment recommendations in two thirds of the cases examined, with reasons for departures remaining “something of a mystery” due to poor documentation (Ostrom et al., 2002, p. 3).
Some studies further show that judges and prosecutors also waver in the support of the PSA. DeMichele and colleagues (2018) find that court officials surveyed as part of a multistate convenience sample perceive factors embedded in the actuarial process—like pending charge—to be important in determining pretrial release. Yet, only 48% of judges report that the risk instrument informs their consideration of release decisions and many further believe that the tool contributes to systemic racial disparities. Similarly, Stevenson (2018) draws on court data to show how patterns of implementation of the PSA in Kentucky varied across counties, with little overall change in state-level racial disparities pre- and post-reform. Moreover, Stevenson’s (2018) assessment shows that despite bipartisan support and extensive consultations with stakeholders, judges deviated from PSA recommendations in as much as two thirds of cases (p. 373).
While these studies echo research on implementation issues tied to departures in other contexts (Cohen, Pendergast, & VanBenschoten, 2016; Schmidt et al., 2016), they rarely frame circumvention in formal decisions as indicative of resistance to new policies. Instead, departures motioned by prosecutors—prompted more directly by judicial intervention—are usually seen as compliant with exemptions or motivated by (largely stable) focal concerns related to the perceived dangerousness or blameworthiness of defendants (Johnson et al., 2008; Lehmann et al., 2019; Stemen et al., 2015). The notion of circumvention as resistance has been more systematically applied to other criminal justice actors such as probation or corrections officers (Miller & Maloney, 2013; Rengifo et al., 2017; Rudes, 2012) perhaps due to the reduced scrutiny of judicial actions or the low rate of reforms that limit the study of cycles of change and adaptation. However, it is likely that reforms marked by a structured process of decision-making or reduced involvement of key participants can elicit similar adaptation strategies that undermine implementation across all justice institutions (Werth, 2017; Wormith et al., 2012).
Taken together, these studies show that agreement with new mandates among judges and other criminal justice actors is far from rare, or uniform, and that departures from guidelines tend to be sensitive to both case attributes as well as other attributes that reflect changes in actor-level authority and position within workgroups (Werth, 2019). We draw on this research to examine a formal action that signals circumvention from reform guidelines through departures from PSA-recommended supervision levels. However, unlike other studies, we focus on motions by the prosecutor and their subsequent adjudication by judges and examine how these decisions are shaped by both case-level factors as well as courtroom-based exchanges. We also expand prior contributions by studying nonconformity beyond pretrial decisions to include judicial adoption of new case processing guidelines in open hearings. Below we situate this perspective in the broader literature on markers of open judicial review.
Departures in Case Review
Studies on judicial decision-making and courtroom protocols vary in how they measure quality of case review with some studies documenting the participation of relevant parties, and others focusing on broader forms of procedural justice. Research on pretrial hearings indicates that between one third and one half of these proceedings are typically decided without feedback from attorneys (Allan et al., 2005; Ottone & Scott-Hayward, 2018; Suffet, 1966; Yule & Schumann, 2019). Other contributions suggest that these hearings also lack proactive engagement by judges (Anleu & Mack, 2015; Rengifo et al., 2020).
Findings from these and other studies suggest that judges are more likely to deviate from new case review practices not only if these are seen as reducing their discretion or autonomy but also if new practices are interpreted as adding additional layers of work or oversight or reducing their local authority (Hannah-Moffat et al., 2009; Johnson, 2005; Jones & Goldkamp, 1991; Lynch & Omori, 2014). Specific strategies of nonconformity range from narrow case review (Anleu & Mack, 2015) to spotty record-keeping (Kaiser & Spohn, 2018; Ostrom et al., 2002). A similar set of strategies has been documented in connection to other justice reforms that add structure to specific routines. For instance, Miller and Maloney (2013) conceptualize the relative completeness and usability of case management notes by probation officers as markers of their compliance with new supervision and classification protocols. This is consistent with studies suggesting that deviations from process-exchanges may have more opacity for specific cases (Laskorunsky, 2018; Rengifo & Fowler, 2016), and when new mandates are seen as illegitimate or contrary to perceived institutional values (Lynch, 1998; Rengifo et al., 2017).
Research has also shown that varying versions of process may be implemented by justice workers according to specific attributes of alleged offenses or people arrested. In Gonzalez Van Cleve’s (2016) ethnography of courts in Chicago, the “mope” is a particular type of defendant seen by courtroom actors as fully responsible for their failures (p. 134), and as such, “unworthy of basic due process rights” (p. 62). In Yngvesson’s (1988) study of a district court in Massachusetts, court officials deploy a range of processing depending on whether cases are seen as low-level “garbage” (p. 414) or deserving of full-fledged legal consideration and process. Other case-level characteristics, such as whether defendants have formal legal representation, have been found to positively influence case processing patterns such as the overt use of explanations for sentencing decisions by judges (Anleu & Mack, 2015). This can be seen in social welfare policy-making in the labeling of the “deserving”—who are seen as individuals struggling as a result of their environment—and the “undeserving”—who are seen to have contributed to their own hardships through personal choices (Applebaum, 2001). Similar rhetoric of inferiority or dangerousness is mobilized to support the harsh treatment of defendants by court officials (Goldson, 2002; Gonzalez Van Cleve, 2016) and police (Rengifo & Pater, 2017).
The deployment of varying forms of decision-making across cases and contexts is enabled by systems that favor discretion by judges and other authority figures. In these settings, visible characteristics of defendants may be used by attorneys as cues of dangerousness or moral worth to shape upward or downward departures, and by judges as proxies for case complexity when managing case review. Justice reforms anchored on the use of actuarial instruments such as New Jersey’s new model of pretrial justice are likely to trigger varying forms of circumvention by court actors as they limit the discretion that has long-buttressed local patterns of case processing and adjudication tied to decarceration and due process ambitions. These reforms also alter the substantive nature of judgments made by courtroom workgroups by outsourcing recommendations of supervision to risk scores and by narrowing decisions and protocols for case review. Thus, while evidence-based approaches contribute to decarceration initiatives by creating a more informed environment for decision-making, these efforts may be limited by the resistance they generate, and the more automated use of priors for arrest and other factors that downplay the impact of a criminal justice system encumbered with racial bias (Eckhouse et al., 2019; Gonzalez Van Cleve, 2016; Kutateladze et al., 2014; Rengifo & Stemen, 2015). Our work on departures from case review contributes to the growing literature on new assessments of justice reform by examining “less visible” forms of circumvention (Johnson, 2005, p. 767), as an apparent lack of structure or scrutiny may signal contrasting patterns of bias or compliance.
Current Study
This study answers two research questions: How do prosecutors and judges adapt their actions and exchanges to new guidelines for case review and pretrial decision? And, to what extent are these responses shaped by defendant-level legal or extralegal factors? To address these questions, we map two distinct types of departures from reform guidelines: First, we document prosecutor requests to increase recommended pretrial release levels (upward departures vs. no departure) and track subsequent challenges by the defense and/or adoption by judges. Second, we study judges’ engagement in case review as levels of departure from case processing during hearings—specifically, whether judges neglect to communicate PSA scores or specific conditions of temporary discharge to defendants, and whether they dismiss the involvement of prosecutors/defense attorneys (comprehensive vs. levels of limited review).
We argue that the examination of these actor-level behaviors provides a novel, ground-level assessment of implementation issues tied to smart decarceration initiatives centered on curbing pretrial detention. Importantly, our study provides contrasting perspectives of the form and use of discretionary power by courtroom actors in an evolving institutional context—one based on “decisions” defined by motions by prosecutors and another more closely linked to “process” defined in terms of judges’ approach to case review. Given the exploratory nature of our research, we do not test specific hypotheses. However, we consider that the actuarial foundation of the reform—the limitation of actors’ discretion and visibility—may have prompted patterns of circumvention among prosecutors and judges akin to others observed in the sentencing literature (Ulmer, 2019). Similarly, it is plausible that prior findings on case-level and institutional drivers of racial disparities in judicial sanctions and treatment remain visible in new contexts marked by risk-based decision-making (Koepke & Robinson, 2018; Omori & Lautenschlager, 2018).
Method
Design and Setting
To address our research questions, we draw on observational data collected in pretrial release hearings in the courts of Union and Hudson counties. These two courts operate in rather similar socioeconomic contexts. According to the 1-year estimates of the American Community Survey (U.S. Census Bureau, 2018), for example, the local unemployment rate is between 7% and 8%, and about one in four persons receive social security income. Importantly, the majority of the resident population is Latinx or Black, and combined, these two jurisdictions processed approximately 15% of all reform-eligible defendants in 2018 (NJJCS, 2018).
Under the new system of pretrial justice adopted in New Jersey, release hearings determine the specific monitoring level and other conditions for people detained on most charges. 1 The reform’s decision-making framework tied matrix-type format-specific recommendations for PMLs to specific ranges of PSA scores. The PSA instrument was designed to collate personal information on legal factors such as pending charges, severity of current offense, prior violent convictions, and prior failures to appear. Only one extralegal factor—age—contributed to the assessment. Ratings were set to be maintained by pretrial services staff, scoring each defendant separately for failure to appear (FTA) and new criminal activity (NCA) on a scale of 1 through 6 (higher scores signaling higher risk), and a binary flag marking whether the predicted NCA was violent in nature. The new PSA-based PMLs ranged from ROR to detention, with four intermediate supervision levels (PML 1, PML 2, PML 3, and PML 3+) denoting reporting frequency (bimonthly, monthly, weekly) and medium (phone, in-person). For example, the recommendation of ROR was associated with NCA/FTA scores of 1 to 2, with higher score ranges linked to more intensive supervision levels (NJJCS, 2018).
The new model enacted in New Jersey preserved some of the old courtroom-based protocols for hearings. For example, the open review of individual cases was triggered in similar ways by the calling of a case by the docket number or the defendant’ s last name, followed by a public reading of charges, warrants, or detainers. Next, judges were expected to explain the risk assessment process to defendants and to communicate specific scores and recommended monitoring level. At that point, they opened the floor for attorney participation or they moved forward with the announcement of a formal decision. Feedback by attorneys was explicitly prompted by action of the judge, and whenever used, it involved the confirmation of the monitoring level already mentioned, or the request of a new one. Sessions typically ended with judges communicating release conditions and noting the date of the next court appearance.
Sampling
Given the impossibility of relying on a conventional sampling frame due to a lack of public records of first appearances in New Jersey, we used regularly scheduled open court sessions as our primary sampling unit. Upon random selection of these sessions during our fieldwork period, we included all cases reviewed with no restrictions placed on the inclusion of cases across the mode of appearance, charges, or PSA score. Overall, we aimed to observe 30 sessions per jurisdiction between November 2018 and February 2019, to replicate the sampling strategy and approach of a separate set of pre-reform observations conducted as part of the broader comparative study (Rengifo et al., 2020). Consistent with this strategy, we attended 59 sessions across the two study sites for a total of 65 hours of observed proceedings. Our fieldwork observations represented approximately 10% of all eligible sessions for observation per jurisdiction (of 292 sessions scheduled in a typical year).
Data Collection
To record case outcomes and protocols, fieldworkers used a paper-and-pencil checklist modeled after similar studies and adapted using feedback by local actors and pilot data (Gonzalez Van Cleve, 2016; Phillips, 2004). 2 The training of raters involved the use of a fieldwork manual with detailed coding instructions. Prior to fieldwork, raters visited each court twice, with briefings scheduled before and after each session. The first of these sessions focused on open-coding and journaling strategies, courtroom access, location, and an overview of the checklist instrument. The second session involved the coding of actual cases and a review of issues associated with speed, accuracy, and consistency in reporting.
The paper-and-pencil checklist was designed to capture general protocols for each court session (on-time/late opening, courtroom actors present, etc.), as well as anonymized case-specific information on matters involving defendants in custody. In total, we compiled observations of 59 randomly selected court sessions yielding a total of 399 individual cases (27 sessions and 298 cases in Hudson, 32 sessions and 101 cases in Union). Most hearings resulted in a formal decision, although, on occasion, scheduled hearings were postponed or interrupted.
For this article, we focused on the subset of cases observed with a formal decision involving an ROR or a release with PMLs 1 to 3/3+ (N = 372). 3 This includes cases that fit the formal criteria for detention consideration, given the reform’s decision-making framework, but were redirected for pretrial release review by the prosecutor’s office. Approximately 12% of the cases were observed by two or more observers to verify the quality of our observations. Measures of interrater reliability reached a minimum of “substantial” convergence of defendants’ demographic information, decisions, and interventions, with kappa estimates ranging from .76 to 1.00 for categorical variables involving departures and the race/gender of defendants (Landis & Koch, 1977).
Measures
Our dependent variables captured the following set of actor-specific practices:
Prosecutors—Upward departures are a dichotomous measure indicating whether the prosecutor motioned the override of the PSA-recommended monitoring level to intensify community-based monitoring (1 = Yes, 0 = No; M = 0.094, SD = 0.292). We also track the adoption of these requests by presiding judges (1 = Yes, 0 = No; M = 0.743, SD = 0.443).
Judges—Limited case review is a count measure that included three explicit behaviors by judges during hearings directly linked to the new protocols of pretrial review. A score of 1 captured the absence of the behavior (0 = presence of the behavior) across the following set of independent activities: Whether the judge (a) communicated any risk assessment scores to the defendant, (b) communicated any conditions of supervised release beyond the frequency/type of reporting requirement, and (c) prompted both the intervention of the prosecution or the defense. 4 The resulting variety score of judicial departures associated with the processing of defendants ranged from 0 (comprehensive case review) to 3 (limited case review; M = 0.503, SD = 0.692).
We modeled variation in these actions as related to key extralegal factors of defendants including observed sex (1 = Female, 0 = Male), and age (5-point category from 1 = Age 16–25 to 5 = Age 55+). We specified the defendants’ race/ethnicity using dummy variables that identified them as non-Latinx Black or Latinx. Non-Latinx White is used as the reference category. 5 To expand our assessment of potential markers of social stratification in the courts we divided Latinx defendants as those that required a language interpreter and those who did not (all interpreter-mediated events we observed involved the use of Spanish language). We also coded the number of charges mentioned using a three-point category (from 1 = One charge to 3 = Three or more charges), and nature of the top, most serious charge (1 = Person or weapons offense, 2 = Property, 3 = Drugs, 4 = Order/other offenses). In addition, we recorded whether warrants/detainers are mentioned during the hearing in connection to the defendants (1 = Yes, 0 = No), and capture the relative length of each hearing in minutes. Risk assessment scores, whenever mentioned, are recorded separately for FTA, or NCA (each scored from 1 to 6). Consistent with the structure of the PSA, we recorded whether it was mentioned that a defendant was deemed at risk of new violent criminal activity (1 = Yes, 0 = No).
We also included an 8-item judicial behavior scale that captured the general engagement of the presiding judge with defendants. This approach allowed us to analytically separate general case processing activity by judges from the more specific scripts tied to the processing of the new pretrial hearings. The scale included dichotomous items that captured whether the judge (a) engaged the defendant off their regular script, (b) made visual contact with the defendant, (c) checked on their work/education status, (d) asked the defendant whether they understood the process, (e) asked the defendant whether they understood the decision, (f) explained to the defendant any penalties for noncompliance, (g) signaled interest in the defendant’s ultimate success/well-being, and (h) referred the defendant to social services or programs. 6 Each of these items was coded using a bivariate, close-ended response (1 = Yes, 0 = No). These items were added together to create a variety score, with higher scores signaling a more engaged review of cases. Finally, our main models included judge fixed effects with supplementary models using a dummy specification to account for broader unobserved differences across the two study sites.
Analytical Approach
Our analyses proceed as follows: We first describe the nature of the sample and summarize patterns in formal motions and decisions, and case review protocols. Next, we predict variation in the two measures of circumvention described above drawing on case characteristics. Given differences in levels of measurement, we specify a logistic model to explore departures by prosecutors (1 = Upward departure, 0 = No departure) and a Poisson model to explore constraints to case review (from 0 = Comprehensive review to 3 = Limited review). Both sets of models, however, use the same set of common case-level covariates, fixed effects, and both predict the likelihood of a more restrictive use of discretion (a departure to enhance monitoring or to limit case review).
Results
Our study sample consisted of observed hearings with a pretrial decision (N = 372). As shown in Table 1, few defendants were female (15%), and most were relatively young (55% coded as being younger than 35 years, in our 5-point categorical scale). About 48% of defendants were classified as non-Latinx Black and 14% as non-Latinx White. As noted in the previous section, we split the Latinx category into two groups, given that all language-mediated hearings we observed involved a Spanish interpreter. Of all defendants observed, 8% were Latinx with an interpreter and 31% were Latinx with no interpreter.
Descriptive Statistics
About a third of the cases in the study sample were associated with a person or weapon offense as the top charge (36%), followed by drug-related charges (21%) and property crime charges (21%). In a third of the hearings observed, current charges were not mentioned. In part, this may have been due to the fact that case review involved short exchanges (M = 4.092 min, SD = 3.565) and judges were not particularly engaged with defendants (M = 2.532 in the 8-point scale of judicial treatment). It is worth noting as well that final decisions were skewed toward the more restrictive options—35% for PML 3/3+ versus 24% for ROR.
Our data show that prosecutors motioned for a departure in recommended monitoring levels in about 16% of the cases observed, with the majority consisting of upward departures (n = 35, or 9.4%). In contrast, we did not record a single case where the defense sought to depart from the monitoring recommendation tied to the risk assessment process. Similarly, a substantial fraction of upward departures requested by the prosecutor were uncontested by the defense (n = 13/35, or 35%), which explains, at least in part, the high rate of adoption by judges (n = 26/35, or 74%). For downward departures, we found complete alignment between prosecutorial requests and adoption by judges (n = 24/24). Findings also showed that judges circumvented the base case review process in 39% of the cases, as they skipped one or more of the three markers of abridged case processing considered by the variety score (M = 0.503). The prevalence of these markers was rather uniform across items, from judges failing to read conditions of release in 16% of the cases observed to judges neglecting to prompt attorney feedback in 18% of the cases. Thus, in relation to our first research question, we find that prosecutors depart with some regularity from recommended monitoring levels, and these motions carry over in the form of final decisions by judges. In addition, results show that circumvention through the case review process is more prevalent than those related to prosecutorial motions.
At the bivariate level, we did not find a systematic association between the race/ethnicity of defendants and departures by judges or prosecutors. However, case processing protocols for Latinx defendants trended toward more restrictive actions, and these patterns differed remarkably across interpreter usage and courtroom actor. Whereas a higher fraction of Latinx defendants with an interpreter were subjected to prosecutorial motions to increase their pretrial release level (30% vs. 7%–9% across other groupings by race/ethnicity), Latinx defendants without an interpreter received a more limited review of their cases (M = 0.625 vs. M = 0.414 across other groups).
Table 2 summarizes the results of our multivariate models predicting departures by prosecutors and judges using case-level factors and judge fixed effects. These models are set up to answer our second research question on potential links between case-level legal and extralegal factors to action by prosecutors and judges. On the left panel of Table 2, we show estimates of upward departures by prosecutors derived from a logistic model. On the right panel, we show estimates of case review departures by judges derived from a Poisson model. Coefficients are presented as odd ratios to facilitate their interpretation. All models rely on the same set of predictors, which are entered sequentially to account first for associations between the departure outcomes and extralegal factors, and next for a broader set of covariates.
Regression Models Predicting Departures by Prosecutors and Judges
Note. We omit judge-level fixed effects. For Models 1 to 2, the pseudo R2 corresponds to the Nagelkerke R2. For Models 3 and 4, the pseudo R2 corresponds to the McFadden R2. In Model 2, we drop 28 cases due to a lack of variation in the outcome associated with one of the observed judges in the sample. OR = odd ratio; RSE = robust standard error; IRR = incident rate ratio; AIC = Akaike information criterion; BIC = Bayesian information criterion; GOF = goodness-of-fit.
p < .05. **p < .01. ***p < .001.
Results from Models 1 and 2 show that upward departures by prosecutors are hardly associated with any of the variables included in the models; although combined, the fuller specification correctly classified about 91% of the cases in the sample. We do find that longer proceedings are significantly associated with a greater likelihood of departures, which may imply a proxy association between case complexity and nonconforming behaviors for prosecutors. However, this relationship is nonsignificant when employing a more direct measure of case seriousness—number of pending warrants. Supplementary models based on a smaller sample of cases with more complete data similarly show that upward departures are insensitive to PSA risk scores or to the number or type of current charges but significantly higher in cases processed in Union county. 7 While the low number of cases prevented us from fitting a conditional model for the adoption of upward departure motions by prosecutors, bivariate associations tracked the patterns described above.
Models 3 and 4 in Table 2 describe a different response associated with judicial departures from comprehensive case review of pretrial release conditions. This variety score count was skewed toward zero as a large fraction of cases in the sample were processed in full compliance with the review protocols defined above (61%). Count models are well suited to model the distributional properties of this response variable. In particular, we rely on Poisson regression as we find no evidence of overdispersion, and model estimates are not significantly different from those generated by more complex models and, in fact, show a better fit to the data. As in Models 1 and 2, we report robust standard errors (RSEs) to mitigate other sources of bias, and specify the duration of the hearings (in min) as an exposure variable to account for different windows for the potential observation of courtroom behaviors.
Estimates listed in the right panel of Table 2 show that judicial review is more sensitive to case-level factors relative to departures by prosecutors. This is not surprising, given that the process of case review is less regulated and less scrutinized than decisions tied to formal departures. In particular, we find that older defendants, as well as Latinx defendants without interpreters, systematically have a more limited review of their release conditions (Model 3, incident rate ratio [IRR] = 1.267, RSE = 0.111 and IRR = 2.126, RSE = 0.590, respectively). These associations remained statistically significant in Model 4, which included additional legally relevant covariates. However, only general judicial behavior was related to variation in the count outcome, with a narrower review of pretrial release conditions associated with less engaged patterns of general judging (IRR = 0.824, RSE = 0.066). Results from Model 4 also show that Latinx defendants without an interpreter were more likely to have a more limited review of their release conditions (IRR = 1.699, RSE = 0.420). The fuller specification of these associations was also linked to better model fit as shown by the nonsignificant deviance parameter and the smaller Akaike information criterion/Bayesian information criterion (AIC/BIC) values.
We examined the robustness of our estimates to other specifications (e.g., right-censored Poisson, negative binomial models) and found that these did not change our findings and resulted in poorer fit. Unlike Models 1 and 2, for example, we did not find that case review varied across the two study sites. However, using the subset of cases with complete offense information, we find that defendants with less serious charges had reduced case review by judges (IRR = 1.271, SE = 0.094). 8 We also explored all substantively relevant two-way interactions and found that these were not significantly related to variation in departures by prosecutors or judges and did not improve the fit of the estimations. In addition, we find that patterns observed in connection to interpreter usage were not indicative of underlying differences in process or decisions tied to legal citizenship utilizing data from court observations where this information was systematically mentioned. Finally, we checked for influential cases and multicollinearity and did not find evidence of potential threats to our statistical routines.
Taken together, our multivariate results did not show that some case-level factors such as charge seriousness were related to departures by prosecutors or judges. However, we did find that some defendants do receive less process during their hearings. We unpack this observation in the next section along with broader considerations about pretrial reform and implementation.
Discussion
This study explored the extent to which actions by prosecutors and judges during pretrial hearings conformed to new protocols for decision-making enacted in New Jersey. Results showed that prosecutors departed from PSA-recommended monitoring levels in approximately 16% of cases in the sample, with most departures requesting increased supervision. Judges subsequently granted three of every four of these upward departure requests and all downward departure requests. In contrast, the defense played a more passive role with no departures from risk-based recommendations noted across cases, courts, or defendants, and little in terms of the number or effectiveness of their challenges to the prosecution.
Our results also show that departures from case processing by judges were more frequent relative to those observed by prosecutors, with 39% of cases linked to one or more markers of limited review of release conditions and other reform protocols such as the explicit review of PSA scores and the engagement with counsel. Estimates from our multivariate models further suggest that case-level factors influenced the patterning of these departures in different ways. Whereas formal decisions by the prosecutor were largely insensitive to factors associated with the relative seriousness of cases or the demographic characteristics of defendants, the limited deployment of process by judges was associated with cases involving older defendants and Latinx defendants without interpreters (relative to non-Latinx White defendants). Finally, we note that cases with fewer markers of general judicial behavior were also associated with a more limited use of specific release protocols during case review.
Our findings, though exploratory, confirm results from prior studies that show that judges and other bureaucracies signal uneven patterns of adoption of new institutional mandates, especially when these involve new policies that narrow discretion or signal a normative shift in the justification for criminal sanctions (Rengifo et al., 2017; Ulmer, 2019; Werth, 2017). We expand this literature to document how patterns of circumvention may be shaped more specifically by attributes of both actors and actions with less alignment with guidelines associated with low-effort/visibility behaviors by principals with more local authority (circumvention of open case review by judges), and more alignment in connection to high-effort/visibility behaviors by principals with less local authority (departure motions by prosecutors). This specification adds to prior studies noting that justice actors may deploy new mandates strategically by favoring components that imply less effort and scrutiny (Hannah-Moffat et al., 2009) or imply less direct challenges to the emerging status quo (Miller & Maloney, 2013). The lack of circumvention behaviors by defense attorneys noted in our study is consistent with this model as it may be indicative of their fragile position in the courtroom and the view of bail reform as advancing their institutional goals (DeMichele et al., 2018).
Our study also adds depth to patterns of reform implementation by contrasting the relative variation in actor-level strategies described above with the path dependence of more system-level attributes and exchanges (e.g., the quick pace of adjudication, the high rate of adoption of prosecutorial motions, and the lack of proactive case review by judges). The fact that these markers of old pretrial systems (Phillips, 2004; Rengifo et al., 2020; Suffet, 1966) remain a feature of new institutional arrangements sheds light to the limited ability of specific reforms to target structural features of court workgroups and their local, collective practice of justice (Ulmer, 2019). Finally, our findings provide preliminary evidence pointing to deployment of more restrictive forms of case processing for specific groups of defendants that conform to varying stereotypes of disadvantaged/“undeserving” populations (Latinx individuals without interpreters or older individuals). However, this is far from a unidimensional story (Hannah-Moffat et al., 2009) as other defendants, notably non-Latinx/Black defendants and Latinx defendants with interpreters, do not face more shallow proceedings, nor do they face more direct actions by prosecutors. As such and contrary to prior work on sentencing practices, the mobilization of more punishing discretionary power did not target all defendants in the same way, perhaps due to the availability of more restrictive options, like detention, or the influence of case-level factors that we were unable to track like prior record. In terms of the specific findings of our study, it is possible that the presence of an interpreter may have been taken by judges as an implicit signal of a subpopulation of Latinx defendants deserving standard attention during case review. In contrast, Latinx people without interpreters may have been seen as an “undeserving” group, given stereotypes related to a lack of understanding of the law or potential dangerousness.
The analyses presented in this article are bounded by a number of factors, including our reliance on a relatively small sample with limited variation in local settings and covariates, and the sole use of observations to infer characteristics of defendants and their cases. While there is substantial internal consistency in our assessments of these characteristics, in some cases, we were able to validate them externally using information provided by attorneys/judges during hearings (“age” as explicitly mentioned by courtroom actors, or “gender,” in terms of the pronouns used in open court). For other covariates including race/ethnicity, we were unable to develop a similar strategy for external validation, given that these markers were rarely signaled during formal proceedings. In addition, we were unable to match our observations with official court records, given that case-level observations were anonymized in terms of defendants’ names or docket identifiers. As a result, the categorization employed in our study may have flattened racial/ethnic differences within and across groups of individuals pending adjudication due to the absence of additional data sources to capture the complexity of ethnic/racial affiliation and identity. This is an important limitation, given the centrality of these concepts to studies of differential treatment and bias in the courts. Given that other research designs based on administrative data by criminal justice agencies confront similar limitations in data quality and confounders (Fratello et al., 2013; Goodman, 2008), future studies may consider mixed methods approaches that can document gaps and contradictions in domains of self-representation and bureaucratic management. Furthermore, while we control for the effects of judge/jurisdiction, our observations say little about the discourses and frameworks that support the actor-level strategies we observed in courtrooms. Critically, our case-level approach provides a narrow perspective on how the practice of bail reform is shaped by institutional factors such as the stability of the court workgroup or broader dynamics of stratification (Omori & Lautenschlager, 2018).
Implications for Smart Decarceration Research
Our findings on the significance of departures from process and decisions, and the relative path dependence of more structural features of case processing have implications for all three of the smart decarceration outcome goals such as reducing jail and prison populations, addressing disparities, and ensuring public safety (Pettus-Davis & Epperson, 2015). In particular, we show that the use of actuarial instruments can be effectively adopted by local court workgroups to structure community-based alternatives to detention. However, we note that, in our study, the deployment of these alternatives seldom involved the use of unconditional releases, opening the door to subsequent loops of surveillance and failure that could undermine the front-end decarceral benefits of the reform (Epperson & Pettus-Davis, 2015). Our results also lend credence to the notion that risk-informed instruments can contribute to building protocols that help to address some disparities within the criminal justice system, but not others, particularly those involving more invisible outcomes such as case review patterns for defendants with limited English proficiency. It is worth noting that these and other dimensions associated with characteristics of defendants or associated court cases are seldom available in official data repositories. Future smart decarceration efforts may consider this limitation when mapping opportunities for infrastructure development or when seeking to explore protocols and populations marginalized from standard accounts about the operation of criminal justice institutions. In our view, it is in the periphery of these conventional narratives of policy-making and case management that new programs and theories are built.
Considering the cost of bail reform implementation and the proliferation of risk assessment adoption, it may be wise for scholars to develop means for new tools to be adapted to courtroom workgroups and cultures. As emphasized in the article, these “blind spots” are integral for ensuring that bail reform measures are implemented to promote decarceration. Furthermore, as displayed in the article, forms of discretion adapt to the constraining nature of the risk assessment tool. The nature and consequence of these differences signal the need for smart decarceration initiatives to consider more explicitly how incremental change may unfold across an expanded set of indicators of process and punishment, integrating innovations in case work processing with system-level changes to better ground the theory and practice of new models of justice.
Footnotes
Authors’ Note:
We would like to thank Popy Begum, Katherine O’Toole, Luis Torres, and Dale Dan-Irabor for their research assistance in the completion of this project.
