Hochul’s Regressive Bail Rollbacks Untethered From Evidence
The bail reform debate turns on data from data from New York State’s Office of Court Administration. But that data cannot answer the question of whether reform works.
From the moment it was adopted in January 2020, New York’s bail reform law has faced intense opposition from law enforcement and other defenders of the criminal legal status quo. Those forces succeeded in pressuring then-Governor Andrew Cuomo to roll back part of the law just months after it was implemented. (He did so by threatening to shut down the entire state government during the peak of the COVID-19 outbreak in New York City.) Governor Kathy Hochul’s 2022 budget—passed over the weekend—will roll the initial reforms back even more, by giving judges even more freedom to confine people pre-trial.
The goal of New York’s bail reform law was simple: to make it more likely that people awaiting trial would not be locked up in a county jail due to their inability to afford bail. The reform established a large number of offenses—subject to a host of exceptions—where judges were not allowed to use monetary bail, but had to either release the defendant subject to certain conditions (curfew, electronic monitoring, etc.) or “on recognizance” (i.e., no conditions). Pre-reform, judges had been free to use these options; now they were required to do so. The July 2020 rollbacks, imposed just months after the original reforms took effect, made offenses eligible for bail that the original bail reform law said could face only non-monetary options. The 2022 rollbacks do the same.
Perhaps the most frustrating aspect of the fight over bail reform is that it is completely untethered from any empirical evidence. The debate turns primarily on the pretrial release data provided by the New York State’s Office of Court Administration (OCA). This rare, commendable example of transparency details state-wide case-level data on more than 200,000 arrests and arraignments in New York courts from January 2020 to June 2021, with updates planned for every six months.
But the OCA data cannot answer the question of whether bail reform “works” for three reasons. One can be easily fixed. The second may pose a more serious challenge. And the third likely defies any sort of correction, but it is nonetheless critically important and points to more significant flaws in how we handle all assessments of “success” in the criminal legal system.
This may seem like a wonky argument about New York’s pretrial release data. On one level, it is. But there is a deeper, far more important lesson here for our increasingly “data-driven” age. The types of data stories we tell are inescapably shaped by the data that we have. When the data cannot tell the story that matters—as is the case here—what we get instead are misleading accounts. And these accounts appear scientific because they purport to be based on “the data.”
By digging into the OCA data, I want to explain why the debate over New York’s bail reform law is fundamentally misguided—and, despite all the numbers being thrown around, fundamentally data-free. In the process, I also want to show how important it is to think critically about the data we use in public debates over major policy issues.
The first, and easiest, issue to address is that the OCA data starts on Jan. 1, 2020. This is the same day bail reform went into effect and thus provides no before-and-after comparison. Unfortunately, efforts by journalists to get the OCA to release data from 2019 and earlier—data the agency has and could easily release—have not been successful, and Hochul’s rollbacks failed to amend the statute creating the public data to compel the agency to do so. This means that when people criticize bail reform by finding allegedly high levels of pretrial-rearrests among certain categories of defendants who have not been locked up while their cases advance, they cannot credibly argue that bail reform made things worse. There simply is no systemic, state-wide “before” data to compare to the present data.
Another related limitation is that it is nearly impossible to identify which releases are due to reform. The major shift in the initial bail reform law was to hold that judges had to use nonmonetary approaches in certain cases where before they had the discretion to either use nonmonetary options or impose bail. But discussions fail to separate the cases whose rules were changed under the reform law from those whose were not. Moreover, even if we focus on offenses that had their rules changed by bail reform, it is impossible to say for a given case if the judge would have imposed some sort of restrictive bail amount on that defendant absent the reform law. Having some pre-2020 data could help develop a rough sense of the extent of changes in bail-setting practices, but it would still be impossible to say anything certain about any specific case. It is important to note, then, that any claim that bail reform “caused” a particular bad outcome is based on untestable assumptions about what would have happened without the reforms.
The second major problem is more serious and gets to the heart of what it means to say that bail reform has “worked.” The animating logic of bail reform is the belief—well substantiated by data—that locking someone up generally increases the risk of later reoffending. Jails, especially violent and mismanaged facilities like Rikers Island, traumatize the people incarcerated there, and may increase the risk of arrests and conviction. Confinement can also lead to job loss or the loss of housing, both of which are risk factors for later offending (and, in the case of homelessness, later victimization as well).
We are thus rolling back more and more of the new bail law based on zero evidence of its long-term impact–impact that we don’t really measure, and haven’t had time to if we did. Even though there are likely major public safety benefits of reform.
The problem with data about pretrial outcomes, however, is that it only follows a person’s case until either the case is disposed (for misdemeanors) or transferred to the courts that handle felony cases. (In New York’s complicated judicial bureaucracy, all cases start in one of several lower courts, but felonies are resolved in higher-level ones.) So this data doesn’t provide a complete picture of a person’s criminal trajectory. If greater pretrial releases lead to a small increase in while-on-release rearrests in exchange for bigger, longer-run declines in total criminal history, the OCA data will suggest that bail reform is “failing” because it can only show us the costs.
Using the OCA data to evaluate bail reform runs the risk of trying to do cost-benefit analysis with a dataset that only tells you the costs. That never ends up well–or accurate.
Fixing this is possible, but it’s more challenging. Understandably, the OCA data is anonymized: each case is assigned a random ID number, and it’s not even possible to link cases within the dataset. In theory, the OCA could provide deanonymized data to researchers, who then would be able to observe the number of future criminal legal system contacts specific defendants have, even if they take place outside the time of pretrial release. The OCA has this data, but it obviously raises concerns about privacy—concerns that are particularly valid given the growing push for “second chance” laws that spare people from being permanently shadowed by old criminal convictions.
But even if the OCA was willing to provide this sort of data, it still wouldn’t be able to provide any answers for current policy debates. Bail reform went into effect on the eve of the ongoing disruptions from the COVID-19 pandemic. It would take several years of post-reform data to be able to say anything about the longer-run impact of bail reform on overall rates of violent and antisocial behavior. This is all the more so because making any firm claims from data on the chaotic early years of the pandemic will be particularly fraught at best and impossible at worst.
Moreover, that sort of data would never be able to shed light on potentially significant knock-on benefits from sparing people pretrial detention. It could not identify, for example, the person who ends up never having a criminal record because the pretrial release of a family member allowed for critical support that steered their life in another direction.
We are thus rolling back more and more of the new bail law based on zero evidence of its long-term impact–impact that we don’t really measure, and haven’t had time to if we did. Even though there are likely major public safety benefits of reform.
Which brings us to the third major problem. Not only does our pretrial data capture only the costs of more releases and few to none of the benefits, it also captures just the costs of early releases, not the costs of additional detentions. There are two huge failures here. These failures are not unique to bail reform data, but they plague almost all cost-benefit analyses of the criminal legal system.
First, our pretrial data points out each person who gets rearrested while on release, a rearrest that would have been avoided had that person been detained. (But we need to be careful even here: About 1,000 people who have rearrest records in the OCA data appear to have been detained in NYS jails at the time of that arrest. This reflects how those institutions struggle to maintain safety and order.)
But pretrial data cannot identify the people who are detained but would not have been rearrested for anything had they been released. This is a notorious problem for the politics of criminal legal reform, that only one type of error—the easily politicized, emotionally salient preventable offense, not the needless confinement—is visible to the public. Opponents of the original bail reform law were quick to exploit such cases in their successful push to get Governor Cuomo to roll back part of the law in 2020.
The second missing cost is that we never estimate the social and human costs of confinement. Confinement affects those locked up in jails as well as their families and communities. Administrative data gathered by the courts can never shed light on this. For example, the risk of death by suicide in jails (50 per 100,000) is nearly twice the overall risk for men aged 25-44 (28 per 100,000) and almost five times greater than the overall risk for Black men (12 per 100,000). A 2018 study showed that pretrial release led to significant increases in wages and employment that lasted for years. People are regularly physically and sexually assaulted in jails. And as we saw during the first wave of COVID-19, jails are vectors for spreading contagious diseases among those confined or working there and to their communities. Families incur significant expenses to try to raise bail; people worry about loved ones being harmed at Rikers; and so on. Even if fewer confinements causally result in more offending—offending that also harms and destabilizes marginalized communities—those extra costs must be balanced against these human costs of confinement as well.
It is impossible to say this early on whether bail reform has made us safer or less safe. But there is extensive evidence of the harmful, crime-causing effects of incarceration, which strongly suggests that we should give the reforms more time to operate to be able to assess if they are working. And the data on pretrial releases cannot support the claims of failure that opponents of the reform try to pin on it.