Thomas Hoepker / Magnum Photos

A Jail Population of 3,700 or Lower is Achievable: Technical Appendix

Michael Rempel

March 12, 2024

This appendix documents the underlying assumptions and limitations behind projected jail population reductions published in Vital City.

Establishing a Baseline 

First, it is necessary to assume a jail population that would exist in the absence of policy reforms. For the current article, people were deducted from the average daily jail population of 6,200 in the second half of January 2024 — including 5,250 people held in pretrial detention. However, the proper starting point could significantly differ by the time of Rikers’ scheduled closure in August 2027. Independent of policy, jail populations continually change in response to shifting crime, arrest and demographic trends. 

The JFA Institute has developed sophisticated methods for projecting future jail populations based on past trends in crime, arrests, population demographics, jail admissions and daily population numbers. The Data Collaborative for Justice is amidst an equivalent modeling process for New York City. Notably, our own work indicates that margins of error grow to several thousand people when projections extend far into the future. In other words, it is not possible to precisely anticipate where non-policy-related trends would lead the daily population to land three years from now in 2027, when Rikers must close.

Discounting for Policy Carve-Outs and Resource Limitations

Second, final policies are unlikely to mirror strong principles articulated at the concept stage. For example, the high-level goal might be to “stop setting bail on first arrests,” but fully formed policies will inevitably establish carve-outs and afford decision-makers a degree of case-by-case discretion. This makes unavoidable the use of quasi-subjective judgments regarding the percentage of the time a desired outcome might occur (such as people meeting certain criteria no longer facing jail time). Another limiting factor is whether sufficient resources exist to meet any new demand created for services in the community. As a conservative default, for the subset of strategies reliant on new community-based diversion investments, this article largely assumes such resources could be made available and decision-makers (judges, prosecutors, etc.) would actually take advantage of them a quarter of the time.

Discounting for Imperfect Implementation

Third, even after establishing carve-outs, years of experience teaches us that everyday practice in courtrooms and other decision-making settings will likely deviate even further from policymakers’ intentions. Practitioners on the ground may miscomprehend what they are supposed to do (e.g., due to insufficient training); revert to the mean (i.e., overuse discretion to favor old routines); or willfully deviate because they disagree. That is why this article offers a range of potential jail reductions. The upper bound reflects “quality implementation,” where decision-makers are broadly faithful to intended policies. The lower bound reflects “reasonable implementation,” allowing for further implementation flaws by applying to the initial projection a standard discount of 25% (mirroring the author’s prior work). 

Finalizing Strategy-by-Strategy Projections 

Fourth, strategies have to be ordered from earlier to later decision points to avoid double-counting people already removed from jail. Strategies toward the bottom of the pretrial list show a lower projection than if policymakers had not instituted other reforms. If they solely took steps to reduce case processing delays, it would have yielded a projected jail reduction of 1,200 people in the “quality implementation” scenario, whereas a lower 700 number appears below, given that strategies applied at earlier decision points, such as people’s initial court arraignment, would have already led many people to be released.

Rounding to Avoid Misleading Claims of Precision

Finally, all numbers shown below are rounded to the nearest 50 to avoid a false aura of precision around an exercise that is impossible without blending art and science. Total jail reductions for all strategies combined are rounded to the nearest 100.

════════════════════════════════════════

A Working Set of Jail Reduction Projections

The brief descriptions provided below informed the upper bound of each jail reduction range. The lower bound is simply 25% less. (Any divergences from a 25% difference are simply a result of rounding.) Except where explicitly noted, strategies are not applied to homicides. In the interest of brevity, information provided below does not spell out literally all steps in the underlying math.

1. Adhere to the Release Assessment

Jail Reduction Range: 800-1,100 

Based on results from multiple publications, New York City’s empirically based release assessment recommends about 75% of bail-eligible violent felony (VFO) cases for release on recognizance (ROR), 17% for a middle “consider all options” category, and 8% for restrictive conditions other than ROR. In effect, the tool classifies only this final 8% with a high likelihood of missing court. Yet arraignment judges set bail or remand 55% of the time in 2022.

Before advancing a credible projection of what might change, it is worth clarifying that the purpose of algorithmic assessments is to inform decision-making, not to yield perfect adherence. As a working assumption, it was therefore projected that judges would still set bail or remand in 20% of VFOs recommended for ROR, 40% recommended for “consider all options,” and 100% not recommended for ROR. These assumptions are identical to those in an earlier 2021 report. If judges proceeded this way, they would achieve a 45% relative reduction in VFO bail-setting.

Sources cited two paragraphs above also found that judges are deviating from the release assessment with bail-eligible nonviolent felonies, but because these deviations are considerably smaller than with violent felonies (and to err conservative), no jail reduction projection is included.

  • Jail population impact: Of 5,250 people detained pretrial, 2,450 involved (non-homicide) VFOs. Reducing them by 45% translates to 1,100 fewer people detained. Discounting by 25% translates to 800 fewer detained

2. Stop Setting Bail on First Arrests 

Jail Reduction Range: 150-200

The release assessment nearly always recommends first arrests for ROR. The only exception arises when an individual is homeless. It was assumed policies directing judges to “stop” or minimize bail-setting on first arrests could, in a potential real-world scenario, perhaps prompt the release of 50% of those still detained. (The final math also draws on data showing that, in 2022, 42% of VFOs in which judges set bail or remand lacked a prior arrest in the past two years.) 

  • Jail population impact: Reducing first-arrest VFOs still in pretrial detention by 50% translates to 200 fewer people detained. Discounting by 25% translates to 150 fewer detained.

3. Use Mandatory Programming for High-Need Cases

Jail Reduction Range: 350-500

Unfortunately, no available data can credibly estimate, after applying the previous strategies, what percentage of people in pretrial detention might have “high needs” making them suitable for mandatory programming or an equivalent “intensive” supervised release track. As a working stipulation, it was assumed expansion of such programming could yield the release of 20% of people still detained pretrial (on any charge except homicide). This low 20% figure considers that mandatory programming depends on making possible the administration of an assessment capable of identifying treatable needs and, also, reflects that the city’s treatment capacity might make programming implausible in many more cases than the numbers shown below.

  • Jail population impact: Reducing by 20% the non-homicide VFOs still in pretrial detention and the 1,350 people held on nonviolent felony and misdemeanor charges translates to 500 fewer people detained. Discounting by 25% translates to 350 fewer detained.

4. Increase Targeted Housing and Mental Health Options

Jail Reduction Range: 100-150

A recent study by the New York City Criminal Justice Agency found that in a sample of people facing charges from 2019 to 2021, 6% were homeless prior to their arrest. The study did not specify whether this percentage is higher for detained people, although that is likely for reasons including that homeless people are indigent (by definition) and tend to lack strong community ties, leading them rarely to afford bail when it is set.

Data from Correctional Health Services indicates that about 21% of people in city jails currently have a serious mental illness (SMI). Absent data on the overlap between homelessness and SMI (but bearing in mind that the homeless percentage is likely an underestimate), the two percentages were added to total 27%.

Considering treatment capacity limitations, it was assumed targeted housing and mental health options at the point of arraignment could shrink people remaining in pretrial detention by 25%.

  • Jail population impact: Assuming 27% of people still in pretrial detention are homeless or suffer from a SMI, and then releasing 25% of this subgroup translates to 150 fewer people detained. Discounting by 25% translates to 100 fewer detained.

5. End Unaffordable Bail

Jail Reduction Range: 250-350

As cited in the main article, courts in a range of states have issued rulings that setting unaffordable bail is unconstitutional. Yet unaffordable bail persists in New York. Given the pervasiveness of this practice in the status quo, it was assumed that clear judicial directives or other policies intended to “end” detention on unaffordable bail might actually mean reducing it by about 25%. Specific limitations to reducing detention in more cases include: scientific challenges of accurately assessing what people can afford; logistical challenges of sharing valid ability-to-pay information with the court prior to arraignment; and the reality that some judges would either continue to assert their authority to set unaffordable bail in certain cases or instead increase their use of remand.

According to a study using 2021 data, 77% of pretrial jail admissions were held on bail, while the others were remanded and, thus, would not be affected by this strategy.

  • Jail population impact: Assuming 77% of people still in pretrial detention are held on bail (not remand) and then reducing their number by 25% translates to 350 fewer people detained. Discounting by 25% translates to 250 fewer detained.

6. Implement Automatic Early Bail Reviews

No projection: This will leverage better implementation of ending unaffordable bail.

Holding an automatic bail review within one week would ostensibly serve as a “second look,” improving implementation of prior strategies for curtailing unaffordable bail at the point of arraignment. 

7. Curb Incarceration of Women, Gender-Expansive People and People 55 and Over

Jail Reduction Range: 50-100

The current article applies projections created for an earlier 2022 report concerning the decarceration of women and gender-expansive people. (Disclosure: I served as one of this report’s coauthors.) Though offering more detail, that report recommended the same basic steps as those outlined in the current article: a universal needs assessment of women and gender-expansive people at jail intake, increased gender-responsive diversion investments and more supportive housing for these vulnerable populations. The projection was that these investments could yield the safe release of 40% of women facing non-homicide charges and 20% charged with homicide. The homicide projection reflected that the well-regarded SHERO program has a record of providing transitional housing and services to women facing extremely serious VFO charges, homicide included. For men 55 and over, the current article’s default assumption of 25% uptake for diversion programming was applied — except, unlike for women and gender-expansive people, no projection was made for people charged with homicides. Mirroring the earlier report cited just above, these projections were applied both to people held pretrial and people in jail on a pending parole violation.

  • Jail population impact: Deducting women already released via earlier projections and then applying the above assumptions translates to 60 fewer women detained. (Public jail data codes gender as binary.)
  • Proceeding analogously with men 55 and over translates to 40 fewer detained. Discounting this total of 100 by 25% translates to 50 fewer detained (after rounding downward). 

8. Flag Chronic Medical Conditions

Jail Reduction Range: 50-50

Correctional Health Services (CHS) data as of November 2023 indicates that from 1% to 29% of people held in NYC jails have one of eight chronic medical conditions. The data does not indicate the overlap among people with multiple conditions. Instead of adding up reported percentages for the eight conditions — a recipe for a gross overestimate — the upper bound of 29% for any one condition was conservatively assumed. 

Unfortunately, a credible jail reduction projection necessarily hinges on engagement and planning with experts at CHS. For purposes of the current exercise, a standard (but admittedly underinformed) working assumption in the absence of intensive CHS engagement was that CHS might present a release request in 25% of chronic medical condition cases and, in turn, the court or parole might follow such a CHS recommendation 25% of the time.

  • Jail population impact: Applying these assumptions to people still in pretrial detention or held on parole violations translates to 50 fewer people detained. Discounting by 25% also rounds to 50 fewer detained.

9. Shrink Case Processing Delays

Jail Reduction Range: 500-700

Adapting national best practices, the effect of the Brooklyn case-delay pilot was to reduce average case processing times in indicted felony cases by 28%. This 28% reduction is adopted here — but only after first incorporating data analysis conducted for an earlier 2021 jail reduction roadmap to arrive at a properly sized denominator consisting solely of indicted felonies pending a disposition. Notably, because best practices for reducing case delay apply to homicides as well as other cases, homicides are included in the projected jail reduction.

Caveat: The assumption of a 28% reduction in case processing times (and, hence, in length of stay for people in pretrial detention) presumes that the average effect seen in the Brooklyn pilot would apply to all cases. But this may not be true. The Brooklyn model may instead be disproportionately effective, or not, with people experiencing the very longest of case processing times. Any non-linearity in the effect of reform for people with varying lengths of stay in the status quo would change its net effect.

  • Jail population impact: Applying a 28% reduction to indicted felonies still in pretrial detention (including homicide charges and including people held on parole violations stemming from a pending new charge) translates to 700 fewer people detained. Discounting by 25% translates to 500 fewer detained.

10. Detention Time Limit of 180 Days

No projection: This will incentivize judges to improve their case processing.

State legislation setting pretrial detention time limits of 180 days, or equivalent judiciary-led initiatives, could help to incentivize judges to implement effective caseload management practices more consistently. This would improve implementation fidelity to case processing best practices projected just above.

11. Reboot the Early Release Program

Jail Reduction Range: 200-250

In late March 2020, the Early Release (6A) Program swiftly enrolled 54% of the 553 people then held on a city jail sentence. A logical assumption is that rebooting the program would mean returning to this enrollment rate. 

  • Jail population impact: Releasing 54% of 450 people currently held on a city jail sentence translates to 250 fewer people incarcerated. Discounting by 25% translates to 200 fewer incarcerated.

12. Invest in Critical Tools and Trainings

No Projection: This will improve implementation of other reforms.

The effect of offering judges better decision-making tools, bench cards and more regular and effective trainings would be to significantly improve the implementation of all other jail-reduction strategies.

════════════════════════════════════════

Note: To avoid underestimating what is currently a modest upward trend in the city’s jail population, the 6,200 number represents a rounded average of the daily population in only the second half of January 2024 (January 16-31). Where possible, numbers for specific components of the jail population (e.g., pretrial, serving a jail sentence, probation violations, etc.) similarly reflect a rounded average from the second half of January 2024. However, due to a lack of data for every date, certain numbers involving specific charges or demographic subgroups are based on exact figures on January 31, 2024.

Cumulative Totals

Average Jail Population in January 2024: 6,200 (Including 5,250 Held Pretrial)

Total Jail Reduction Range: 2,500-3,400

Resulting Daily Jail Population Range: 2,800-3,700 

Conclusion 

The purpose of this article’s projection exercise is not to postulate an exact “truth” of the matter. It is to indicate the rough magnitude of what decision-makers might potentially accomplish by implementing jail reduction strategies backed by evidence. The goal was for all estimates provided above to err conservative (i.e., to project a lesser, and in some places a far lesser, jail reduction than what might ideally materialize from strong leadership by policymakers followed by good implementation).

Putting aside the overall toplines, the specific projections offered above also provide a window into which specific strategies would be more and less impactful when viewed in relation to each other. Clearly, reforming judges’ overuse of bail in cases legally and empirically inappropriate for it and reducing outsized case processing delays could have an especially large effect in propelling the city toward closing Rikers Island on schedule.

As for the fine print, consistent with limitations stated in the main article, projecting the effects of policy reforms is an incurably hazardous undertaking. This reflects that each projection requires making a judgment regarding a reasonable fraction of the time a new policy might plausibly alter decision-making, alongside a stipulated implementation discount set at 25% to create a range of possibilities. By comparison, our ongoing work at the Data Collaborative for Justice suggests that pure science can take us further when it comes to projecting the effects of nonpolicy changes on the jail population, deriving, for example, from shifting arrest, arraignment and jail admission trends.

In short, it would not be shocking for actual jail reductions flowing from proposed policies to be greater or less than even those at the upper and lower bounds of each range. Readers are free to try out their own alternative assumptions. Please contact the author for further information about formulas that could be modified for readers wishing to try out the ramifications of other sets of assumptions.

A final but essential caveat is that there are countless good ideas for safely reducing the use of jail in New York City that this article neglects. The current issue of Vital City curates many such promising directions. And, yet, having collected all possible ideas, it is also unknown which of them, if any, city and state actors will take up. 

What the current projection exercise does make crystal clear is that the possibility exists for policymakers to achieve sizable jail reductions if they possess the will, knowledge and planning acumen to do so.