This article was originally published in Vol. 41, Issue 1 of the Cardozo Law Review. 

Risk assessments are a common feature of federal decisionmaking across a wide variety of areas of regulation. Perhaps the most widely used definition describes risk assessment as “the process of using risk factors to estimate the likelihood (i.e., probability) of an outcome occurring in a population.”<fn>John Monahan & Jennifer L. Skeem, Violence Risk Assessment: The State of the Science, in Principles and Practice of Forensic Psychiatry 713 (Richard Rosner and Charles Scott eds., 3d ed. 2017) (quoting Helena Chmura Kraemer et al., Coming to Terms with the Terms of Risk, 54 Archives Gen. Psychiatry 337, 340 (1997)). Thus, “[r]isk assessment can be defined as the process through which risk factors and protective factors are used to estimate the likelihood that an outcome will occur.” Sarah L. Desmarais & Evan M. Lowder, Safety and Justice Challenge, Pretrial Risk Assessment Tools: A Primer for Judges, Prosecutors, and Defense Attorneys 3 (2019), http://‌www.safetyandjusticechallenge.org/‌‌wp-content/‌‌uploads/‌‌2019/‌‌02/‌‌Pretrial-‌Risk-‌Assessment-‌Primer-‌February-‌2019.pdf [https://‌perma.cc/‌X9VP-‌4Q8V]. The First Step Act, the focus of this Essay, defines the term “risk and needs assessment tool” as: an objective and statistically validated method through which information is collected and evaluated to determine— (A) as part of the intake process, the risk that a prisoner will recidivate upon release from prison; (B) the recidivism reduction programs that will best minimize the risk that the prisoner will recidivate upon release from prison; and (C) the periodic reassessment of risk that a prisoner will recidivate upon release from prison, based on factors including indicators of progress and of regression, that are dynamic and that can reasonably be expected to change while in prison. First Step Act of 2018, Pub. L. No. 115-391, sec 101, § 3635(6), 132 Stat. 5194, 5208–10.</fn> Measures for such outcomes are common throughout federal administrative law, particularly where regulations have impacts on large populations. For example, the federal Regulatory Improvement Act specified the use of risk assessment for any “major rule” by a federal agency.<fn>Regulatory Improvement Act of 1998, S. 981, 105th Cong. § 623(a)(1).</fn> The Environmental Protection Agency (EPA) uses risk assessments regarding human health impacts and ecological risks with the assistance of standing advisory groups of experts, as well as outside peer review from experts, including the National Academy of Sciences, and with detailed guidelines for conducting these assessments.<fn>About Risk Assessment, U.S. Envtl. Protection Agency, https://www.epa.gov/‌risk/‌about-‌risk-‌assessment [https://‌perma.cc/‌PHY6-‌XDV6] (last updated June 05, 2019); Risk Assessment Guidelines, U.S. Envtl. Protection Agency, https://www.epa.gov/‌risk/‌risk-‌assessment-‌guidelines [https://‌perma.cc/‌KC8M-‌V2FG] (last updated June 05, 2019).</fn> In a 1984 report, the National Academy of Sciences provided detailed guidance on how federal agencies should incorporate risk assessment, and distinguish it from risk management.<fn>Nat’l Research Council, Risk Assessment in the Federal Government: Managing the Process (National Academies Press 1984), https://www.ncbi.nlm.nih.gov/‌books/‌NBK216615 [https://‌perma.cc/‌J6UZ-‌RXTU].</fn> A series of more recent National Institute of Standards and Technology (NIST) reports have provided updated guidance on risk assessment, including for all federal information systems.<fn>E.g., Special Publ’n, Guide for Conducting Risk Assessments (2012), Nat’l Inst. of Standards and Tech., https://nvlpubs.nist.gov/‌nistpubs/‌Legacy/‌SP/‌nistspecialpublication800-‌30r1.pdf [https://‌perma.cc/‌4VFD-‌EDHZ]; Special Publication, Managing Information Security Risk (2011), Nat’l Inst. of Standards and Tech., https://‌nvlpubs.nist.gov/‌nistpubs/‌Legacy/‌SP/‌nistspecialpublication800-‌39.pdf [https://‌perma.cc/‌E4WM-‌HMQF].</fn>

In contrast, federal criminal actors traditionally did not use risk assessments, despite the impact that decisions regarding criminal detention, sentencing, incarceration, and release all have on large populations. For example, the 1984 Federal Bail Reform Act asked judges to predict the potential failure of a defendant to appear or the danger of the commission of a new crime, but without the benefit of empirical evidence.<fn>Shima Baradaran & Frank L. McIntyre, Predicting Violence, 90 Tex. L. Rev. 497, 505–07 (2012).</fn> Risk assessment disappeared from federal sentencing, as formal guidelines tasked judges with assigning backward-looking sentences based on retribution for past crimes and not based on the risk of future crimes.<fn>See generally John Monahan & Jennifer L. Skeem, Risk Redux: The Resurgence of Risk Assessment in Criminal Sanctioning, 26 Fed. Sent’g Rep. 158 (2014) [hereinafter Monahan & Skeem, Risk Redux]; Kirk Heilbrun et al., Risk Assessment for Future Offending: The Value and Limits of Expert Evidence at Sentencing, 53 Ct. Rev. 116 (2017); Michael Tonry, Legal and Ethical Issues in the Prediction of Recidivism, 26 Fed. Sent’g Rep. 167 (2014).</fn>

This state of affairs has changed in recent years. The role of risk assessment is increasingly prominent at all stages of the criminal justice system, including policing, pretrial detention, sentencing, corrections, reentry, and parole.<fn>See, e.g., John Monahan & Jennifer L. Skeem, Risk Assessment in Criminal Sentencing, 12 Ann. Rev. Clinical Psychol. 489, 493–94 (2016) [hereinafter Monahan & Skeem, Risk Assessment] (describing the use of risk assessment in sentencing); Sonja B. Starr, 27 Fed. Sent’g Rep. 205, 205 (2015) (“[W]e are already in the risk assessment era.”).</fn> In its most recent 2017 revision, the American Law Institute, in the influential Model Penal Code, endorsed consideration of risk in the sentencing process, and specifically, to potentially divert lower-risk defendants to reduced or alternative sentences.<fn>Model Penal Code § 6B.09(3) (Am. Law Inst., Proposed Final Draft 2017).</fn> In a separate work, John Monahan and I have examined the use of risk assessment in sentencing in Virginia and the challenges of implementation by judges.<fn>These findings are described in Anne Metz et al., Risk and Resources: A Qualitative Perspective on Low-level Sentencing in Virginia, 47 J. Community Psychol. 1476 (2019); Brandon L. Garrett et al., Judicial Reliance on Risk Assessment in Sentencing Drug and Property Offenders: A Test of the Treatment Resource Hypothesis, 46 Crim. Just. & Behav. 799 (2019); John Monahan et al., Judicial Appraisals of Risk Assessment in Sentencing, 36 Behav. Sci. & L. 565 (2018). For an overview, see Brandon L. Garrett & John Monahan, Judging Risk, 108 Calif. L. Rev. (forthcoming 2019), https://‌papers.ssrn.com/‌sol3/‌papers.cfm?‌abstract_‌id=‌‌3‌1‌9‌040‌3 [https://‌perma.cc/‌H763-‌BLPA].</fn>

Pretrial risk instruments have been implemented in federal district courts, although, as I will describe, the history of that program is quite uneven. Similarly, the use of risk and needs assessments to inform reentry programs in federal prisons has been uneven, according to a Department of Justice (DOJ) Inspector General’s Report.<fn>See U.S. Dep’t of Justice Office of Inspector Gen., Audit of Federal Bureau of Prisons’ Management of Inmate Placements in Residential Reentry Centers and Home Confinement ii (2016), https://‌oig.justice.gov/‌reports/‌2016/‌a1701.‌pdf [https://‌perma.cc/‌WEJ7-‌V7P8.</fn> As part of a federal effort to rethink the use of risk assessments for federal prisoners regarding programming and release, the Formerly Incarcerated Reenter Society Transformed Safely Transitioning Every Person (First Step) Act, enacted in 2018, makes risk assessments a centerpiece of its approach towards federal prisoners.<fn>First Step Act of 2018, Pub. L. No. 115-391, 132 Stat. 5194.</fn> The Act mentions risk no less than one hundred times. The Act relies on risk assessments in several different ways, including to allocate prison programming and ultimately, prisoner release.<fn>Id.</fn> The law has been called perhaps the most far-reaching federal sentencing reform in a generation.<fn> Deborah Barfield Berry, Senate Passes First Step Act with Push from Criminal Justice Groups; Bill Goes to House, USA Today, https://‌www.usatoday.com/‌story/‌news/‌politics/‌2018/‌12/‌18/‌first-‌step-‌act-‌passes-‌senate-‌criminal-‌justice-‌bill-‌moves-‌house/‌2356547002 [https://‌perma.cc/‌U7EE-‌EE8L] (last updated Dec. 19, 2018).</fn> That said, despite the prominent endorsement of consideration of risk in the 2017 revision of the Model Penal Code, the Federal Act does not suggest that risk factors into sentencing.

The Act has many notable features. It tasks the Attorney General with duties to conduct “ongoing research and data analysis.”<fn>First Step Act of 2018, Pub. L. No. 115-391, sec. 101, § 3631(b)(3), 132 Stat. 5194, 5195.</fn> It calls for empirically-validated recidivism reduction programming in prisons.<fn> Id. sec. 101, § 3632(b).</fn> It calls for unusual transparency in the use of risk assessments, including the publication on the DOJ website of the risk and needs assessment system with annual review and statistical validation, as well as evaluation of any “unwarranted disparities,” including as “among similarly classified prisoners of different demographic groups.”<fn>Id. sec. 101, § 3631(b).</fn> No use of algorithmic decisionmaking in any government context has ever called for such ongoing review to make public the analysis of whether bias has entered into decisionmaking. In another novel provision, the Attorney General must periodically report to Congress concerning an “assessment of budgetary savings resulting from this subchapter,” including savings from decreases in recidivism, together with “a strategy to reinvest the savings.”<fn>Id. sec. 101, § 3634(7).</fn>

However salutary these features are, the devil will be in the details as this Act is implemented. A range of federal actors are tasked with applying empirically-validated methods in contexts in which data may be lacking, at least initially. The Act does not specify the particular risk assessment to be used by federal corrections officials, but rather, it tasked a newly formed group with selecting or developing a risk assessment instrument to be used to classify all eligible federal prisoners (sixty-eight crimes are excluded from the coverage of the Act).<fn> Id. sec. 101, § 3632(d)(4)(D) (“A prisoner is ineligible to receive time credits under this paragraph if the prisoner is serving a sentence for a conviction under any of the following provisions of law . . . .”).</fn> Over time, the Attorney General is tasked with adjusting the approach in response to data concerning its implementation. Important lessons can be learned from efforts to implement risk assessments at the state level.

As the federal system is developed, there must be careful attention paid to how federal prison staff apply it and exercise their discretion in implementation, in research, and in litigation.<fn>See id. sec. 101, § 3632(f) (requiring that prison staff receive biannual auditing, including of their “interrater reliability,” presumably requiring, then, that there be testing done of their consistency in decisionmaking); id. § 103 (requiring that the Comptroller General conduct an audit every two years to report on the implementation of the new risk assessment used in federal prisons).</fn> There is also reason for caution, given how risk and needs assessments have been implemented in other state and federal settings, including pretrial and during sentencing.<fn>See generally Megan Stevenson, Assessing Risk Assessment in Action, 103 Minn. L. Rev. 303 (2018); Richard Berk, An Impact Assessment of Machine Learning Risk Forecasts on Parole Board Decisions and Recidivism, 13 J. Experimental Criminology 193, 193 (2017).</fn> Finally, there may be litigation concerning these new federal risk assessments. The Fifth Circuit recently affirmed a federal judge’s order that the pretrial system in Harris County, Texas, violates the Due Process Clause because it employed a “flawed procedural framework.”<fn>O’Donnell v. Harris County, 882 F.3d 528, 536 (5th Cir. 2018), superseded by O’Donnell v. Harris County 892 F.3d 147 (5th Cir. 2018).</fn>

In this Essay, in Part I, I describe the use of risk assessments in federal criminal cases, beginning with the adoption of a federal postconviction risk assessment, the use of risk assessment pretrial, risk and needs assessment in federal prisons, and finally, summarizing the main provisions of the First Step Act. In Part II, I describe the First Step Act and the risk assessment approach adopted to date. In Part III, I ask what questions will arise as it is implemented and as it is reviewed and challenged in litigation.

1. Criminal Risk Assessment in the Federal Courts
2. Federal Postconviction Risk Assessment

As early as 1923, federal officers used actuarial risk assessment instruments for postconviction supervision.<fn>See James B. Eaglin & Patricia A. Lombard, Fed. Judicial Ctr., A Validation and Comparative Evaluation of Four Predictive Devices for Classifying Federal Probation Caseloads 9 (1982), http://‌www.fjc.gov/‌sites/‌default/‌files/‌2012/‌ClssProb.‌pdf [https://‌perma.cc/‌4WPK-‌GFR6].</fn> In the 1970s, probation officers were required to “classify persons under supervision into maximum, medium, and minimum supervision categories dependent upon the nature and seriousness of the original offense, extent of prior criminal history, and social and personal background factors in the individual case.”<fn>Admin. Office of the U.S. Courts Prob. and Pretrial Servs. Office, An Overview of the Federal Post Conviction Risk Assessment 5 (2018) (quoting Administrative Office of the U.S. Courts, Guide to Judiciary Policies and Procedures, X-A Probation Manual, no. 7, Feb 15, 1979, at § 4004).</fn> However, they used a wide variety of probation or parole instruments. In the early 1980s, some of these tools were compared and evaluated to try to introduce some consistency to the practice.<fn>Eaglin & Lombard, supra note 23.</fn>

The instruments selected were found not to be very predictive, and, in response to concerns with their effectiveness, in the 1990s, the Federal Judicial Center developed a new instrument.<fn>Thomas H. Cohen & Kristin Bechtel, Removal of the Non-Scored Items from the Post-Conviction Risk Assessment Instrument: An Evaluation of Data-Driven Risk Assessment Research Within the Federal System, 81 Fed. Prob. 37 (2017); see also Pat Lombard & Laural Hooper, RPI FAQs Bulletin, Fed. Jud. Ctr., Aug. 1998, at 5.</fn> That instrument, too, had limitations, including that it could not consider changes over time, and, after considering commercial options, the Administrative Office of the U.S. Courts developed a Federal Post Conviction Risk Assessment (PCRA) to assist with postconviction supervision and treatment in 2009.<fn>Cohen & Bechtel, supra note 26; see also Lombard & Hooper, supra note 26.</fn> In 2004, the Judicial Conference of the United States endorsed adopting a “strategic approach” in probation to focus on evidence-based practices.<fn>Judicial Conference of the U.S., Rep. Proc. Jud. Conf. U.S. (Sept. 21, 2004), at 15.</fn> The probation instrument developed was used to identify higher-risk offenders that would benefit from additional supervision and services, or lower-risk offenders that do not require the same degree of supervision. The instrument was developed using data regarding over 185,000 offenders on probation or supervised relief, and, while high-risk offenders had failure rates of over thirty percent in the first six months after release and ninety-four percent over sixty months, the low-risk offenders had failure rates of only ten percent at thirty-six months and fifteen percent at six months.<fn>James L. Johnson et al., The Construction and Validation of the Federal Post Conviction Risk Assessment (PCRA), 75 Fed. Prob. 16 (2011).</fn> The instrument was revised, with the new version called the PCRA 2.0, to eliminate unscored test questions that did not add predictive power to the instrument and to also add scales to predict future violent arrest.<fn>Ralph C. Serin et al., Using a Multi-level Risk Assessment to Inform Case Planning and Risk Management: Implications for Officers, 80 Fed. Prob. 10, 10 (2016).</fn>

Federal probation has developed training manuals and conducted training so that officers understand the predictive validity of the instrument, as compared with unstructured officer assessments.<fn>Id.</fn> Research suggests that, since the adoption of these PCRA instruments, probation officers have fewer contacts with the low-risk scored persons, and there was no change in recidivism rates. However, there was no corresponding increase in supervision contacts with the higher-risk persons, perhaps, the authors concluded, because the federal system has had an increase in the proportion of high-risk persons.<fn>Thomas H. Cohen et al., The Supervision of Low-Risk Federal Offenders: How the Low-Risk Policy Has Changed Federal Supervision Practices Without Compromising Community Safety, 80 Fed. Prob. 3, 11 (2016).</fn>

1. Federal Pretrial Risk Assessment

Risk assessment in federal district courts provides a cautionary tale. Beginning in 1982, federal legislation created pretrial services offices in every federal judicial district, except the District of Columbia, and authorized pretrial and probation officers to collect and report on information regarding release recommendations, supervision, and outcomes to the DOJ and federal courts.<fn>18 U.S.C. § 3152 (2012).</fn> The Bail Reform Act of 1984 required federal officers to consider defendant dangerousness and flight risk when making these pretrial decisions, requiring detention for certain charges and identifying factors to be considered when making pretrial release decisions.<fn> Id. §§ 3141–150.</fn> Judges must ultimately make the pretrial release determination, after hearing from the parties and pretrial services officers.<fn>Id. § 3142.</fn> In the 1980s, an increase in both drug and immigration enforcement corresponded with an increase in the number of defendants detained pretrial. A 2007 study found that far more people were now being supervised pretrial, who previously would have been released from supervision.<fn>James Byrne & Jacob Stowell, The Impact of the Federal Pretrial Services Act of 1982 on the Release, Supervision, and Detention of Pretrial Defendants, 71 Fed. Prob. 31 (2007).</fn>

In response, the Office of the U.S. Courts and the Office of Federal Detention Trustee, which runs pretrial detention in the federal system, began to investigate whether there were “relevant predictors of pretrial risk outcome” that could be used to identify persons “most suited for pretrial release without jeopardizing the integrity of the judicial process or the safety of the community.”<fn>Marie VanNostrand & Gena Keebler, Pretrial Risk Assessment in the Federal Court, 73 Fed. Prob. 33 (2009).</fn> In 2009, it published a report, Pretrial Risk Assessment in the Federal Court, recommending development of a risk assessment instrument. The tool that was developed used data concerning over 200,000 defendants between fiscal years 2001 and 2007. The instrument measures eleven scored items, including prior felony convictions, failures to appear in court, nature and type of current charge, level of education, and citizenship status. A score results in a low, moderate, and high-risk classification. The federal Pretrial Risk Assessment (PTRA) was implemented in November 2009, and by August 2011, it was being used in most federal districts.<fn>Timothy P. Cadigan & Christopher T. Lowenkamp, Implementing Risk Assessment in the Federal Pretrial Services System, 75 Fed. Prob. 30, 30 (2011).</fn> The goal was to help federal officials assess the likelihood that federal defendants would be rearrested for violent crimes, failure to make court appearances, or violations of conditions of release. A series of studies has validated the predictive ability of the instrument. A more recent study revalidated the instrument and found that it predicted new criminal arrest activity equally as between persons of different races and genders.<fn>Thomas H. Cohen et al., Revalidating the Federal Pretrial Risk Assessment Instrument (PTRA): A Research Summary, 82 Fed. Prob. 23, 28 (2018).</fn> The instrument is not binding; it is advisory, and judges have discretion to consider the PTRA score or not.

However, there is also evidence that pretrial release rates have declined over the past decade: researchers at the Probation and Pretrial Services Office of the Administrative Office of the U.S. Courts confirmed that the decline is observed even if one controls for changes in the federal defendant population, such as the increase in immigration filings and the decrease in financial crime filings.<fn>Thomas H. Cohen & Amaryllis Austin, Examining Federal Pretrial Release Trends over the Last Decade, 82 Fed. Prob. 3 (2018).</fn> Even defendants with light criminal history profiles saw higher detention rates. The cost of federal pretrial detention is almost $1.5 billion a year.<fn>Matthew G. Rowland, The Rising Federal Pretrial Detention Rate, in Context, 82 Fed. Prob. 13, 13 (2018).</fn> Yet, following the adoption of the PTRA, federal pretrial detention reached record high levels.<fn>Id. at 15.</fn> This is despite the fact that all federal districts have judge-supervised treatment programs, together with informal community-based programs, which could provide additional alternatives to detention pretrial.<fn> Id. at 20.</fn>

Thus, the PTRA’s implementation “has not been associated with rising pretrial release rates,” and, instead, during the very time period that the PTRA was implemented, the federal system “has become more oriented towards pretrial detention than release over the last 10 years.”<fn>Cohen & Austin, supra note 40, at 10.</fn> Why is this so? One reason may be that judges were not involved in the design or adoption of the instrument.<fn>Id. at 11.</fn> However, additional reasons may have to do with the manner in which the information is presented and its relevance to the legal framework that federal judges apply. After all, the Bail Reform Act and other federal laws already specified considerations that judges should follow, and the PTRA is not integrated into those preexisting standards.<fn> Id.</fn> Indeed, judges may view the PTRA as inconsistent with the demands of the Bail Reform Act. Interestingly, the researchers also found that, even in the “few districts” that include PTRA scores in bail reports to judges, pretrial release rates have also declined.<fn>Id. at 11–12.</fn> The authors of that study concluded that “changing court culture is a difficult task” and developing a risk instrument is not enough.<fn>Id. at 12.</fn>

1. Risk and Needs Assessment in Federal Prisons

While for decades federal prison officials made assessments of prisoner dangerousness subjectively, by the 1970s, new actuarial tools were beginning to be used in corrections.<fn>Nathan James, Risk and Needs Assessment in the Federal Prison System, Cong. Res. Serv. 2 (2018), http:/‌/‌fas.org/‌sgp/‌crs/‌misc/‌R44087.‌pdf [https://‌perma.cc/‌N7KZ-‌BEXC].</fn> The Federal Bureau of Prisons (BOP) developed a tool to determine inmate security and custody classifications. However, it had no assessment to decide whether inmates should be placed in rehabilitative programs; that decision was based on discretion and any evaluations by psychologists or case managers.<fn>Id.; see also Bureau of Prisons, U.S. Dep’t Justice, Inmate Security Designation and Custody Classification, Program Statement P5100.08 (2006), http://www.bop.gov/‌policy/‌progstat/‌5100_‌008.‌pdf [https://‌perma.cc/‌22FQ-‌J8XK].</fn>

The BOP had also been using risk and needs assessments to decide whether to place federal prisoners in residential reentry centers (RRC), or halfway houses, as well as direct home confinement, while serving the remainder of their sentences, pursuant to the Second Chance Act of 2007.<fn>18 U.S.C. § 3624(g)(2) (2012).</fn> The DOJ’s Office of Inspector General conducted an audit of that program in 2016, following DOJ efforts to improve reentry programs in the federal system.<fn>U.S. Dep’t Justice, Roadmap to Reentry, Reducing Recidivism Through Reentry Reforms at the Federal Bureau of Prisons (2016), https://‌www.justice.gov/‌archives/‌reentry/‌file/‌844356/‌download [https://‌perma.cc/‌33UB-‌LLRM].</fn> The Inspector General auditors concluded that inmate risk and needs were not being carefully considered: “contrary to BOP policy, BOP guidance, and relevant research, BOP’s RRC and home confinement placement decisions are not based on inmate risk for recidivism or need for transitional services.”<fn>See U.S. Dep’t Justice Office Inspector Gen., Audit of Federal Bureau of Prisons’ Management of Inmate Placements in Residential Reentry Centers and Home Confinement ii (2016), https:/‌/‌oig.justice.gov/‌reports/‌2016/‌a1701.‌pdf [https://‌perma.cc/‌4B4K-‌5D2L].</fn> They also found that the BOP was underutilizing direct home confinement for low-risk prisoners, which also meant that there was not room in RRCs for higher-needs and higher-risk prisoners before their release dates.<fn>Id.</fn> They found that the BOP was not doing a sound job of assessing the quality of programming provided by the RRC or halfway house providers. <fn> Id. at iii, 6 (“Finally, we found that BOP does not have performance measures that evaluate the efficacy of its RRC and home confinement programming, nor does BOP have procedures in place that adequately assess the quality of services provided by RRC contractors.”).</fn>

By 2018, legislation had been introduced in Congress that would extend risk and needs assessments into decisions regarding prison programming; that legislation would later become the basis for part of the First Step Act.<fn>Id.</fn> Yet, the experience with the Second Chance Act provided grounds for caution regarding the BOP’s resources and ability to conduct evidence-based programming and reentry work.

1. The Adoption of the First Step Act

For years, advocates had been pushing for federal sentencing reform. The Smarter Sentencing Act of 2013 and the Sentencing Reform and Corrections Act (SRCA) of 2015 had been introduced and did not pass; the SRCA was reintroduced in 2017. In early 2018, the Prison Reform and Redemption Act was introduced by Democratic Senator Sheldon Whitehouse and Republican Senator John Cornyn and renamed the First Step Act. That Bill did not include sentencing reforms but focused on federal prison conditions and programs.<fn>For a description of this legislative history, see Ames Grawert & Tim Lau, How the First Step Act Became Law—and What Happens Next, Brennan Ctr. for Just. (Jan. 4, 2019), https://‌www.brennancenter.org/‌blog/‌how-‌first-‌step-‌act-‌became-‌law-‌and-‌what-‌happens-‌next [https:// ‌perma.cc/‌28GC-‌XLBM].</fn> A compromise was reached in November 2018, when provisions from the SRCA were included in the First Step Act.<fn>New Compromise on Federal Criminal Justice Reform Should Be Priority for Congress, Brennan Ctr. for Just. (Nov. 12, 2018), https://‌www.brennancenter.org/‌press-release/‌new-compromise-‌federal-‌criminal-‌justice-‌reform-‌should-‌be-‌priority-‌congress [https://‌perma.cc/‌W9GZ-‌9ZDQ].</fn> The Act had overwhelming bipartisan support. Senator Chuck Grassley, one of the supporters, called it a “once in a generation” piece of legislation.<fn>Berry, supra note 14.</fn> A wide range of civil rights groups such as the National Urban League, the American Civil Liberties Union, #cut50, as well as conservative groups, such as the Texas Public Policy Foundation, praised the legislation.<fn>Id.</fn>

That said, some chiefly opposed the sentencing changes; a coalition of civil rights groups argued that not only did federal prisons already have the authority to improve conditions and assignment of prisoners, but that:

[B]y building and placing a “risk and needs assessment” algorithm in the hands of the Attorney General—one not required to be designed or tailored for the individuals it is meant to judge—we risk embedding deep racial and class bias into decisions that heavily impact the lives and futures of federal prisoners and their families.<fn>Letter from The Leadership Conference on Civil and Human Rights et al. to Members of the House of Representatives (May 21, 2018), https:/‌/‌civilrights.org/‌resource/‌‌vote-no-first-step-act-2 [https://‌perma.cc/‌TM7J-‌L6XY].</fn>

They also opposed overboard exclusions from early release.<fn>Id. (“The majority of people in prison will eventually be released. Categorically excluding entire groups of people from receiving early-release credits will undermine efforts to reduce prison overcrowding and improve public safety since such exclusions weaken the incentive to participate in recidivism-reduction programming.”).</fn> Similarly, the American Federation of Government Employees, AFL-CIO, and the Council of Prison Locals called for sentencing reform and endorsed the SRCA, but not the First Step Act. They highlighted that federal prisoners are suffering from “sustained cuts” to staffing levels, which reduces access to rehabilitative programming.<fn>See Letter from E.O. Young, Nat’l President, American Fed’n of Gov’t Emps. Council of Prison Locals C-33, to Chairman Grassley, Ranking Member Feinstein, Chairman Goodlatte, Ranking Member Nadler, and Senator Durbin (May 8, 2018), https:/‌/‌www.afge.org/‌‌globalassets/‌documents/‌bop/‌‌first-step-letter-5-11-18.pdf [https://‌perma.cc/‌PQB9-‌FM4G].</fn> They argued that “[t]he current risk assessment system in use by the BOP has been developed and refined over many years, and hastily creating a new system that is untested could put the safety of correctional officers at risk.”<fn>Id. at 2.</fn> Further, they argued that there was not sufficient funding to implement such a system, given staffing shortages and that the current budgets had already eliminated far too many federal corrections positions.<fn>Id. at 3.</fn>

Other groups endorsed the First Step Act, including #cut50, which called it a “MAJOR breakthrough in the long-fought battle for criminal justice reform,” although noting that they were “disappointed in some of the concessions made to appease the minority of GOP members who had concerns about this bill.”<fn>About the First Step Act, #FirstStepAct, https://www.firststepact.org‌/‌about [https://‌perma.cc/‌CMQ9-‌ZUNJ]; see also H.R. 5682 (115th): First Step Act, govtrack, https://‌www.govtrack.us/‌congress/‌bills/‌115/‌hr5682 [https://‌perma.cc/‌P7VJ-‌Z9Z5].</fn>

The federal First Step Act, enacted on December 21, 2018, thus includes as a compromise both sentencing and federal prisoners related provisions. Regarding sentencing, the Act includes a range of measures to alter federal criminal sentencing, including the (1) broadening of the Safety Valve provision, (2) reduction and restriction (but not elimination) of mandatory minimum drug penalties, (3) eliminating “stacking” of penalties under 18 U.S.C. 924(c)(1)(A), and (4) making the Fair Sentencing Act of 2010 retroactive to apply to about 3,000 people convicted of crack offenses before that Act took effect. Those sentencing changes all took effect immediately and will impact tens of thousands of defendants each year. Already, there have been over 800 sentence reductions and 643 early releases under the Fair Sentencing Act provisions of the First Step Act.<fn>Press Release, U.S. Dep’t of Justice, Department of Justice Announces First Step Act Implementation Progress (Apr. 8, 2019) [hereinafter Press Release], https://‌www.justice.gov/‌opa/‌pr/‌department-‌justice-‌announces-‌first-‌step-‌act-‌implementation-‌progress [https://‌perma.cc/‌MJ6C-‌2M49].</fn>

The Act also, regarding federal prisons, addresses adoption of Medication Assisted Treatment (MAT) in facilities for inmates with substance abuse problems; improvements to federal prison conditions, including banning shackling of pregnant women; provision of funding of $75 million per year for new programming and assessments aimed at reducing recidivism; de-escalation training to BOP employees; and permitting “good time credit,” to reduce sentences by up to about fifty days a year for good behavior (a provision that will benefit large numbers of prisoners).

The focus here, however, is the set of provisions in the Act that tasks the DOJ and the BOP with design and implementation of a new risk assessment instrument. Where federal prisons had used actuarial tools informally in the past, and not for decisions regarding rehabilitative programs or release, the Act adopts risk assessment more broadly. A new risk and needs instrument was to be created under the Act. This instrument will be used to classify all 1.8 million federal prisoners. Unlike the sentencing provisions that take effect immediately, the risk assessment portions of the Act, which apply to all prisoners, take effect over time. The timelines imposed for this effort are fairly short: 210 days to select or develop a risk assessment instrument (and release it “publicly on the Department of Justice website”) and then 180 days to implement and administer this new system to all federal prisoners and set their risk score at one of four levels: minimal, low, medium, and high.<fn>See First Step Act of 2018, Pub. L. No. 115-391, sec. 101, § 3632(a), sec. 102, § 3621(h)(1), 132 Stat. 5194, 5196, 5208.</fn> The somewhat lengthier timeline extends the time period for implementation, responding, at least in part, to the concern that there was not sufficient time to implement this new approach.

1. Evaluating the First Step Act

A risk and needs assessment has now been developed under the First Step Act, with a report describing the development and design of that tool; it is pending approval as of this writing.<fn>Id. sec. 101, § 3632(a) (“Development of risk and needs assessment system . . . In carrying out this subsection, the Attorney General may use existing risk and needs assessment tools, as appropriate.”); U.S. Dep’t Justice, The First Step Act of 2018: Risk and Needs Assessment System (2019) [hereinafter Department of Justice Report] (describing development and adoption of the risk and needs assessment system).</fn> Assuming the instrument is approved, evaluations will be ongoing under the Act, which requires reassessment.<fn>First Step Act of 2018 sec. 101, § 3632(a)(4) (“[R]eassess the recidivism risk of each prisoner periodically, based on factors including indicators of progress, and of regression, that are dynamic and that can reasonably be expected to change while in prison . . . .”).</fn> There are certain safeguards built into the legislation, however, which, if implemented well, can help to prevent use of biased, unfair, or nonpredictive assessments, as well as poor decisionmaking based on the assessment scores. In this Part, I evaluate the approach adopted in the Act.

1. Developing Risk and Needs Assessment

The Act contains many of the elements of a sound risk and needs assessment approach. At a basic level, it involves the adoption of risk assessment through legislation. Federal risk assessment, and many state efforts to use risk assessment, had often in the past been adopted by judges, prison officials, or others informally and without involving the public or key decisionmakers. The Act was enacted through a highly bipartisan process, with public input and deliberation. The Act specifies prisoners must be classified into one of four risk categories—minimum, low, medium, and high.<fn>See id. sec. 101, § 3632 (describing the “Development of risk and needs assessment system” and providing: “(a) In general. Not later than 210 days after the date of enactment of this subchapter [enacted Dec. 21, 2018], the Attorney General, in consultation with the Independent Review Committee authorized by the First Step Act of 2018, shall develop and release publicly on the Department of Justice website a risk and needs assessment system”).</fn> The Act does not specify what thresholds or what likelihoods of recidivism should be reflected in defining each of these categories, and it notes that the Attorney General may select existing risk and needs assessment tools, if appropriate.<fn>Id. sec. 101, § 3632(a).</fn> The Act states that “evidence-based recidivism reduction programs” will be adopted and assigned to prisoners based on the risk and needs assessment tools selected.<fn>Id. sec. 101, § 3632(b).</fn> The Act includes a broad definition of such programs, but it states that they must be “shown by empirical evidence to reduce recidivism” or “based on research indicating that it is likely to be effective in reducing recidivism.”<fn>Id. sec. 101, § 3635(3)(A).</fn> The Act states that prisoners will be grouped in housing based, if practicable, on these risk and needs scores.<fn>Id. sec. 101, § 3632(c).</fn> Further, and most importantly, a range of incentives will apply to prisoners who participate in these “evidence-based recidivism reduction programs,” including increased phone and visitation privileges, transfer to institutions closer to release,<fn>Id. sec. 101, § 3632(d)(1)–(2).</fn> and time credits, in which a prisoner who completes this programming “or productive activities” can earn ten days of time credit for every thirty days of successful participation, and, for lower-risk prisoners, an additional five days of credit for thirty days of participation.<fn>Id. sec. 101, § 3632(d)(4).</fn> The Act then lists a lengthy series of crimes that make inmates presumptively ineligible for rehabilitative programming, which override any risk and needs determination.<fn>Id. sec. 101, § 3632(d)(4)(D) (“A prisoner is ineligible to receive time credits under this paragraph if the prisoner is serving a sentence for a conviction under any of the following provisions of law . . . .”).</fn> In addition, prisoners may also have their risk level updated over time, with more frequent risk assessments for prisoners with less than five years until release.<fn>Id. sec. 101, § 3632(d)(5) (“A prisoner who successfully participates in evidence-based recidivism reduction programming or productive activities shall receive periodic risk reassessments not less often than annually, and a prisoner determined to be at a medium or high risk of recidivating and who has less than 5 years until his or her projected release date shall receive more frequent risk reassessments. If the reassessment shows that the prisoner’s risk of recidivating or specific needs have changed, the Bureau of Prisons shall update the determination of the prisoner’s risk of recidivating or information regarding the prisoner’s specific needs and reassign the prisoner to appropriate evidence-based recidivism reduction programming or productive activities based on such changes”).</fn>

The Act specifies in some detail the process by which the risk and needs instrument is to be designed and evaluated over time, as well as who will engage in that work. The Act requires the appointment of an Independent Review Committee,<fn>Id. sec. 101, § 3632(a); see also id. sec 107, § 3631 (describing composition, appointment, and duties of the Independent Review Committee).</fn> consisting of a group that includes researchers who have studied risk assessment to vet possible instruments and report to the Attorney General, who makes the final decisions regarding the system to be adopted and who has duties under the Act to conduct “ongoing research and data analysis.”<fn>Id. sec. 101, § 3631(b) (describing Attorney General duties to conduct “ongoing research and data analysis” on “(A) evidence-based recidivism reduction programs relating to the use of prisoner risk and needs assessment tools; (B) the most effective and efficient uses of such programs; (C) which evidence-based recidivism reduction programs are the most effective at reducing recidivism, and the type, amount, and intensity of programming that most effectively reduces the risk of recidivism”); see also id. sec. 101, § 3632(a).</fn> That Committee was established in April 2018, with the selection of the nonprofit Hudson Institute to host the Committee, and with the National Institute of Justice contracting with outsiders to help develop the Risk and Needs Assessment System.<fn>Press Release, supra note 67.</fn> The Act also specifies that the “crimogenic needs” of prisoners should guide the provision of resources for rehabilitative programming in prisons.<fn>First Step Act of 2018 sec. 101, § 3632(a)(5)(B); see also Monahan & Skeem, Risk Assessment, supra note 8, at 497 (“[V]ariable risk factors have been confused with causal risk factors under the rubric of ‘needs,’ ‘criminogenic needs,’ or ‘dynamic risk factors.’ The latter phrases are often misused as synonyms for causal risk factors—they typically reference risk factors that theoretically can be changed through intervention to reduce risk, but empirically have not been shown to do so. Most ‘needs’ are variable risk factors, given the current state of evidence.”).</fn> It does not define what those risks or needs might be or what types of programming might be linked to those risks or needs. As noted, the federal prisons currently use an instrument, but it is only designed to help set security levels and predict misconduct while in prison.<fn>Charles Colson Task Force on Federal Corrections, Transforming Prisons, Restoring Lives: Final Recommendations of the Charles Colson Task Force on Federal Corrections 32 (2016), https://‌www.urban.org/‌research/‌publication/‌transforming-prisons-restoring-lives/‌view/‌full_‌report [ https://‌perma.cc/‌QL8B-‌UYEP].</fn>

However, there are risk and needs instruments that have been developed in other settings: the Act states that an existing model or instrument may be adopted. The “dominant paradigm” in corrections is the Risk-Needs-Responsivity (RNR) model.<fn>James, supra note 49, at Summary.</fn> Questions are asked concerning static factors, such as age, gender, past criminal offenses, or prior probation violations. Questions are also asked concerning “dynamic” factors relating to “crimogenic needs,” such as those related to employment, substance abuse, and mental health.<fn>For a description, see James Austin, The Proper and Improper Use of Risk Assessment in Corrections, 16 Fed. Sent’g Rep. 1, 5 (2004).</fn> Next, services are provided that are tailored to the person in question (called the “responsivity principle”).<fn>See James, supra note 49, at 6 (“The responsivity principle is further divided into two elements. The general responsivity principle states that cognitive-behavioral and social learning therapies are the most effective form of intervention. The specific responsivity principle states that treatment should consider the relevant characteristics of the offender (e.g., the offender’s motivations, preferences, personality, age, gender, ethnicity, and cultural identification, along with other factors).”).</fn> There may be a “natural limit” to the predictive power of these instruments, as John Monahan and Jennifer Skeem have explained.<fn>Monahan & Skeem, Risk Redux, supra note 7.</fn> However, when tailoring programs, there is evidence that it is beneficial to target resources towards high-risk individuals. There is corresponding evidence that intensive programming can be counterproductive for low-risk individuals.<fn>See James, supra note 49, at 7 (“This could be because placing low-risk offenders in intensive programming interrupts support structures or self-correcting behaviors that already exist, or because it exposes low-risk offenders to high-risk offenders who may have a negative influence on low-risk offenders’ thoughts or behaviors.”); see also James Bonta & D.A. Andrews, Risk‒Need‒Responsivity Model for Offender Assessment and Rehabilitation (2007).</fn>

The risk and needs assessment system selection was announced in July 2019, after a process of consultation between experts and stakeholders, the BOP, the National Institute of Justice (NIJ), and the members of the Independent Review Commission.<fn> See Department of Justice Report, supra note 69, at 14–16.</fn> The new tool, validated as described in the DOJ Report upon its release using BOP datasets, is the Prisoner Assessment Tool Targeting Estimated Risk and Needs (PATTERN).<fn>See id. at 43.</fn> The report announcing this new tool states that, while operating “under an extremely short timeline,” the tool achieves a “high level of predictive performance and surpasses what is commonly found” in risk assessment tools used in the United States.<fn> Id.</fn> Further, the tool uses dynamic factors, allowing inmates to improve their scores over time.<fn> Id.</fn> The developers describe their assessment of how the tool performed on BOP data and their finding that it improves upon current BOP risk assessment, using the BRAVO-R tool in its predictive ability and “parity across race and ethnic classifications.”<fn> Id. A dataset of BOP releases from 2009–2015 was used to develop and validate the new PATTERN instrument. Id. at 49.</fn> The dynamic factors used in the tool include infractions during incarceration, number of programs completed by the inmate, number of technical and vocational courses, federal industry employment, drug treatment, drug education, and payment of victim restitution.<fn> Id. at 45–46.</fn> The static factors used include age at time of first conviction, age at time of assessment, whether the crime of conviction was violent, and whether the individual was identified as a sex offender.<fn> Id.</fn> In addition, the BOP expanded assessment and accommodation of disabilities, including dyslexia, as required by the Act.<fn>Id. at 61.</fn>

The Act states that the risk and needs instrument will be annually reviewed, validated, and released publicly on the DOJ website.<fn>First Step Act of 2018, Pub. L. No. 115-391, sec. 101, § 3631(b)(4), 132 Stat. 5194, 5195.</fn> No prior legislation has taken that step. As a result, the public and researchers can evaluate it each year. The Act also calls for a “statistical validation of any tools that the risk and needs assessment system uses” every year.<fn>Id. sec. 101, § 3631(b)(4)(D).</fn> Criminal justice legislation does not normally call for such statistical validation either; this language is remarkable.

In addition, this annual review is to include an “evaluation of the rates of recidivism among similarly classified prisoners to identify any unwarranted disparities, including disparities among similarly classified prisoners of different demographic groups . . . .”<fn> Id. sec. 101, § 3631(b)(4)(E).</fn> No prior legislation has taken that step either. This is an important response to calls that algorithmic decisionmaking in general must be carefully scrutinized to prevent bias. Finally, the Act calls for training of prison officials on the use of the risk instrument, in a continuing fashion.<fn>Id. sec. 101, § 3632(f); id. sec. 103, § 3621(5) (calling for evaluation of whether “officers and employees of the Bureau of Prisons are receiving the training described in section 3632(f)”).</fn> The Act also requires biannual assessments of prison officials’ use of the assessments.

1. Implementation and Resources

Within thirty days of the enactment of the law, the NIJ was to create an Independent Review Committee to assist in designing this new risk assessment approach. That risk assessment tool accounts for eighty percent of the $75 million allocated for the Act.<fn>Id. § 104(a).</fn> However, due to the longest partial federal government shutdown in history, which began just hours after the signing of the First Step Act, the NIJ was operating “under a lapse in appropriations and following our contingency procedure” in January 2019.<fn>Beryl Lipton, The First Step Act’s First Steps Are Stalled, Muckrock (Jan. 22, 2019), https://‌www.muckrock.com/‌news/‌archives/‌2019/‌jan/‌22/‌algorithms-first-step-act [https://‌perma.cc/‌7TL9-‌Z56F].</fn> The Commission was not formed by January 21, 2019, but on April 8, 2019, it was announced. The Committee also assisted in approving a screening tool to implement a dyslexia screening requirement, an evaluation to identify candidates for the new medication-assisted treatment (MAT) program, and procedures for “compassionate release” sentence reductions.<fn>Press Release, supra note 67.</fn>

On July 19, 2019, on schedule, the DOJ announced that a risk and needs instrument had been developed, the PATTERN, with a “high level of predictive performance.”<fn>See Department of Justice Report, supra note 69, at 63.</fn> As noted, one question that the Act does not answer is what thresholds will be set to determine who is placed in each of the four high, medium, low, and minimal risk categories.<fn> There have been attempts to use judges to develop such cut-offs. See generally John Monahan & Eric Silver, Judicial Decision Thresholds for Violence Risk Management, 2 Int’l J. Forensic Mental Health 1 (2003).</fn> That is a crucial question, as is what threshold levels are for what is considered a need that should be addressed through rehabilitative efforts. However, “[t]he ultimate description of a defendant’s risk as low, moderate, or high in a given jurisdiction is a policy decision, not a scientific one.”<fn>Desmarais & Lowder, note 1, at 4.</fn> Those policy choices were delegated to the new Independent Review Committee.<fn>Whether this Committee is an appropriate host for these efforts has been questioned; lawmakers have raised concerns regarding the selection of the Hudson Institute, with Senators on both sides of the aisle asking why a group described as an “opponent” of the Act was tasked with implementing it. Samantha Michaels, Trump Keeps Celebrating Prison Reform. His Administration’s Latest Move Could Sabotage It, Mother Jones (Apr. 11, 2019) (quoting Senator Mike Lee), https://www.motherjones.com/‌crime-justice/‌2019/‌04/‌trump-first-step-act-hudson-institute-risk-assessment-committee [https://‌perma.cc/‌9U73-‌J2EW]. As the Leadership Conference on Civil and Human Rights noted, it is not clear whether all six members of the team chosen for the Committee satisfy the legislative criteria; only three appear to have experience with design or implementation of risk assessment tools. Id. Letter from American Civil Liberties Union et al. to David B. Muhlhausen, Dir., Nat’l Inst. of Justice, Dep’t of Justice (April 12, 2019), https:/‌/‌civilrights.org/‌resource/‌statement-for-the-record-of-the-aclu-justice-roundtable-and-the-leadership-conference-in-response-to-department-of-justice-doj-april-3-and-5-listening-sessions [https://‌perma.cc/‌UN5W-‌KCGX]; see also 18 U.S.C. § 3632 (a)–(b).</fn> The report announcing the PATTERN describes the rate of recidivism associated with each category, minimum, low, medium, and high, selecting these “cut points” using recidivism base rates for the relevant populations and using, for example, half the base rate (twenty-four percent) for the low-risk threshold, under one-quarter the base rate (ten percent) for the minimum risk category, and two-thirds above (or eighty percent) for the high category.<fn>See Department of Justice Report, supra note 69, at 51.</fn> This Committee did not carefully explain how those crucial choices were made. In Virginia, while the Sentencing Commission did develop, with the assistance of researchers, the risk assessment tool, the policy goal, in terms of numbers of individuals to be diverted, was set by the Legislature.<fn>See Garrett & Monahan, supra note 10, at Section II.A.</fn> Such an approach is preferable.

The DOJ has solicited comments on the PATTERN, and as of this writing, the review process is still underway. Commenters, including this author, have raised a series of concerns with the PATTERN’s approach, including regarding selection of risk cutoffs, the reliance on technical violations and minor crimes in calculating risk scores, the use of dynamic factors, and more generally, the lack of transparency concerning the design and validation of the instrument.<fn>See, e.g., Letter from The Leadership Conference on Civil & Human Rights, to David B. Muhlhausen, Dir., Nat’l Inst. of Justice (Sept. 4, 2019), https://civilrights.org/‌resource/‌comment-letter-to-department-of-justice-on-pattern-first-step-act [https://perma.cc/‌LE4R-TZQ5]; Brandon L. Garrett & Megan T. Stevenson, Comment on PATTERN, Duke Law: Just Science Blog (Sept. 15, 2019), https://sites.law.duke.edu/‌justsciencelab/‌2019/‌09/‌15/‌comment-on-pattern-by-brandon-l-garrett-megan-t-stevenson/ [https://perma.cc/‌RZ5L-CQLV].</fn>

Once a risk system is considered and approved, the Committee must engage in a further series of reviews called for by the statute, as it assists the Attorney General, who has the responsibility to review existing systems, recommend additional “evidence-based recidivism reduction” programs,” research and analyze data on new programs, and oversee training programs, as well as monitoring the BOP’s use of the system.<fn> First Step Act of 2018, Pub. L. No. 115-391, § 107(a), (e), 132 Stat. 5194, 5215; see also id. sec. 101, § 3632(a)–(b), 3633. First Step Act of 2018 sec. 101, § 3633 states: “Prior to releasing the System,” and in consultation with the Committee, the Attorney General “shall” (1) “review the effectiveness of evidence-based recidivism reduction programs that exist” prior to enactment of the Act; (2) “review available information” concerning recidivism reduction programs in state prisons; (3) “identify the most effective evidence-based recidivism reduction programs;” (4) “review the policies for entering into evidence-based reduction partnerships;” and (5) direct the Bureau of Prisons regarding “evidence-based recidivism reduction programs” and more.</fn> The reporting and data analysis of the use of the PATTERN and these other changes will continue for years to come.

Another crucial question is whether adequate resources will be provided for both the needed programming and the implementation of the new risk assessments. A notable feature of the Act, which distinguishes it from many other pieces of criminal legislation, is that it calls for a cost-benefit analysis over time. The Attorney General must periodically report to Congress concerning an “assessment of budgetary savings resulting from this subchapter,” including savings from the transfer of prisoners to prerelease custody and the decrease in recidivism, together with “a strategy to reinvest the savings” in both law enforcement activities and further recidivism reduction programs.<fn>Id. sec. 101, § 3634(7).</fn> This model is distinct from a Justice Reinvestment model, in which savings are reinvested in further use of alternatives to incarceration and rehabilitation programming.<fn>Susan Turner & Julie Gerlinger, Risk Assessment and Realignment, 53 Santa Clara L. Rev. 1039, 1045 (2013).</fn>

However, concerns have already been raised regarding whether the Administration plans to request much of the $75 million called for under the Act.<fn>Justin George, First Step Act Comes up Short in Trump’s 2020 Budget, Marshall Project (Mar. 12, 2019, 6:00 AM), https:/‌/‌www.themarshallproject.org/‌2019/‌03/‌12/‌first-step-act-comes-up-short-in-trump-s-2020-budget [https://‌perma.cc/‌P4HN-‌UBYU].</fn> The Inspector General’s audit of the reentry programming in the BOP raised a similar problem, regarding not only the nonuse of a risk assessment, but the lack of adequate resources to conduct sufficient reentry programming. Studies of risk assessment efforts at the state level have found similar problems regarding adequacy of resources and of implementation. If decisionmakers do not share the normative goals of a risk assessment, have their own very different risk thresholds, or lack adequate resources to implement it, then the approach may not be consistently followed.<fn>Note, Bail Reform and Risk Assessment: The Cautionary Tale of Federal Sentencing, 131 Harv. L. Rev. 1125, 1145 (2018) (“The risk threshold is the most important part of risk assessment tools because it marks the compromise between the presumption of innocence, decarceration, and public safety.”).</fn>

One example comes from studies of risk assessment in Virginia. In the words of the Model Penal Code: “On risk assessment as a prison-diversion tool, Virginia has been the leading innovator among American states.”<fn>Model Penal Code § 6B.09(3) (Am. Law Inst., Final Draft, approved May 24, 2017).</fn> Alex Jakubow, John Monahan, and I found that many eligible offenders did not receive alternative sentences that they were entitled to under the Virginia scheme.<fn>Garrett et al., supra note 10.</fn> Many judges explained, in responses to a survey by the same research team, that they had resource constraints on their ability to follow the risk assessment recommendations. When asked about the availability of local treatment resources that allow for alternative sentencing—such as outpatient drug or mental health programs—seventy percent of these Virginia judges rated the existence of these resources as “less than adequate,” and five percent rated their local resources as “virtually non-existent.”<fn>Monahan et al., supra note 10.</fn> Judges’ likelihood of imposing nonjail alternative sentences on offenders increased from forty-four percent in the most resource-poor jurisdictions to seventy-one percent in the most resource-rich jurisdictions. The authors concluded that these findings confirm the “treatment resource hypothesis”—the availability of resources in the community for non-incarceration supervision and treatment—to be one factor accounting for the wide variation among courts and individual judges to the extent to which low-risk offenders actually receive a sentence of community treatment that does not include incarceration in a local jail.<fn> Id.</fn>

1. Litigation and Federal Risk Assessment

Inmates may file legal challenges to the First Step Act. The Act itself does not speak to judicial review but rather to review in which the Attorney General must submit reports to Congress two years after enactment and then every five years, along with internal review processes.<fn>First Step Act of 2018, Pub. L. No. 115-391, sec. 101, § 3634, 132 Stat. 5205–07 (“Beginning on the date that is 2 years after the date of enactment of this subchapter [enacted Dec. 21, 2018], and annually thereafter for a period of 5 years, the Attorney General shall submit a report to the Committees on the Judiciary of the Senate and the House of Representatives and the Subcommittees on Commerce, Justice, Science, and Related Agencies of the Committees on Appropriations of the Senate and the House of Representatives . . . .”).</fn> Perhaps it should have addressed judicial review to help to consolidate and organize the inevitable legal challenges. Challenges may focus on the degree to which prison officials consistently follow the recommendations made by the risk instrument. Challenges may focus on the design of the instrument itself. Such challenges will not raise the issues raised in the Loomis litigation in Wisconsin, in which the design of the risk instrument used in sentencing was not made public.<fn> State v. Loomis, 881 N.W.2d 749 (Wis. 2016); see also Alyssa M. Carlson, Note, The Need for Transparency in the Age of Predictive Sentencing Algorithms, 103 Iowa L. Rev. 303 (2017).</fn>

One barrier to any such challenge is the limitation on the ability of prisoners to challenge discretionary decisions regarding prison programming and systems permitting early release.<fn>One source of such limits is the Prison Litigation Reform Act (PLRA), 42 U.S.C. § 1997(e) (2012).</fn> In the past, though, courts have consolidated prisoner challenges to BOP regulations, such as regarding methods for calculating good time credit, while also generally deferring to the presumed reasonableness of BOP regulations.<fn>See, e.g., Barber v. Thomas, 560 U.S. 474, 480 (2010) (“Because the BOP’s administration of good time credits affects the interests of a large number of federal prisoners, we granted the consolidated petition for certiorari to consider petitioners’ challenge.”). For examples of federal deference to BOP interpretation of statutes regarding good time credits, see Sash v. Zenk, 428 F.3d 132, 136–37 (2d Cir. 2005); Tablada v. Thomas, 533 F.3d 800, 809 (9th Cir. 2008).</fn> With a standard risk and needs instrument adopted, questions will be raised concerning implementation of the instrument. One question will be disparities in outcomes, which may also enter into the system where there are categorical exclusions for certain criminal offenses.<fn>James, supra note 49, at 12 (“Data from BJS is not detailed enough to assess the implications of some of the exclusions outlined . . . with a high level of specificity, but it can provide insight into the characteristics of the federal prison population by the offense type for which they were incarcerated.”).</fn> Another question that may arise is whether officials consistently follow the recommendations made by the instrument. Moreover, if rehabilitative programming is emphasized for high-risk inmates, will lower risk inmates receive earlier release? As that programming is evaluated, what standards will inform how effective it is?

Federal courts have reviewed the constitutionality of state and local risk assessment, but not in the prison or post-conviction context. For example, the Fifth Circuit affirmed a federal district judge’s order finding that the cash bail system in Harris County, Texas, violate the Due Process Clause where it adopted a “flawed procedural framework,” in which bail decisions by presiding judges were found to have been arbitrary.<fn> ODonnell v. Harris Cty. 882 F.3d 147, 154 (5th Cir. 2018).</fn> Judges departed from release recommendations by pretrial services as much as sixty-six percent of the time.<fn>Id. at 160. However, the court held that judges need not be required to provide written decisions explaining their reasons for denying release or bail. Id. (“We decline to hold that the Constitution requires the County to produce 50,000 written opinions per year to satisfy due process.”). The court also relied on empirical analysis of outcomes in the county. Id. at 162 (citing Paul Heaton et al., The Downstream Consequences of Misdemeanor Pretrial Detention, 69 Stan. L. Rev. 711, 786–87 (2017)).</fn> Similar litigation is pending in other circuits.<fn>For an overview, see Brandon L. Garrett, Wealth, Equal Protection, and Due Process, Wm & Mary L. Rev. (forthcoming), https://papers.ssrn.com/‌sol3/‌papers.cfm?abstract_‌id=‌3313358 [https://‌perma.cc/‌E3XU-‌7494]; see also Lorelei Laird, ABA Files Amicus Brief Challenging Money Bail System as House of Delegates Considers Resolution, ABA J. (Aug. 10, 2017, 2:18 PM), http://‌www.abajournal.com/‌news/‌article/‌aba_‌amicus_‌brief_‌money_‌bail_‌5th_circuit [https://‌perma.cc/‌C78B-‌K5YZ] (describing the role of the American Bar Association in the Fifth Circuit as well as pending Sixth and Eleventh Circuit litigation). In addition, the Fifth Circuit has reversed as overly expansive an injunction entered on remand that called for immediate release of any inmate not provided with a hearing within forty-eight hours. O’Donnell v. Goodhart, 900 F.3d 220, 228 (5th Cir. 2018).</fn> In addition, local jurisdictions have used risk assessment while under federal court supervision due to jail overcrowding or litigation concerning pretrial procedures or conditions.<fn> See, e.g., Balla v. Idaho State Bd. Of Corr., 595 F. Supp. 1558, 1571 (D. Idaho 1984) (finding that a system of classifying inmates based on “a primary foundation of risk assessment” is more objective and adequate); Braggs v. Dunn, 257 F. Supp. 3d 1171, 1219 (M.D. Ala. 2017) (finding that the agency violated the Eighth Amendment in failures regarding mental health care in prison, including based on failure to use an appropriate risk assessment tool to determine suicide risk).</fn> Such cases have, so far, not involved challenges to risk and needs assessments conducted by local jails or prisons, however.

III.     Improving Federal Risk and Needs Assessments

If the First Step Act is just a first step, and if it is successfully implemented, how could it set the stage for further efforts? First, will it succeed on its own terms? It is hard to say without information regarding its implementation. One important criticism of risk assessment tools is that they can rely on information that is correlated with race in a manner that exacerbates racial bias.<fn>Sandra G. Mayson, Bias In, Bias Out, 128 Yale L.J. (forthcoming 2019); Shima Baradaran Baughman, Dividing Bail Reform, Iowa L. Rev. (forthcoming), https://papers.ssrn.com/‌sol3/‌papers.cfm?‌abstract_‌id=‌3358662 [https://‌perma.cc/‌MD64-‌VQ6Z]; Sonja B. Starr, Evidence-Based Sentencing and the Scientific Rationalization of Discrimination, 66 Stan. L. Rev. 803, 806 (2014).</fn> The Act is notable in that it calls for ongoing evaluation to examine whether such bias exists in its implementation. Whether the risk tool selected under the Act does so, whether it relies on other demographic variables that raise risk of bias, and whether it is carefully reviewed over time to address any resulting bias remains to be seen.

Second, is it a model for sound use of risk assessment in criminal justice? John Monahan and I, in prior work, set out key features of a sound use of risk assessment in criminal justice: (1) public specification of criteria for risk assessment instruments; (2) defining the relevant risks and needs to be measured; (3) making the risk instrument public and accessible to researchers; (4) presenting risk information in a comprehensible way to decisionmakers; (5) structuring decisionmaking to better make use of that information; and (6) accompanying these reforms with ongoing monitoring, through judicial review and by making data accessible to researchers.<fn>Garrett & Monahan, Judging Risk, supra note 10.</fn> Some scholars object to use of prediction generally on grounds that just punishment should not be based on future outcomes; as John Monahan and I have described, use of risk assessment in sentencing is incompatible with a retributivist approach to punishment, but it is compatible with a limited retributivist approach.<fn>See generally Bernard E. Harcourt, Against Prediction: Profiling, Policing, and Punishing in an Actuarial Age (2007).</fn> The Act, however, focuses not on sentencing, but on prison conditions and release, which does not as directly implicate punishment theory.

The First Step Act satisfies many of the criteria that we set out for use of risk instruments in criminal justice generally. There are public specifications of at least basic criteria and some definition of relevant risks. However, much of the empirical validation appears to come only after the system is already adopted. The risk instrument is to be made public and accessible. We do not know how the risk information will be presented, although we do know that there will be training on it, including both initial training and continuing education.<fn>First Step Act of 2018, Pub. L. No. 115-391, sec. 101, § 3632(f)(1)–(2), 132 Stat. 5194, 5204.</fn> We do not know how information will be presented or structured into existing decisionmaking.

We do know that there will be ongoing auditing and review of the use of the new risk assessment system. Remarkably, the statute includes “a requirement that [BOP staff] demonstrate competence in administering the System, including interrater reliability, on a biannual basis.”<fn>Id. sec. 101, § 3632(f)(4).</fn> I am not aware of any federal statute, and certainly not in criminal law, that requires an examination of interrater reliability. We do not know whether the approach adopted will actually improve outcomes, including regarding recidivism.<fn>Megan Stevenson, Assessing Risk Assessment in Action, 103 Minn. L. Rev. 304 (2018).</fn> The Act, while not creating a perfect model, represents an extremely large-scale adoption of risk assessment, and a step forward in thinking about how to appropriately legislate such assessments.

Third, the Act does not introduce risk assessments to the initial sentencing process. A system in which lower risk offenders would be better identified and released pretrial could make an important contribution at the federal level. As discussed, there is evidence that the pretrial assessments are often not followed when release decisions are made for several important reasons, including poor integration with bail statutes.<fn>Id.</fn> Sentencing guidelines similarly do not reflect any consideration of risk. There are alternative sentencing and diversion programs in the federal courts, including pilot programs that focus on treatment alternatives to incarceration.<fn>U.S. Sentencing Comm’n, Federal Alternative-to-Incarceration Court Programs (2017) (“Many federal district courts around the country, with the support of the Department of Justice (DOJ), have begun creating specialized court programs to increase the use of alternatives to incarceration for certain types of offenders, most commonly for those with substance use disorders.”); Thomas E. Ulrich, Pretrial Diversion in the Federal Court System, 66 Fed. Prob. 30 (2002) (describing history and usage of pretrial diversion in federal courts, but noting that drug treatment is a condition of participation for a small percentage of participants).</fn> For a risk-based approach to function, there must be resources dedicated to such efforts.

Conclusion

Risk assessments now are used in federal courts pretrial, postconviction, and in federal prisons. These assessments have not enjoyed clear successes in the past, but in some states (and not in others) assessments have resulted in change in detention practices. As one observer has noted, despite many cautionary lessons from uneven or failed efforts to implement risk assessments, “[i]f risk assessments are paired with adequate safeguards, sustained reductions in incarceration and progress toward equal treatment may be possible.”<fn> Note, supra note 116, at 1125.</fn> The First Step Act, which has the ambition to transform reliance on risk, has the potential to introduce a more legitimate, transparent, and validated approach. Whether it will be successfully implemented, using a sound risk assessment instrument, however, remains to be seen, as of this writing. Prior efforts in the federal system, as well as at the state and local levels, suggest reason for caution and care. It may take longer than anticipated, and a greater investment of resources, to use risk assessments to improve