Introduction

Every year, more than 2.3 million felony and 10 million misdemeanor criminal cases are prosecuted by over 2,300 individual prosecutor’s offices in the United States (Natapoff, 2012). Each day, over 450,000 people await trial in jail because they cannot post bail (Cohen & Reaves, 2007). Prosecutors have broad discretion to make decisions that affect the defendant, his or her liberty, the outcome of the case, and the defendant’s future. Despite the volume of prosecutions, this work occurs largely out of public view. Thus, the mechanisms of prosecution have been termed a “black box” with some research using expressions such as “opening Pandora’s box” or “unlocking the black box of prosecution” (Kutateladze et al., 2016; Pfaff, 2011).

There are many decision points where a prosecutor can select a particular course of action, each of which can affect the final outcome of the case. However, there has been relatively little research on the process preceding guilty pleas, and “what research existed remained scattered and limited…more collaborative work was needed to unite shared ideas” (Johnson et al., 2016, p. 480). Prior research of the prosecution process has focused on pleas and sentencing, with most studies of prosecutorial outcomes examining a single decision or outcome among many that occur during the course of a prosecution. Among the peer-reviewed publications on aspects of prosecution other than plea bargains and sentencing, half focused on the initial screening decision and most examined only a single discretion point (Donnelly & MacDonald, 2018; Kutateladze & Andiloro, 2014; Leslie & Pope, 2017).

The aim of this study is to fill the gap in the literature by focusing on bail—a critical ‘front end’ component of the adjudication process. A growing body of research has examined the possible cascading effects of bail decision-making, and this research aims to contribute further by incorporating one of the first post-arrest decision points (Kurlychek & Johnson, 2019). The study also expands upon past prosecutorial discretion work by examining the bail request patterns of prosecutors in conjunction with downstream outcomes. The scope of bail is vast, with an estimated 11 million individuals detained prior to conviction annually worldwide, and almost half a million detained daily in the United States—one-fifth of all incarcerated people (Dobbie et al., 2018). Most jurisdictions in the United States participate in a monetary bail system (Pre-trial Justice Institute, 2017) and its impact is enormous. Defendants unable to post bail are more likely to lose income, housing, and employment, and are more likely to plead guilty and be imprisoned compared with those released on their own recognizance (Allan et al., 2005; Dhami, 2002; Heaton et al., 2017; Petersen, 2019a, 2019b;). In addition to many economic outcomes, significant racial and ethnic disparities are often observed in the bail outcomes and pre-trial detention rates, raising concerns about the extent of impartiality in the treatment of defendants (Arnold et al., 2018; Assefa, 2019; Demuth, 2003; Demuth & Steffensmeier, 2004; van Brunt & Bowman, 2019). To verify the validity of increasing support for bail/pre-trial reform, the current research builds on the recommendations of prior research and examines the effects of defendants’ race and ethnicity on multiple decision points, with a focus on the prosecutor’s bail request and its impact on intermediate and final case outcomes.

In particular, this study heeds the recommendations of experts in prosecution research (e.g., Donnelly & MacDonald, 2019; Johnson, 2015; Spohn, 2015; Wooldredge et al., 2015) to examine the indirect and cumulative effects of prosecutorial and judicial decision-making on the intermediate and final outcomes of criminal cases. Accordingly, the study adds to a growing body of literature on the cumulative effects of discretionary decision-making in the life-course of criminal cases by crafting improved measures of discretion within a relatively recent dataset and examining the prosecution process, beginning with decision-making regarding bail. The findings are directly applicable to the policies and practices of prosecutors’ offices nationwide.

Literature Review

Theoretical Perspectives

Uncertainty Avoidance and Focal Concerns

Some notable theoretical perspectives have been widely adopted to guide research that examines racial and ethnic disparities in various criminal court processing outcomes. Scholars associated with critical sociology and criminology traditionally proposed hypotheses regarding overt and intentional accounts of racial/ethnic disparities such as racial threat theory (Blalock, 1967; Crawford et al., 1998; Stolzenberg et al., 2004) or conflict theory (Chambliss & Seidman, 1971; Quinney, 1970). Recent perspectives are less critical, suggesting it is more complex and nuanced than previously assumed (Walker et al., 2012; Wooldredge et al., 2015). That is, as proponents of justice, prosecutors and judges try to fully assess legitimate factors such as offense type and severity, offender’s criminal history, and harm incurred by victims and the community. However, they also need to devise short-cuts and stereotypical attribution in their decision-making under practical constraints. For example, according to the perspectives of uncertainty avoidance and causal attribution (Albonetti, 1991), courtroom actors tend to develop “patterned responses” by considering not only relevant legal factors but also specific extralegal offender characteristics such as race and ethnicity when constrained by bounded rationality and insufficient time and information (March & Simon, 1958) as well as inherent uncertainty when predicting offenders’ future criminal behaviors.

Steffensmeier and colleagues (Steffensmeier & Demuth, 2000; Steffensmeier et al., 1998) present a primary theoretical framework that guides contemporary research on the racial and ethnic disparities in the criminal justice process by integrating many insights and perspectives to propose a “focal concerns” perspective. To better account for the consistent patterns of racial and ethnic disparities observed in sentencing data, focal concerns perspective posits that court actors consider three primary focal concerns when they make decisions under the constraints of time and information: the offender’s blameworthiness or culpability, dangerousness and risk of future criminal involvement, and practical constraints and consequences associated with decision-making. In addition to relevant information about offender and offense characteristics, extra-legal defendant characteristics (such as race/ethnicity, gender, and socioeconomic status) may also be used as proxies when courtroom actors assess blameworthiness, culpability, and dangerousness of defendants. For example, courtroom actors may draw on racial stereotypes or racialized/gendered perceptions to assess the harms done by the offense and threats posed by the offenders (Mann et al., 2006; Steffensmeier et al., 1998).

Cumulative Disadvantage

In existing literature on cumulative disadvantage, the importance of initial contact with the criminal justice system on subsequent criminal processing and broader life outcomes is acknowledged. The concept of cumulative disadvantage was first described by Merton (1968) with the term the “Matthew Effect,” whereby scientists might advance or decline in their field on the basis of early opportunities and work. In criminology, Sampson and Laub (1997; see also Laub & Sampson, 2003) proposed “a life-course theory of cumulative disadvantage” to emphasize a developmental model of cumulative continuity from adolescent delinquency to adult crime. However, this focused on the long-term consequences of negative life events associated with adolescent delinquency (e.g., arrest, official labeling, school failure) by highlighting how such negative consequences accumulate over the life-course of individuals, affecting their criminal behavior. Extant research more relevant to our study has underscored the need to examine cumulative disadvantages that accrue across the entire life-course of criminal cases, with emerging interest in the cumulative effects of initial decision-making on downstream outcomes within criminal justice processing (Donnelly & MacDonald, 2018; Johnson, 2015; King, 2019; Kurlychek & Johnson, 2019; Martinez et al., 2020; Omori, 2019; Spohn, 2015; Wooldredge et al., 2015). Cumulative disadvantage (or “accelerating bias”) in the context of prosecution refers to a series of events that occur during the process that make negative future events more likely. For example, defendants who cannot post bail are detained pre-trial. Those who are held in jail while awaiting trial are more likely to plead guilty to obtain credit time to a sentence and thus end the process (Heaton et al., 2017). Those who plead guilty are more likely to be convicted, making it more difficult to finding future employment (Wooldredge, 2012). Thus, even small racial and ethnic disparities identified at the initial stage of criminal case processing can accrue over the life of a prosecution.

There are competing viewpoints of the mechanisms of cumulative disadvantage. One side holds that the effects refer to “temporal growth” in existing inequalities, while the other envisions “additive and interactive” relationships (Kurlychek & Johnson, 2019, p. 293). Taken together, it is recognized that some combination of static and dynamic factors can influence future possibilities and outcomes, and that these factors change over time. Further, these disadvantages can accumulate to systematically disadvantage certain individuals or groups. The distinction between the two viewpoints is important, because each requires a different type of data and approach. For example, if a study measures growth in inequality, longitudinal data are needed. For instance-based inequality, models capable of estimating both direct and indirect effects are key (Kurlychek & Johnson, 2019; Wooldredge et al., 2015).

Integration of Diverse Theoretical Perspectives

Despite a decreasing pattern of overt racism and unwarranted racial disparities within the contemporary criminal justice system, the debate is still unsettled (Spohn, 2000, 2015). Drawing on the theoretical perspectives of uncertainty avoidance, focal concerns, and cumulative disadvantage, we posit that bail and pre-trial decisions are most prone to disparate outcomes. Compared with other criminal decision-making processes, such as indictment, plea bargaining, conviction, incarceration, and sentencing, bail request and pre-trial release tend to be cursory and patterned decisions that occur shortly after an arrest when there is usually insufficient time and little definitive information on the characteristics of the offenses and offenders (Assefa, 2019; Colbert, 2011), and sometimes even an absence of legal representation to enable a full evaluation of the defendants (Donnelly & MacDonald, 2018). Thus, ‘causal attribution’ and ‘patterned response’ based on some stereotypical characteristics of defendants may contribute to racial and ethnic disadvantages. In turn, any disparities that arise in these initial phases will affect subsequent court processing outcomes because failure to post bail or be released on their own recognizance will allude to defendant culpability and future dangerousness, which are the primary ‘focal concerns’ of those who make decisions on case processing. As well as having fewer opportunities to fight for their innocence, innocent-but-detained defendants are more likely to accept a guilty plea for favorable outcomes (Heaton et al., 2017; Reiman & Leighton, 2020). This is why evidence of racial disparity observed on single decision-making points may offer an incomplete picture of the complex, dynamic, and systematic cumulative disadvantages. The current study explicitly employs theoretical frameworks and methodological approaches recommended by the literature to account for both direct and indirect causal pathways to racial and ethnic disadvantage, and thus better understand the locus and magnitude of unwarranted disparities within the justice system.

Prior Research

Disparity or Discrimination?

Contemporary research has made significant progress in examining and understanding race-group differences for some of the most critical decision points in the criminal justice process (Spohn, 2000; Walker et al., 2012; Zatz, 2000). In most studies, Black and Latino defendants are at significant disadvantage net of legal factors, although there is still a lack of agreement as what mechanisms work to create disparate outcomes (Kutateladze et al., 2014, p. 516). Baumer argues that there are several underlying rationales that motivate almost all research on race: detecting racial discrimination, evaluating the effect of policy interventions on different racial groups, and assessing how race influences criminal justice actors (2013, p. 24). However, as Spohn (2000) states, “forty years of research [on race and sentencing] has not settled the debate” (p. 2).

While research consistently finds disparities in criminal justice outcomes between racial groups, the strength of the findings has varied somewhat (Johnson et al. 2016; Shermer & Johnson, 2010). Kutateladze et al. (2014) suggested that the inconsistency in prior findings reflects the variety of decision points examined, methods of operationalizing race and discretion, and the types of prosecutions under examination. Similarly, recent research has become even more sensitive to detecting racial disparities in certain situations among certain types of defendants (Johnson & Spohn, 2016; Spohn, 2000). Other studies contend that there is no difference in treatment between races, while others argue that any difference is indirect, conditional on other factors, and sometimes moderated through other variables (Baumer, 2013; Spohn, 2000; Steffensmeier et al., 1998; van Wingerden et al., 2016). Overall consistency in findings regarding minority groups and case outcomes has not settled the debate regarding the source and mechanisms of racial disparities.

Although research on the impact of race in criminal justice processing has identified the impact of racial stereotypes, whereby Black and Latino defendants are perceived as violent or gang members, less is known about the experience of Asian defendants in the criminal justice system. Seminal research has characterized Asian-Americans as a “model minority” based on educational and professional success and underrepresentation in serious or violent crimes (Johnson & Betsinger, 2009; Kutateladze et al., 2014; Wong et al., 1998). The positive stereotype carries different connotations compared with stereotypes of Black and Latino individuals, and there is a paucity of research on the topic of Asian people in the criminal justice system. In light of this unique status, an analysis of Asian-American defendant’s experiences with criminal justice processing is warranted and sorely needed.

Earlier Decisions Matter

Despite extensive research on racial disparities and discrimination in the criminal justice process, the focus has been mainly on plea and final sentencing decisions (Johnson, 2015; Kutateladze et al., 2014; Spohn, 2000; Steffensmeier et al., 1998). Any focus on the last stages of the process disregards the influence of prior actions by courtroom actors and the previous outcomes that resulted in a final disposition and sentence. To examine a single point in the mid or latter stages of the justice process might mask disparities that originate at other points. Scholars are increasingly taking note of the successive points of disparate treatment for minority groups during the justice process (Donnelly & MacDonald, 2018; Johnson, 2015; Kutataledze et al., 2014; Sloan, 2022; Starr & Rehavi, 2013; Tuttle, 2019; Wooldredge et al., 2015). To make progress on this issue, researchers have recommended analyzing prosecution as a dynamic set of interrelated decision-making points (Baumer, 2013). If it has been found collectively that racial disparity exists during the final stages, it is reasonable to consider that disparities exist in earlier decision-making and are worthy of further study.

Although it is beyond the scope of this study to further incorporate even earlier processes of police decision-making into our analytic model, it should be noted that the norms and practices the police rely on when they make arrest decisions can significantly affect the disproportionality in the racial/ethnic composition observed in the subsequent processes of criminal prosecution and adjudication. A voluminous body of literature has documented that young black males are significantly overrepresented in the police stops and arrests—especially in drug-related arrests (e.g., Fielding-Miller et al., 2020; Garrison, 1997; Walker, Spohn, & DeLone, 2007).

After police officers make arrests of suspects, one of the key early decisions made by a prosecutor—even before other courtroom actors (e.g., judge, public attorney) intervene is whether to request bail, and for what amount. Similarly, a judge must weigh a variety of factors to determine which type of release conditions, including detention awaiting trial, are appropriate to ensure the defendant’s return to court. The impact of these decisions is significant (Assefa, 2019; van Brunt & Bowman, 2019). The United States Supreme Court recognized the importance of freedom prior to conviction that enables an “unhampered preparation of a defense and serves to prevent the infliction of punishment prior to conviction” in Stack v. Boyle. (1951), and early criminal justice research identified the deleterious effects of pre-trial detention on the defendant’s ability to mount a successful defense (Foote, 1954).

Research into bail practices has to date been limited (especially compared with research into sentencing practices), despite being one of the earliest decisions made by a prosecutor. However, Turner and Johnson (2007) argue that the lack of focus on bail does not detract from its importance. Existing research on bail practices (distinct from pre-trial detention) has consistently found that Black and Latino defendants are subject to higher bail amounts than White defendants, even after controlling for offense severity and prior criminal history (Ayres & Waldfogel, 1994; Turner & Johnson, 2007). The full impact of bail and pre-trial detention is not just evident in their immediate effects. Both bail and pre-trial detention are influential mediators that significantly influence downstream outcomes. For example, defendants held in pre-trial detention are more likely to be convicted, incarcerated, and to receive longer sentences after conviction (Dobbie et al., 2018; Gupta et al., 2016; Leslie & Pope, 2017; Martinez et al., 2020; Oleson et al., 2016; Sacks and Akerman, 2014; Sacks et al., 2015). Understandably, there are still gaps in research on cumulative disadvantage, particularly with regard to cumulative decisions made as part of a prosecution and the techniques used to examine those decisions.

Hypotheses

The theoretical and empirical foundation of this research gives rise to three hypotheses regarding a possible pattern of disparity throughout the multiple phases of criminal case processing for minority defendants, as follows:

  1. 1.

    Black and Latino defendants would be subjected to more outcome-specific disadvantages, such as increased bail amounts, detention rates, and higher indictment rates than similarly situated White defendants.

  2. 2.

    The cumulative effects of racial disparity embedded in prior decision-making such as bail requests would be a significant predictor of downstream outcomes.

  3. 3.

    Asian defendants would be subject to less punitive outcomes and cumulative disadvantages than White defendants.

Research Context: New York County

This research is a detailed analysis of prosecutorial decision-making that took place in over 40,000 felony cases in the New York County (Manhattan) District Attorney’s Office from 2013 to 2017. New York City has received significant and continuing criminal justice reform efforts over the past few decades. The Vera Institute of Justice was formed in the 1960s to administer The Manhattan Bail Project, a study of the feasibility of personal recognizance bonds as an alternative to cash bail to ensure appearance for trial (Phillips, 2012). The landmark study demonstrated that defendants with strong community ties could be trusted to return to court. Accordingly, the defendant interview system was implemented, serving as an early model for pre-trial programming across the United States. More recently, in January 2020, New York State implemented bail reform laws that eliminated cash bail for most non-violent felony and misdemeanor offenses. The law was amended in April 2020 to include additional charges and situations for which cash bail could be required, although the spirit of the law remains intact. Judges are now required to consider the least restrictive option that will ensure defendants return to court. Prior to the COVID-19 crisis, the pre-trial detainee population at Riker’s Island (New York City’s primary jail) was estimated to have decreased by 40% as a result of bail reform. In addition to minimizing the number of individuals detained awaiting trial, bail reform could also reduce unwarranted racial disparity present in the bail decision-making process (Assefa, 2019; MacDonald & Raphael, 2020; van Brunt & Bowman, 2019).Footnote 1

Prior research performed in the same jurisdiction by Kutateladze and Andiloro (2014) indicated that, compared to similarly situated white defendants, Black and Latino defendants were more likely to be detained, to receive a custodial plea offer, and to be incarcerated. However, this study differs their study in numerous significant ways. First, data used in the current study (2013–2017) is more recent than in the Kutateladze and Andiloro’s (2014) study (2010–2011), and it is possible that recent criminal justice reform efforts might have further reduced racial disparity and be reflected in this dataset. Second, Kutateladze and Andiloro’s (2014) examined both felony and misdemeanor charges in their pre-trial detention model, whereas the current research examines only felony charges. We decided to focus exclusively on felony because misdemeanor cases are becoming less relevant in the study of bail decision-making due to the recent bail reforms within the New York County. Because racial/ethnic disparities in pre-trial detention were particularly pronounced for misdemeanor offenses, New York State have continuously restricted and eventually eliminated cash bail for most non-violent felony and misdemeanor offenses in the past decade. Third, Kutateladze and Andiloro (2014) examined only a subset of all felonies, namely “…robbery, weapons, burglary, drugs and domestic violence cases” (p. 41). These charges encompass less than half of the total felony volume of the current study’s dataset. Most of all, in our data, almost all felonies screened during the time frame had a monetary bail request of some amount, the most important variable for the purpose of this study. Fourth, the control variables in this research differ significantly from their research, most notably with regard to the scope and breadth of criminal history data used in each study. For example, Kutateladze and Andiloro (2014) included simple binary measures of prior arrests and prior incarcerations, limited to prior Manhattan arrests and incarcerations, as statewide data were not available at the time of their study. The data for the current study include criminal convictions from throughout New York State, which were available to the DANY analytics team via a data feed from the New York State Department of Criminal Justice Services, established years after the Kutateladze and Andiloro (2014)’s study. Consequently, criminal history information in the current study is much more expansive. For example, 24% of White and 47% of Black felony defendants in the current study are categorized as having one or more prior felony convictions, whereas Kutateladze and Andiloro (2014) reported that 10% of White and 24% of Black felony defendants had a prior felony conviction (Table 10, p. 74).

Data and Method

The current dataset enables a quantitative examination of the impact of legal and extralegal factors on multiple discretionary points throughout the life of a prosecution. The dataset includes almost all felony cases accepted for prosecution between 2013 and 2017 that were prosecuted by one of six “trial bureaus” at the New York County District Attorney’s Office (DANY). The trial bureaus primarily prosecute arrests by the NYPD, as opposed to the DANY Investigations Division that proactively investigates white-collar crime in Manhattan. The court data were sourced from DANY’s case management system, which was built in-house by the Information Technology department and extensively modified over the course of almost 30 years. The system contains rich information about defendants, characteristics of the prosecutor, charges, court events, and prosecution outcomes. Criminal history data were obtained from the New York State Department of Criminal Justice Services and contain all New York State convictions. These datasets were tidied and integrated into a data warehouse used by the planning department for day-to-day reporting and analysis. The data for this study were extracted from the data warehouse, which contained markers for three successive discretionary decision points over the life of a case. This dataset exceeds the threshold set by Vance and Oleson (2014) who state that the ideal dataset should include nuanced information on original charges, case processing measures, and conviction charges.

Dependent Variables

This study examines outcomes for racial and ethnic groups across three dependent variables: bail request, pre-trail detention, and indictment. The first dependent variable—bail request—refers to the bail amount requested by the Assistant District Attorney (ADA) during initial case screening and prior to criminal court arraignment. In our data, almost all felonies screened during the time frame had a monetary bail request of some amount. In practice, an ADA drafts a complaint based on the information delivered by the arresting officer. The ADA examines the facts and circumstances of the instant offense, and then goes on to draft a complaint with the charges and alleged facts of the case when the defendant’s criminal history is available. Then, they make a bail recommendation that will be delivered to the arraignment judge a few hours later. In most instances in the study jurisdiction, the ADA drafting the complaint and the ADA presiding over arraignments are not the same, and the arraignment ADA has little familiarity or time to review the case. The screening ADA reviews the available facts and circumstances of the case and enters a bail request into the case management system, which is used by the arraigning ADA during criminal court arraignments. The distribution of bail amount is right-skewed, and a log transformation is applied prior to modeling.

At criminal court arraignment, the defendant is informed of the charges against him or her, the prosecutor argues whether the defendant should be subject to bail or released on his or her own recognizance. The defendant’s lawyer presents their arguments for either lower or no bail. Judges who may have access to pre-arraignment screening reports and other documentation then set the conditions of release pending further proceedings. On average, the entire process between arrest and criminal court arraignment takes place in less than 24 h (McKinley, 2014).

The second dependent variable Is pre-trial detention, which is coded 1 for defendants who had some amount of monetary bail set at criminal court arraignment and coded as 0 for defendants who were released on their own recognizance. After being charged with a felony and criminal court arraignment, an ADA can choose whether to seek a felony indictment by presenting the case to a grand jury or to reduce the charges to a misdemeanor and prosecute the matter in criminal court. The third dependent variable is indictment, coded as 1 when a felony complaint was elevated to supreme court, and 0 when it was not.

Independent Variables

The primary independent variables are the race or ethnicity of the defendant, measured using dummy variables for White, Black, Latino, and Asian defendants, with White as the reference category. We control for gender and age. Gender is a dichotomous variable coded as 1 for male defendants. Age is an ordinal measure with categories under 18, 18–24, 25–35, and over 35, with 25–35 the omitted reference category.

Additional independent variables were included to account for the severity of the offense, case characteristics, and the defendant’s criminal history. We constructed dummy variables measuring the statutory severity, presence of violence, and offense type for the top 13 most common charges, yielding categories such as ‘Class B violent robbery’. All other charges were grouped into an ‘Other’ category, and a Class D grand larceny offense was used as the omitted reference category. We also captured whether the case was flagged as domestic violence, the defendant’s defense attorney type, and whether the defendant had another pending case in Manhattan at the time of the instant offense.

The criminal history of the defendant was controlled for using four distinct measures: prior misdemeanor, felony, violent convictions, and the number of bench warrants issued for the defendant. These variables differed significantly from those in the prior research performed in the same jurisdiction (Kutateladze & Andiloro, 2014; Kutateladze et al., 2014). Data for the prior research included measures of prior arrests and prior incarcerations, although these were only prior Manhattan arrests and incarcerations, as statewide data were not available at the time. The data for the current study comprise criminal convictions throughout New York State, which was available to the DANY analytics team via a data feed from the New York State Department of Criminal Justice Services established after the Kutateladze and his colleagues’ research (Kutateladze & Andiloro, 2014; Kutateladze et al., 2014).

Another set of defendant-related variables concerned whether the defendant, at the time of the offense, was tracked internally by the Crime Strategies Unit as a ‘crime driver’ in Manhattan. Specialized software known as the Arrest Alert System helped prosecutors and analysts organize persons of interest and issue notifications of any arrests pertaining to those individuals in New York City (Tallon et al., 2016). The Arrest Alert System also contained information about whether the defendant was under New York State parole supervision at the time of the offense.

Analytic Strategy

We created a series of multivariate regression models to investigate racial and ethnic differences at each of the three separate decision points. The first model comprised only race and ethnicity as independent variables, the second comprised race/ethnicity and all legally relevant and other control variables, and the third comprised race/ethnicity, the control variables, and measures of prior decision-making. Each model was built upon the previous one(s) and included measures of the preceding decision points to model the cumulative effects of discretionary decision-making. The bail request model was an exception because the prosecutor’s bail request was the first decision point. Two of the three regression models used logistic regression because the outcome was binary. The prosecutor’s bail request expressed in dollars and linear regression was used for the bail request model. In contrast to prior research in the same jurisdiction (e.g., Kutateladze et al., 2014), the decision points for the current study had a clear temporal ordering.

To investigate possible cumulative disadvantage that might occur across multiple stages of prosecution, path analyses were conducted using Mplus version 7.11 (Muthén & Muthén, 2013). In addition to the sequence of the causal link among multiple outcome variables, path analysis enables us to understand the effects of defendants’ race and ethnicity on both pre-trial detention and indictment outcomes by decomposing the total effect into direct and indirect effects via prior decision points. Because the sampling distribution of an indirect effect estimator is not necessarily normal or even symmetrical but usually positively skewed (Preacher & Hayes, 2004, 2008), we used the confidence intervals of indirect effect estimators created empirically by bootstrapping procedures, a nonparametric resampling approach that makes no assumption about normality. Conventional statistical tests of mediation make a parametric assumption of the standard normal distribution of the standard error and are often underpowered when assessing the statistical significance of the estimated indirect effects (Preacher & Hayes, 2004, pp. 720–722). Mplus is widely applied to test total and specific indirect effects because of its flexibility in estimating various types of bootstrap confidence intervals (CIs), such as percentile, bias-corrected, or bias-corrected and accelerated CIs (Hayes, 2013; Preacher & Hayes, 2008, p. 885).

Results

Approximately 85% of defendants in the sample were male, and the average age was 33 years old. In terms of criminal history, there were just under one prior felony conviction and five prior misdemeanor convictions on average per defendant. In addition, the average defendant failed to appear 1.8 times in a prior New York County prosecution. According to available criminal history data, 83% of defendants had no prior violent convictions, approximately 7% were under New York State Parole supervision, and 12% were flagged in internal intelligence databases as gang members or suspected crime drivers.

Although the descriptive statistics suggest significant variation in the dependent variables across race and ethnicity categories, these could be the result of differences in legally relevant indicators of criminal history or crime severity, or extra-legal indicators such as defense counsel type. To account for these pre-existing conditions, we next estimated a series of multivariate models to analyze outcome-specific racial disparities. Throughout our analysis, Model 1 includes only race and ethnicity, while Model 2 has legally relevant and extra-legal characteristics added, and Model 3 includes indicators of upstream decision-making outcomes where applicable (Table 1).

Outcome-Specific Disadvantage

Table 2 shows that Black defendants were subject to significantly higher bail requests compared with White defendants. The inclusion of legally relevant and extra-legal characteristics reduced—but did not eliminate—this disparity. Compared with Model 1, the inclusion of legal and extra-legal controls significantly improved model fit, with Model 2 explaining approximately 26% of the variation in bail requests. The final coefficients indicate that compared with Whites, the bail requests for Black defendants were approximately 3% higher (p < 0.05) with no statistically significant difference between White and Latino dependents or between White and Asian defendantsFootnote 2 at the conventional significance level. In addition, defendants charged with more serious offenses or with more significant criminal histories were subject to more bail requests. Defendants in domestic violence cases or those that were charged in ECAB were associated with decreased bail requests.

Table 1 Descriptive statistics by race and ethnicity
Full size table
Table 2 Regression coefficients and odds ratios predicting bail request, pre-trial detention, and indictment outcomes
Full size table

When relevant controls were applied, converse pattern of findings emerged for racial and ethnic differences in the pre-trial detention outcome. Somewhat surprisingly, Black and Latino defendants were significantly less likely to be detained following criminal court arraignment after the inclusion of legal and extra-legal controls (Model 2) and the prosecutor’s bail request (Model 3). Table 2 shows these counterintuitive results, which begin witih the race/ethnicity-only pre-trial detention model (Model 1) indicating Black and Latino defendants at increased risk of pre-trial detention (p < 0.001). The addition of legal and extra-legal controls changes the direction of the relationship between race and pre-trial detention: Model 2 indicates that the odds of Black and Latino defendants being detained following criminal court arraignment are expected to decrease by 11% (p < 0.01) and 14% (p < 0.001), respectively. The addition of the prosecutor’s bail request further reduces the likelihood of pre-trial detention for Black and Latino defendants. Model 3 suggests the odds of pre-trial detention for Black and Latino defendants are significantly lower than similarly situated White defendants, by 17% (p < 0.001). Once again, no statistically significant difference was observed between White and Asian defendants after controls (Models 2 and 3). Consistent with our prediction, defendants with increased bail requests were much more likely to be detained following criminal court arraignment (2.89, p < 0.001). Similarly, charge severity and criminal history tend to increase the likelihood of pre-trial detention. Characteristics such as domestic violence and very young or very old defendants reduced the likelihood of pre-trial detention.

That Black and Latino defendants were less likely to be detained than White defendants is noteworthy for a number of reasons. Although prior research in the same jurisdiction (Kutateladze & Andiloro, 2014) indicated that Black and Latino defendants were more likely to be detained than White defendants, we believe our findings are more valid and informative in some significant ways. The current research examined only felony charges because misdemeanor charges are no longer eligible for bail decision in New York County. The types of felony offenses are also more inclusive than prior research in which only a subset of all felonies were analyzed. Most of all, the scope and breath of criminal history, as one of the most important control variables in this research, are much more expansive than those of prior research.Footnote 3 Some recent efforts made for bail reforms could have substantially reduced unwarranted racial disparity in New York County, especially among serious felony cases that are constantly being monitored.

Following pre-trial detention, we next predicted the likelihood that a felony complaint would be indicted following presentation to a grand jury. The factors associated with an increased likelihood of indictment were similar to those that increased the likelihood of the dependent variable in prior models: violent charges, prior criminal history, and whether the defendant had another pending New York County case. In contrast to the pre-trial detention models (but consistent with our a priori prediction), each of the indictment models indicates that Black defendants are associated with an increased risk of indictment compared with White defendants (1.12 and 1.11 at p < 0.01 in Models 2 and 3, respectively). There was no statistically significant difference between White and Latino defendants or White and Asian defendants at the conventional significance level when all the control variables were included. The odds ratios indicate that defendants who are held in pre-trial detention are more than two and a half times as likely to be indicted compared to defendants who are not held in pre-trial detention (2.51, p < 0.001). Similarly, an increase in the requested (logged) bail amount is associated with an increased likelihood of indictment (2.13, p < 0.001). It is notable that the race estimates are fairly stable even after accounting for prior decision-making.

These results, in conjunction with the bail request and pre-trial detention models, suggest that although Black defendants are less likely than Whites to be detained awaiting trial, they are subject to increased bail requests and more likely to be indicted, net of other legally relevant factors such as criminal history and charge. Albontetti’s (1987) seminal research on the charging process concluded that prosecutors reject a significant number of cases at screening and attempt to avoid uncertainty by filing charges on cases with favorable odds of conviction. However, rejection of a significant percentage of cases at screening has historically not been the case in Manhattan (Kutateladze & Andiloro, 2014). Prosecutors may use indictment as a case screening tool by choosing not to indict certain cases that were considered less likely to be convicted in supreme court. If nearly all cases are accepted and Black defendants are more likely to be indicted, these results suggest an additional layer of cumulative disadvantage.

Cumulative Disadvantage

The first part of our analysis (Table 2) illuminated a number of circumstances that resulted in racial and ethnic disparities in criminal prosecutions, although it did not fully examine how punitive outcomes such as higher bail requests and pre-trial detention could accumulate and exacerbate disparate outcomes. To provide additional insight into the mechanics of prosecutorial and judicial decision-making, the second part of our analysis examined the indirect effects of defendants’ race on indictment via bail requests and pre-trial detention.

We first decomposed the effect of Black defendant on pre-trial detention outcome into direct effect and indirect effect via bail request. Although Black defendants are less likely than White defendants to be detained awaiting trial, they are still more likely to be subject to pre-trial detention indirectly via increased bail requests of prosecutors (i.e., Black defendant → bail request → pre-trial detention) as shown in Fig. 1 and Table 3. We then decomposed the effect of Black defendant on indictment outcome into direct effect and indirect effects via all the possible combinations of bail request and pre-trial detention (Black defendant → bail request → indictment; Black defendant → bail request → pre-trial detention → indictment; Black defendant → pre-trial detention → indictment). Figure 1 depicts how indirect effects via disparate bail requests extend to the indictment outcome. Table 3 shows that all the indirect effects via bail request (i.e., Black defendant → bail request → indictment; Black defendant → bail request → pre-trial detention → indictment) exacerbate the cumulative disadvantage against Black defendants. However, because the direct effect of Black defendant on pre-trial detention is in the negative direction, the specific indirect effect only via pre-trial detention (not via bail request) ameliorates the cumulative disadvantage against Black defendants. Collectively, these results suggest that the patterns of racial disparity are more nuanced and complex than those reported in prior research. Despite there being a combination of positive and negative cumulative effects, we found significant indirect effects of Black defendants via bail request that contribute to the unwarranted racial disparities in both pre-trial detention and indictment outcomes. However, racial disparity is less obvious and serious than expected due to the process of cumulative advantage resulting from decreased odds of pre-trial detention among Black defendants.

Fig. 1
figure 1

Path Diagram of the Causal Link among Black Defendant, Bail Request, Pre-trial Detention, and Indictment

Full size image
Table 3 Direct and indirect effects of black defendant on pre-trial detention and indictment
Full size table

Discussion

The current study investigated racial and ethnic disparity across multiple prosecutorial and judicial decision points using data on almost all felony prosecutions that commenced between 2013 and 2017 in New York County (Manhattan), New York. Our findings lend credence to the continued focus on the effects of race and ethnicity on prosecutorial and judicial outcomes, especially within the early phases of case processing. First, the three sets of models applied in the analysis revealed mixed evidence of racial disparity in bail requests, pre-trial detention, and indictment outcomes net of other relevant factors. Consistent with our prediction, Black defendants were subject to higher bail requests and more likely to be indicted in supreme court compared with White defendants. However, contrary to our prediction, Black and Latino defendants were less likely to be detained than White defendants. While we hypothesized that Asian defendants would be subject to less punitive outcomes compared to similarly situated White defendants, there was no statistically significant difference between Asian and White defendants with regard to bail requests, pre-trial detention, and indictment.

Legal indicators such as crime severity and prior criminal history were strongly associated with more punitive outcomes together with race and ethnicity, which is consistent with most prior research (e.g., Stolzenberg et al., 2013; Sutton, 2013). Specifically, defendants with more serious criminal histories, those flagged by DANY intelligence systems as ‘crime drivers,’ or those charged with more serious offenses were more likely to experience punitive outcomes. Consistent with the theoretical perspectives of uncertainty avoidance (Albonetti, 1991) and focal concerns (Steffensmeier & Demuth, 2000; Steffensmeier et al., 1998), these findings suggest that prosecutors and judges use a combination of legally relevant factors and other extra-legal defendant characteristics in their decision-making. Especially, dangerousness should not be a factor prosecutors and judges consider within New York County, but the results shows that they might associate specific racial/ethnic groups with more threats to the community. It should also be noted that while the focal concerns perspective allows for the influence of defendant’s race and ethnicity, it posits that legal factors are the primary focal concerns of courtroom decision-makers and these should therefore dominate sentencing outcomes (Steffensmeier et al., 1998).

Black and Latino defendants exhibited varying and inconsistent levels of disparate outcomes compared to similarly situated White defendants, which was counter to our first hypothesis. This highlights the importance of examining multiple decision points when studying prosecutorial and judicial decision-making. If we had only examined a single decision point for this sample, we would have simply concluded either that Black defendants were found to be subject to more punitive bail requests or indictment rates, or that Black and Latino defendants were found to be at reduced risk of pre-trial detention compared to White defendants. Further, if we had used pre-trial detention as the starting point in the sequence of decision-making, we would also have erroneously concluded there was no evidence of racial disparity against Black defendants at both pre-trial detention and subsequent decision points. Previous research has identified early decision-making in the criminal justice process as crucial and often made by actors with the least authority, rendering it ripe for disparities (Sutton, 2013). By examining and incorporating bail request as one of the earliest decisions made by the prosecution, we identified nuanced racial disparity whereby Black defendants were subject to more punitive bail requests in comparison to similarly situated White defendants. The disparity in bail requests between defendants of different racial and ethnic groups has both direct and indirect effects on subsequent decision-making, which makes it an important decision point to study.

The importance of analyzing multiple and successive decision points is further verified by our finding that the prosecutor’s bail request had a positive and significant impact on the likelihood of pre-trial detention, but that Black and Latino defendants were less likely to be detained pre-trial when holding all other factors constant. Path analysis results suggest that race and ethnicity have significant direct effects on the outcomes we examined but may also have indirect effects mediated through prior decision-making such as bail requests. In particular, Black defendants were more likely to be subject to pre-trial detention and indictment indirectly via higher bail requests. However, our findings suggest not merely cumulative but also offsetting racial impacts due to inconsistent patterns of racial disparities at multiple stages of case processing.

Overall, our findings offer qualified support for our theoretical predictions. The criminal court findings are consistent with a focal concerns perspective, which posits prosecutors and judges make decisions with “…incomplete information, [and] race may serve as a key decision-making proxy for offender dangerousness, threat, and culpability” (Kutateladze et al., 2014, p. 540). Prior qualitative research in Pennsylvania also identified the use of racial stereotypes by court actors, who described the criminal records of young black males as “more serious and indicative of future crime risk” than other defendants (Steffensmeier et al. 1998, p. 788). These statements echo our findings, especially at the earliest point of making bail requests when prosecutors experience limited time and incomplete information available to them.

Despite its significant contributions, this study also has some important limitations. First, even after controlling for the effects of all the legally relevant and other factors currently available in the dataset, the omitted variable bias is still likely resulting from some unknown/unaddressed sources of racial/ethnic disparities. For example, selection bias might still exist in our estimation of the effects of race/ethnicity on the prosecutor’s bail request decision because the current data were restricted to the ‘arrested’ felony cases. These cases might be systematically different from the not-arrested felony cases if police have a difference evidence threshold or other norms and practices with black defendants when bringing in cases to the DAs office. In the absence of such information, we could not address this issue explicitly in our analyses (e.g., by using Heckman’s correction for selection bias, which accounts for potential biases introduced when examining only arrested felony offenders out of the entire felony offender population). The performance of our analytic models would have been substantially improved if police arrest decision-making had been considered in conjunction with subsequent decision-making in the criminal prosecution. Future research should address this issue to further investigate how the discretion exercised within each domain of police, prosecutor, and judge is closely interrelated, producing unwarranted disparities in the formal criminal justice process.

nexus among different domains of the system—for the pursuit of racial justice ultimately will require thoughtful examination of the many diverse and interrelated discretionary components of the formal criminal punishment process. Second, it is a single jurisdiction study and various rules, guidelines, norms, and practices are implemented and exercised across different jurisdictions, which limits the generalizability of our findings. Third, despite our emphasis on the cumulative nature of the disparities observed across multiple decision points, this study did not investigate the downstream consequences of bail requests past the indictment stage including guilty plea, conviction, and sentencing. Lastly, as previously discussed, the current study did not incorporate the initial arrest patterns of police officers. Both researchers and practitioners agree that the police have substantial discretion in their arrest decision, which can have important and lasting influences on subsequent prosecutorial decision-making. Future research should examine diverse and interrelated discretionary points simultaneously.

Conclusion

Almost 30 years ago, a panel on sentencing assembled by the National Research Council concluded that future study of the punishment process required the collection of longitudinal data to understand the scope and specific impact of disparate outcomes at different points throughout the process (Kurlychek & Johnson, 2019). Today, the call for further research still rings true. The overwhelming majority of empirical research is limited to snapshots of individual decisions, often ignoring upstream outcomes that may have limited downstream options. Comparatively little research has been conducted on the earliest decisions made by prosecutors, and how those choices impact the continuum of punishment.

Building on the recommendations of past research, the current study contributes to existing research on race and punishment in a number of important ways. Informed by a cumulative disadvantage perspective, this study illuminates previously unknown mechanisms at work in prosecution by investigating prosecutorial discretion and racial disparity across multiple decision points using a robust data set from a large urban jurisdiction. Rather than focusing on a single outcome, this research followed a cohort of over 40,000 felony defendants through three stages of prosecution. Further, the study built upon the extant research and analyzed seldom-investigated prosecutorial outcomes, such as requested bail amounts and indictment. Further, analyses of the indirect effects suggest the presence of accumulating disadvantage for Black defendants. Black defendants were more likely to be detained and indicted indirectly via higher bail requests than comparable White defendants. We call for this line of research on some other extra-legal factors that might affect prosecutorial decision-making such as gender (Didwania, 2021),

Perhaps most importantly, practitioner feedback received throughout the course of the study shed light on the typical foci of frontline prosecutors, case processing nuances, and data limitations. The findings reiterate the importance of a continued societal focus on bail, because of the immediate effects and the impact of bail on downstream consequences. Our findings could also inform the wider research community, practitioners, and the public about the importance of fair and just decision-making throughout the prosecution process. We support the abolition of cash bail system especially for misdemeanor and non-violent felony offenses because unwarranted disparities might persist even after adopting clearly defined bail schedules and rigorous risk assessment tools. Patterned responses such as stereotyping based on some extralegal offender characteristics (e.g., race/ethnicity)—not just legal factors—are unavoidable under the constraints of insufficient time and information as well as the inherent uncertainty when predicting offenders’ risk of flight and pre-trial offending.

It is the authors’ hope that this study makes a contribution to the call for quality research into cumulative disadvantage and the mechanisms of prosecution, and that future studies will build upon its findings.