Printer Friendly

Dangerous Defendants.

AUTHOR. Assistant Professor of Law, University of Georgia School of Law. For extremely helpful input, I am indebted to Laura Appleman, David Ball, Shima Baradaran Baughman, Richard Berk, Mitchell Berman, Stephanos Bibas, Kiel Brennan-Marquez, Jessica Eaglin, Barry Friedman, Lauryn Gouldin, Rachel Harmon, Paul Heaton, John Hollway, Mark Houldin, Douglas Husak, Samuel Issacharoff, Orin Kerr, Seth Kreimer, Jim Jacobs, Craig Konnoth, Joanna Langille, Sophia Lee, Youngjae Lee, Richard Lippke, Stephen Morse, Anna Roberts, David Rudovsky, Tim Schnacke, Larry Schwartztol, Jocelyn Simonson, Megan Stevenson, Alec Walen, Rebecca Wexler, Sam Wiseman, participants of the University of Pennsylvania Fellows' Workshop, participants of the Quattrone Center's Lunch Workshop Series, attendees of the bail panel at CrimFest 2016, and attendees of faculty presentations at Berkeley Law, Boston University School of Law, Brooklyn Law School, Louisiana State University Paul M. Hebert Law Center, Roger Williams University School of Law, Rutgers Law School, Sandra Day O'Connor College of Law, Southern Methodist University Dedman School of Law, University of Georgia School of Law, and University of Utah S.J. Quinney College of Law. For institutional support, I am grateful to University of Pennsylvania Law School, the Quattrone Center for the Fair Administration of Justice, and University of Georgia School of Law. Heather Richard and the editorial team at the Yale Law Journal provided invaluable editorial advice that much improved the piece. And, as always, I am most grateful to Maron Deering, whose patience is unending.
ARTICLE CONTENTS

INTRODUCTION                                                 492
1. A NEW REGIME OF PRETRIAL PREVENTIVE RESTRAINT             502
    A. Origins of Pretrial Restraint for Dangerousness       502
    B. The Third Generation of Bail Reform                   507
II. DEFENDANTS AND NON-DEFENDANTS WHO ARE EQUALLY DANGEROUS  518
     A. No Clear Constitutional Distinction                  521
          1. Pretrial Detention Doctrine                     521
              a. Gerstein and Probable Cause                 521
              b. Salerno and Civil Commitment                523
            2. Pretrial Search Doctrine                      526
     B. No Clear Moral Distinction                           534
          1. "Moral Predicate" Theories                      534
          2. The Causal-Responsibility Argument              541
    C. No Clear Practical Distinction                        545
          1. Benefits and Costs of Preventive Restraint      545
          2. Diminished Costs                                548
              a. Bounded Restraint                           548
              b. Less Cost in Liberty                        549
              c. Notice and Opportunity                      55i
III. PARITY OF PREVENTIVE AUTHORITY                          557
    A. The Parity Principle in Action                        557
         1. What Risk Justifies Pure Preventive Restraint?   557
         2. Policy Implications                              560
         3. A World with Parity                              564
   B. The Parity Principle as Benchmark                      565
 CONCLUSION                                                  566
 APPENDIX                                                    568


INTRODUCTION

There is a nationwide movement underway to radically reconfigure the pretrial system. (1) The current system, which relies on money bail as the primary mechanism for pretrial release, results in the systematic detention of poor defendants. The scale of detention is vast. Approximately eleven million people are arrested each year; on any given day, around half a million of them sit in jail, awaiting trial. (2) Nearly all pretrial detainees have money bail set and would be released if they posted it. (3) Even at the lowest bail amounts, detention rates are high. (4) Reformers from across the political spectrum agree that a system that conditions liberty on wealth is both unjust and inefficient. At least ten states and forty counties have accordingly revised, or are in the process of revising, their pretrial law and policy--and in some cases their state constitutions. (5) If the pace of reform continues, the pretrial process across the nation will soon look very different.

The core reform goal is to untether pretrial detention from wealth and tie it directly to risk. To accomplish that objective, a growing number of jurisdictions are adopting actuarial risk-assessment tools to sort high-risk from low-risk defendants. (6) Until now, courts charged with setting bail and making pretrial custody decisions have, for the most part, assessed risk subjectively. Actuarial risk assessment is intended to improve the accuracy and consistency of these judgments.

It is hard to overstate the momentum behind this shift. A broad array of stakeholders, including national policy groups and large foundations, have advocated the adoption of pretrial risk assessment tools. The Laura and John Arnold Foundation, for instance, aims to ensure "that every judge in America will use a data-driven, objective risk assessment [for pretrial custody determinations] within the next five years." (7) It may succeed. Jurisdictions around the country are increasingly turning to risk assessment as the keystone of pretrial reform. (8)

The risk of core concern in today's pretrial policy debate is not, as it once was, the risk that defendants might abscond or tamper with witnesses. It is, instead, the risk that released defendants will commit other crimes. Reform opponents allege that defendants are too dangerous to be released into the community without significant restraint. The claim is not that they will skip court, harm witnesses, or otherwise obstruct prosecution. It is simply that they will commit new crimes unrelated to their pending charge. (9) In response, reformers assure stakeholders that actuarial risk assessment can reduce detention rates without compromising public safety.

There are many explanations for the reform movement's focus on danger. At a structural level, it reflects the broader turn toward incapacitation in criminal justice at the end of the twentieth century, and the risk-oriented, managerial approach to crime and punishment that Malcolm Feeley and Jonathan Simon dubbed "the new penology." (10) As a practical matter, flight risk may be less of a concern than it once was because it is hard to truly flee from justice in today's hyper-connected world. And in realpolitik terms, elected judges suffer much greater political costs when released defendants commit high-profile crimes than when they fail to show up for court.

Whatever the reasons, "[t]he goal of most criminal justice decisionmakers is to detain defendants who pose a risk to public safety--particularly those who appear likely to commit crimes of violence--and to release those who do not." (11) This is not to say that flight risk is unimportant, just that public safety has dominated the recent reform conversation. (12) In broad strokes, the central goal of the bail reform model has been to reduce pretrial detention by limiting it to the statistically dangerous.

This model holds great promise, but also raises an extremely difficult question: what probability that a person will commit unspecified future crime justifies detention, or even lesser restraints, like GPS monitoring? (13) For defendants who score in the top risk bracket on the Federal Pretrial Risk Assessment Instrument (PTRA), for example, the projected likelihood of rearrest for any type of crime in the pretrial period is ten percent. (14) Defendants classified as high risk by the Florida Pretrial Risk Assessment Instrument (FL PRAI) have a sixteen percent chance of rearrest in a six-month span. (15) And those classified as high risk for violence by the Public Safety Assessment (PSA), the most widely used tool in state systems, have about an eight percent chance of rearrest on a violent charge within six months. (16) Are these probabilities sufficient to justify detention? If not, what probability of future arrest is enough?

The question has received markedly little attention from modern reformers. A generation ago, pretrial restraint to prevent non-case-related future crime--what I will call, for simplicity, "preventive restraint"--was a matter of intense controversy. Critics argued that no probability of future crime was sufficient to authorize preventive detention. (17) Today's bail reform movement, by contrast, has assumed the legitimacy of pretrial preventive restraint and advocates preventive detention as a basic component of a model pretrial system. Advocacy groups like the American Civil Liberties Union (ACLU) have sporadically voiced concerns but have nonetheless signed on to the reform agenda. (18) Among academics, the turn to actuarial risk assessment has engendered both excitement and apprehension, but criticism has centered on its potential to exacerbate race and class inequalities. (19) There has been essentially no public debate about what degree of risk should be deemed sufficient to justify detention or other forms of restraint.

Recent events may soon bring that question to the fore. On January 1, 2017, New Jersey's comprehensive bail reform took effect, including a preventive detention regime that required an amendment to the state constitution. As cases move through the new system, New Jersey's courts are beginning to grapple with what quantum of risk is sufficient to justify detention. (20) Other states pursuing reform are not far behind. (21)

The adoption of risk assessment will require stakeholders to consider what degree of risk justifies restraint, moreover, because the new statistical methodology makes the question unavoidable in a way that it was not before. Many of the scholars who debated preventive detention a generation ago argued that useful prediction was impossible. (22) Laurence Tribe diagnosed an early preventive detention proposal as betraying "the inability to predict with even the slightest confidence" which defendants would commit future crime because, for lack of a better methodology, it relied on broad offense and criminal history categories as proxies for dangerousness. (23) Today's actuarial tools are far from perfect, but they are rapidly improving in sophistication and predictive power. By making it possible to formulate more precise legal standards for dangerousness, they also make it necessary to do so: the design of every risk assessment tool requires a decision about the statistical "cut point" at which a person will be deemed high risk, and detention recommended. (24) New Jersey's new regime authorizes detention on the basis of this recommendation alone, (25) which is to say, on the basis of a certain probability of rearrest in a six-month period. (26) In these circumstances, the choice of a statistical risk threshold may be implicit or explicit, but it cannot be avoided.

One way to start thinking about what level of risk justifies restraint is to ask a related question: is the answer different for defendants than for people not accused of any crime? A thought experiment will clarify this approach. Imagine that State Z implements actuarial risk assessment at the Department of Motor Vehicles (DMV). Like the newest pretrial risk assessment tools, the DMV tool requires no interview. It simply draws on administrative data, primarily criminal history records, to identify people who have at least a sixteen percent chance of arrest in the next six months. State Z proposes to funnel these DMV visitors into short-term detention. Presumably, State Z's proposal would face serious opposition. Detention of DMV visitors solely on the basis of this statistical risk would likely violate commonly held moral commitments and constitutional norms. (27)

Now consider whether there is any reason why State Z would nonetheless be justified in detaining defendants who pose the same degree of risk. (28) If so, it cannot be because that risk alone justifies detention, since it does not justify jailing DMV visitors. The question is whether there is some additional justification for the restraint of equally dangerous defendants.

The Supreme Court's lone decision on pretrial preventive restraint, United States v. Salerno, (29) does not answer this question. Salerno addressed a facial constitutional challenge to the federal preventive detention regime implemented by the Bail Reform Act of 1984. The Court rejected the challenge, holding that no constitutional provision categorically prohibits pretrial detention on the basis of dangerousness alone. Such detention, it held, can pass constitutional muster under some circumstances. But the Salerno Court did not specify what degree of risk is constitutionally sufficient to justify detention. And it said nothing about whether the answer is different for defendants than for others.

Nor has past scholarship confronted this question head on. There was a dynamic debate about the constitutionality of pretrial preventive detention in the 1970s and 1980s, when Congress enacted the first preventive detention statutes. (30) But that debate, like Salerno, centered on the question of whether the Constitution categorically prohibits the practice. A few scholars did raise the question of whether defendants should be uniquely subject to preventive interference. Professor Tribe asserted that they should not: "If two men appear equally likely to commit a violent crime, it is arbitrary to imprison the man who is about to be tried for a past offense while imposing no restraint on the man who is not facing trial." (31) A handful of other scholars have made similar observations. (32) None, though, has given the question more than passing attention.

It is now widely assumed that the state does indeed have special crime-prevention authority in the pretrial realm. Many believe that, as a general matter, the state may not restrain people who are responsible agents solely to prevent them from committing speculative future harm. (33) But defendants are presumed to be different. In both legal and moral terms, they are believed to be uniquely subject to preventive interference.

This Article challenges that broadly held view. It argues that, for purposes of preventive restraint, there is no clear, relevant distinction between defendants and non-defendants who are equally dangerous. (Once again, I use the term "preventive restraint" to mean pretrial restraint to prevent non-case-related future crime. It is synonymous with "restraint for general dangerousness," and distinct from restraint to prevent harm to specific witnesses or other obstruction of justice.) In terms of positive law, and contrary to common belief, there is no constitutional text or doctrine that clearly grants the state more expansive preventive authority over defendants than non-defendants. Similarly, there is no clear normative basis for subjecting defendants to preventive restraint that we would not tolerate for equally dangerous people not accused of any crime. And the practical justifications proffered to support the special preventive restraint of defendants are, at best, incomplete.

For purposes of preventive restraint, then, a defendant and non-defendant who pose equal risk are identical in the only relevant sense, which is riskiness. In line with the bedrock principle that like cases should be treated alike, (34) this conclusion implies a normative principle that I term "parity of preventive authority" or "the parity principle." The parity principle holds that the state has no greater authority to preventively restrain a defendant than it does a non-defendant who poses an equal risk. If a sixteen percent chance of arrest in the next six months is insufficient to justify short-term detention of a non-accused person, it is likewise insufficient to justify pretrial detention.

The Article's objective is not, however, to prove the parity principle conclusively. The goal is simply to demonstrate that an assumed premise of pretrial reform is highly questionable. At the very least, the notion that defendants are uniquely liable to preventive interference demands much more thorough justification than legislatures, courts, or scholars have provided to date. Given the trajectory of pretrial reform, it is both an important and an opportune time to clarify the contours of the state's pretrial powers. This Article aspires to begin the conversation, not to end it.

More broadly, the Article aims to assess the extent to which bail policy is influenced by a perception that people entangled with the criminal justice system are inherently less deserving as legal subjects. Whether conscious or not, that perception can shape law and policy. This Article strives to promote disciplined thinking about whether there is good reason for defendants to be subject to relaxed standards for preventive restraint beyond an assumption that their liberty deserves less protection. (35)

Finally, to the extent that the parity principle is correct, it is important to note that it is not fatal to pretrial preventive detention. The state currently engages in many types of preventive detention outside of criminal proceedings, including civil commitment, commitment of "sexually violent predators," material witness detention, protective custody of substance abusers, immigration detention, quarantine, and the detention of suspected terrorists. (36) At least some of these practices reflect a societal judgment that dangerousness alone can justify detention, even of people who qualify as responsible agents. (37) Many legal commentators reject that judgment and aspire to change it, (38) although not all do. (39) If dangerousness alone can indeed justify detention at some threshold of risk, then the parity principle permits pretrial preventive detention at that threshold.

This Article does not purport to establish what degree of future-crime risk might authorize preventive restraint of non-defendants. But, drawing on current law, it suggests that the threshold cannot be less than a substantial risk of serious violent crime in a six-month span. (40) Pretrial preventive restraint should therefore be limited to defendants who present a risk at or above that threshold. And risk assessment tools should both measure and communicate the likelihood of rearrest specifically for violent crime.

The Article proceeds in three Parts. Part I provides a brief history of pretrial restraint for dangerousness and an introduction to pretrial risk assessment. Part II argues that there is no clear constitutional, moral, or practical basis for treating defendants as a special case for purposes of preventive restraint. Part III outlines the resulting parity principle and draws out its policy implications.

I. A NEW REGIME OF PRETRIAL PREVENTIVE RESTRAINT

A. Origins of Pretrial Restraint for Dangerousness

Until the 1960s, the stated function of the pretrial system was to ensure the appearance of the accused at trial. (41) This remains the central function that the institution of bail is designed to serve. (42) An accused person deposits some security with the court to guarantee his appearance; so long as he does in fact appear, the deposit is returned to him at the conclusion of the case.

But authorities on pretrial law and policy--including pretrial laws themselves--now universally identify a second purpose of the pretrial system: protecting the public from harm at the hands of dangerous defendants. (43) Nearly all U.S. jurisdictions authorize courts to impose pretrial conditions of release on the basis of dangerousness. (44) Some authorize full-scale preventive detention as well. (45) Commentators disagree as to when pretrial law first endorsed restraint for dangerousness. Capital defendants have been excluded from bail since colonial days, and there is some evidence that this exclusion was a public-safety measure. (46) But there is also evidence to the contrary (47) For noncapital defendants, though, U.S. pretrial law was at least purportedly centered on ensuring appearance until the 1960s, when the system underwent a profound shift.

The 1960s saw the first major wave of bail reform. (48) That movement, like the present one, was catalyzed by the realization that the cash bail system discriminated dramatically against the poor. (49) Reformers sought to limit money bail in favor of release without bail, also known as release on recognizance (ROR). Ultimately, reform efforts led to the liberalization of pretrial release policies nationwide. (50)

This first wave of reform also had unintended consequences, however. As both release and crime rates rose in the late 1960s, political pressure built for the development of new methods to contain pretrial crime. In the past, judges had routinely prevented defendants they viewed as dangerous from getting out of jail by setting unattainable bail amounts--a practice known as sub rosa preventive detention. (51) Reform made that method more difficult, and proposals to authorize more explicit forms of preventive restraint began to surface. (52)

Between 1968 and 1984, the pretrial system transformed itself again. During this "second generation" of bail reform, (53) thirty-four states amended their statutory laws to authorize detention without bail or restrictive conditions of release on the basis of dangerousness. (54) Many amended their constitutions along the same lines. (54) Congress passed two statutes authorizing preventive detention for noncapital defendants, first in the District of Columbia and then throughout the federal system via the Bail Reform Act of 1984. (56)

Initially, preventive detention of noncapital defendants was highly controversial. Critics argued that the Constitution prohibits detention of people who are responsible agents solely to prevent intentional future harm. (57) The case that tested that argument was United States v. Salerno. (58) The defendants in the underlying proceeding, Anthony Salerno and Vincent Cafaro, had been detained pursuant to the 1984 Bail Reform Act. (59) They argued that the detention regime, on its face, violated substantive due process, procedural due process, and the Eighth Amendment prohibition of "[e]xcessive bail." (60) The Second Circuit agreed, holding that substantive due process "prohibits the total deprivation of liberty simply as a means of preventing future crimes." (61)

The Supreme Court rejected this view. (62) Justice Rehnquist, writing for the majority, concluded that none of the constitutional provisions invoked by the defendants categorically prohibits preventive detention. (63) Such detention, the Court held, may be constitutionally permissible if the danger is sufficiently acute and the detention at issue is a sufficiently tailored means of addressing that danger. (64) The Court found that the federal detention regime did not fail these criteria as a facial matter. (65)

Salerno effectively ended the preventive detention debate. The Court had answered the core question with a resounding "no": the Constitution does not categorically prohibit preventive detention. Perhaps because the critics' defeat on this point was so total, Salerno was widely perceived as a robust endorsement of pretrial preventive detention and lesser forms of preventive restraint. (66)

In fact, Salerno's holding was relatively narrow. Although the Court held that neither the Due Process Clause nor the Eighth Amendment prohibits preventive detention entirely, it also recognized that each does limit the state's detention authority. The Salerno Court explained that pretrial detention will constitute punishment--and so violate substantive due process--if it is irrational or "excessive" in relation to its regulatory goal, or if it is inflicted with punitive intent. (67) Moreover, the Court implied that even non-punitive detention can violate substantive due process if not carefully tailored to its goal, (68) and might also violate the Excessive Bail Clause if it is excessive "in light of the perceived evil" it is designed to address. (69) Finally, detention regimes that lack the "extensive" procedural safeguards of the Bail Reform Act might violate procedural due process. (70) The Court explicitly left the door open for as-applied challenges pursuant to any of these provisions. (71) Acknowledging that pretrial liberty is the "general rule," (72) Salerno concluded by affirming that "[i]n our society liberty is the norm, and detention prior to trial or without trial is the carefully limited exception." (73)

The years since have belied this statement. Before 1984, around twenty-four percent of federal defendants were detained until trial. (74) By 2010 the number was at least sixty-four percent, (75) with seventy-six percent of federal defendants detained for some period of time. (76) As-applied constitutional challenges are infrequently brought and rarely succeed. Lower courts regularly find that federal pretrial detention is warranted without rigorous analysis of whether detention is an excessive response to the threat at issue. (77)

One reason that Salerno's limits have had so little traction is that they remain amorphous. Any "excessiveness" inquiry necessarily requires some accounting

for the risk at issue and the expected crime-prevention benefit the restraint achieves by reducing that risk. (78) The Salerno Court did not indicate with any precision what degree of risk it thought constitutionally sufficient to justify detention. It noted that the Bail Reform Act "operates only on individuals who have been arrested for a specific category of extremely serious offenses," whom Congress had found to be "far more likely to be responsible for dangerous acts in the community after arrest." (79) The Court also emphasized that under the Act, before detention can be imposed, the government must show that the defendant presents "a demonstrable danger to the community," and the trial court must find "that no conditions of release can reasonably assure the safety of the community or any person." (80) But these standards are extremely vague. What conditions of release provide "reasonable assurance" of safety, and under what circumstances? What probability of crime risk qualifies as "demonstrable danger" ? Salerno itself provides little guidance.

B. The Third Generation of Bail Reform

A third wave of bail reform is now underway. Despite the ambitions of the first two waves, most jurisdictions never fully implemented regimes of preventive detention or non-financial conditions of release. They have instead continued to rely on money bail and sub rosa detention as a crude mechanism for managing pretrial crime risk. (81) Indeed, since 1990, both pretrial detention rates and the use of money bail have risen steeply; it is likely that we now detain millions of people each year for their inability to post even small amounts of bail. (82) This fact has galvanized a third reform movement, driven in part by national policy organizations, that aspires to achieve lasting change by shifting the entire pretrial paradigm from a cash-based to a risk-based model. (83)

The new reform model requires jurisdictions to sort defendants through actuarial risk assessment. (84) The key advantage of actuarial assessment is that it is formulaic. Whereas clinical risk assessment relies on an expert's subjective judgment, actuarial assessment is mechanistic and statistical. It tells us the likelihood that person A will have outcome X, extrapolated statistically on the basis of the past outcomes of other individuals with similar traits. (85)

Currently, the dominant methodology in the pretrial context is the use of a simple checldist tool, or "risk assessment instrument." Statisticians develop such tools by analyzing aggregated pretrial data to identify the traits of defendants that correlate most closely with the outcome of concern. (86) Those traits are deemed "risk factors." (87) The developers then create a checklist that assigns each risk factor a number of points corresponding to how closely it is correlated with the bad outcome in the group data. (88) Having been arrested before age eighteen might be three points, for example, and being unemployed might be two. To calculate an individual's risk score, one simply checks off the risk factors that apply and adds up the points. While some tools are only available as paper checklists and require manual administration, others are offered as automated or partially automated software. (89)

Existing pretrial tools assess the risk of two outcomes: failure to appear (FTA) and rearrest. Most of the existing instruments produce a single score that represents the risk of either one occurring. (90) As Lauryn Gouldin has explained, this aggregation is a problem. (91) The two risks are different in kind, are best predicted by different variables, and are most effectively managed in different ways. (92) It is to be hoped and expected that the next generation of tools will measure each outcome separately.

Around forty jurisdictions are now using risk assessment tools, and the number is growing fast. (93) This momentum is due in large part to the Laura and John Arnold Foundation (LJAF), which developed an instrument designed to function in any jurisdiction (the PSA) and has made it freely available. (94) The PSA improves on other tools by separately assessing the risks of nonappearance, rearrest, and rearrest on a violent charge. (95)

The number of risk factors included in current tools varies from seven to fifteen, and the factors themselves vary across tools. Table 1 provides a basic overview. (96) In addition to these scoring instruments, there is a more advanced methodology on the horizon: forecasting through machine-learning technologies. (97)

Depending on a defendant's numerical risk score, the tools classify that person as low, moderate, or high risk. (98) These classifications do not purport to predict individual outcomes with anything approaching certainty. A classification as high risk does not assert that Person A will fail to appear or be rearrested unless restrained. All it purports to do is rank Person A relative to the rest of the population upon which the instrument was developed. What a "high risk" classification actually asserts is that Person A belongs to a group of people with shared traits that will have higher levels of nonappearance and rearrest than those outside the group." (99)

Nonetheless, each classification does correspond to some statistical likelihood of nonappearance or rearrest. A risk assessment tool is only valid if the group of defendants it classifies as high risk has the highest rate of bad outcomes. To "validate" a tool, therefore, researchers use it to classify members of a sample set where outcomes are known, then document the rate of the relevant outcome for each risk class. (100) Validation studies thus illustrate the statistical likelihood of a given outcome for defendants grouped in each risk class.

Validation studies of existing pretrial tools have documented the following rates of rearrest among defendants classified in the highest-risk group. (101)

Persons classified as "high risk" by the Federal PTRA have a ten percent chance of rearrest in the pretrial period, and so forth.

It is important to note that the rates depend both on the outcome measure and the time over which it is measured. The broader the measure and the longer the measurement period, the higher the rate will be. Forty-two percent of those deemed highest risk by the CPAT, for instance, have some land of "new criminal filing" in the pretrial period, but that includes traffic and municipal case filings. (113) Unfortunately, many of the validation reports that measure "new arrests" do not precisely define that term. (114)

The most recent reform model envisions actuarial risk assessment as the basis for pretrial release and custody decisions. (115) Money bail is not to be used to mitigate danger. (116) Rather, low-risk defendants should be released pending trial; moderate-risk defendants should also be released, though potentially with non-financial conditions tailored to risk level; and high-risk defendants should be detained. (117) The reform model therefore requires that judges have authority to order pretrial preventive detention. In general, reformers advocate leaving judges some discretion to override risk-based recommendations, though they disagree over how much discretion to allow. (118)

Implementing this vision will require significant changes to state law. Twenty-three states still guarantee a broad constitutional right to bail and will have to amend their constitutions to authorize preventive detention without bail. (119) As for statutory law, every state already authorizes judges to order conditions of release to protect "public safety," but the current standards are varied, ambiguous, and often irrational. (120)

Notwithstanding these hurdles, the reform vision is already becoming a reality, and change is happening fast. At least twenty states and countless municipalities have undertaken significant pretrial reform efforts in the last few years. (121) Others are poised to follow. (122)

The reform movement holds great promise, but it also crystallizes fundamental questions about pretrial policy: What degree of statistical future-crime risk justifies restraint? And is the answer different for defendants than for anyone else? Most criminal-law scholars hold that only serious risk--if any at all--could justify the purely preventive restraint of a rational member of the general population. (123) The question this Article aims to address is whether a pending criminal charge alters the analysis.

II. DEFENDANTS AND NON-DEFENDANTS WHO ARE EQUALLY DANGEROUS

There is an understandable basis for assuming that, when it comes to state-imposed deprivations of liberty, defendants are a special case. Few contest the notion that the state has the right to arrest someone if there is probable cause to believe she has committed a crime. (124) Likewise, few deny that the state has the right to detain her or otherwise limit her liberty if necessary to prevent her from fleeing or interfering with witnesses. Because it seems abundantly clear that the state may restrict defendants' liberty in special ways for these reasons, it is widely assumed that the state may also restrain defendants in special ways to prevent future crime.

Yet the fact that the state can impose special restraints for some purposes (to ensure appearance and protect witnesses) does not entail the conclusion that it can do so for others (to prevent non-case-related crime). Whether the state does have authority to impose unique restraints on defendants to prevent non-case-related crime depends on the reason why it is authorized to restrain defendants to prevent flight and harm to witnesses. One possibility is that the state may restrict defendants' liberty to prevent those obstructive harms because defendants simply have a lesser right to liberty than other people. If that is so, then the purpose of the restraint is immaterial. The state's special authority derives from defendants' inferior rights status, so whatever the purpose of restraint, the standards for defendants can be comparatively lax.

But the justification for the state's authority to restrain defendants to prevent obstructive harms may not be that defendants have a reduced right to liberty. Rather, it may simply be that the state's interest in making sure the legal process functions--that defendants show up, that witnesses are safe--sometimes trumps the individual right to liberty. In other words, it might be the case that the powerful state interest in administering a functional criminal justice system would justify restricting anyone's liberty if necessary. It just happens that defendants are the people most likely to derail a prosecution. It makes sense, on this account, that defendants are subject to unique restraints--those necessary to make the system run. Note, though, that reluctant witnesses are subject to similar restraints on the same grounds. (125)

If the reason that the state can impose some special restraints on defendants is its interest in operating a functional criminal justice system, then the legitimacy of those restraints does not mean that the state may also subject defendants to special restraints for other purposes. The justification for the special treatment is a particular state need that happens to uniquely affect defendants. The justification is not that defendants have a reduced right to liberty.

This Part argues for the latter understanding of state authority to restrict individual liberty in the pretrial realm. On close analysis, there is no constitutional doctrine holding that defendants have an inherently reduced right to liberty, or authorizing unique restraint of defendants to prevent non-case-related harms. Nor, for purposes of preventive restraint, is there any moral or practical basis for the state to value defendants' liberty less than other people's. In sum, for purposes of preventive restraint, there is no constitutional, moral, or practical distinction between a defendant and non-defendant who are equally dangerous.

Some readers may object at the outset that, on average, defendants are more dangerous than non-defendants, or that a particular defendant might pose a greater threat than any person the state can identify in the population at large. These points are taken up later, (126) but neither undermines the Article's argument. For clarity of exposition, though, the argument focuses on defendants labeled high risk on the basis of actuarial assessment. As noted above, a high-risk classification currently denotes anywhere between a ten percent and a forty-two percent chance of a new arrest, or an approximately eight percent chance of new arrest for violent crime, over a six-month span. (127) With big-data predictive technologies we could identify plenty of non-defendants who pose an equivalent statistical risk, but presumably these risk levels alone would not justify restraint of a non-defendant. Contrary to popular belief, there is no reason the analysis should differ for defendants.

At risk of repetition, a restatement of terminology may be useful here. The argument that follows concerns pretrial burdens on defendants' liberty or privacy imposed to prevent non-case-related future crime, which I call, collectively, "preventive restraint." Preventive restraint includes "preventive detention," by which I mean detention imposed to prevent non-case-related crime (or detention on the basis of "general dangerousness"). Preventive restraint also includes conditions of release, like supervision or GPS monitoring, imposed to prevent non-case-related crime. For purposes of this Article, "preventive restraint" does not include detention or conditions of release imposed to prevent case-related--or obstructive--harms (flight, harm to witnesses, or other efforts to thwart prosecution).

A. No Clear Constitutional Distinction

To begin with the positive law: One might imagine that constitutional doctrine grants the state more expansive authority to preventively restrain defendants than members of the public at large. (128) Because preventive restraint encompasses both detention and noncustodial conditions of release like GPS monitoring and drug-testing, the Supreme Court's jurisprudence on both pretrial detention and pretrial searches is relevant. (129) A careful reading, however, demonstrates that neither line of cases plainly authorizes special standards for preventive restraint of defendants.

1. Pretrial Detention Doctrine

a. Gerstein and Probable Cause

The most obvious distinction between defendants and non-defendants who are equally dangerous is that the defendants have been charged with a crime. It is a common assumption that a criminal charge supported by probable cause authorizes the state to hold a person until trial, such that the state grants pretrial release at its discretion. If that were the case, constitutional doctrine would indeed grant the state more expansive preventive authority over defendants than equally dangerous non-defendants. The state could detain any defendant on the basis of probable cause alone, regardless of risk.

But that is not the state of the law. The Supreme Court has never held that probable cause alone is a constitutionally sufficient ground for extended pretrial restraint. In Gerstein v. Pugh, the Court held that a judicial determination of probable cause is a necessary condition for extended pretrial restraints on liberty, not that it is a sufficient one. (130) The Court recently reaffirmed the probable-cause requirement in Manuel v. City of Joliet. (131) That decision also asserts only that probable cause is a necessary prerequisite for severe pretrial restraint; the Court did not say that probable cause alone is sufficient to justify such restraint. And Salerno clarified that pretrial liberty is the norm. (132) Serious incursions on pretrial liberty require justification beyond probable cause. (133)

Yet Gerstein and Manuel might still appear to suggest that the state has broader preventive authority over defendants than equally dangerous non-defendants. The fact that probable cause is a necessary condition of pretrial pre-ventive restraint might suggest that it is a necessary condition of any preventive restraint. (134) In other words, one conceivable reading of Gerstein and Manuel is that dangerousness can never justify extended restraint without an arrest supported by probable cause.

But this reading is not plausible. Most simply, it would appear to prohibit other preventive detention schemes the Court has upheld, including the civil commitment of sexually violent predators. (135) The commitment scheme the Court upheld in Kansas v. Hendricks did require a charge or conviction for a sexually violent crime at some point in the past, as well as a finding of probable cause to believe the putative detainee was a sexually violent predator likely to reoffend in the future, (136) but neither of these requirements would satisfy the Gerstein pretrial rule. Gerstein requires probable cause for a pending criminal charge--not for a charge adjudicated in the past, nor for a finding of dangerousness.

The more plausible reading of Gerstein and Manuel is that the Court's attention in those cases was focused on the pretrial realm. The opinions do not contemplate detention regimes falling outside that space. Even within the pretrial context, neither Gerstein nor Manuel specifically discusses restraint for general dangerousness. (138) They thus do not preclude detention for dangerousness without a pending criminal charge. And they cannot answer the question of whether the constitutional criteria for preventive restraint are more relaxed for defendants than for others.

b. Salerno and Civil Commitment

United States v. Salerno might also seem to authorize special standards for preventive restraint of defendants. Salerno is the only Supreme Court case that squarely addresses pretrial preventive restraint. (139) As discussed above, Salerno rejected a facial constitutional challenge to the federal preventive detention regime. (140) It held that neither due process nor the Excessive Bail Clause categorically prohibits pretrial detention for general dangerousness. (141) One might infer that, in authorizing preventive detention of defendants, the Court implicitly held that defendants are different than non-defendants, even if they pose the same degree of crime risk.

Yet nothing in Salerno compels the conclusion that the state has broader authority to preventively detain defendants than non-defendants who are equally dangerous. The most basic reason is that the Court did not consider this comparison. The opinion does not address it.

The Salerno Court's reasoning, moreover, is not specific to the pretrial sphere. The Court upheld the challenged detention regime on the basis of pure instrumentalist balancing. It reasoned that an individual's right to liberty may "be subordinated to the greater needs of society" when "the government's interest is sufficiently weighty," (142) and that "the Government's regulatory interest in community safety can, in appropriate circumstances, outweigh an individual's liberty interest." (143) This form of logic is as applicable to non-defendants as it is to defendants. (144)

The Court itself described its interest-balancing analysis as applying far beyond the pretrial system. It cited "the well-established authority of the government, in special circumstances, to restrain individuals' liberty prior to or even without criminal trial and conviction." (145) As examples of the latter situation, the Court cited the government's power to detain enemy aliens, "individuals whom the government believes to be dangerous" in times of insurrection, "potentially dangerous resident aliens pending deportation proceedings," and "mentally unstable individuals who present a danger to the public." (146) None of these examples involves pending criminal charges. And the Court presented these as a non-exhaustive list of situations in which the state's interest in public safety could simply outweigh an individual's right to liberty. This reasoning suggests that the state could detain anyone to prevent future crime if the projected crime were sufficiently serious and likely to occur. At the very least, Salerno does not foreclose that possibility.

To bring the point full circle: If the state's preventive authority is simply a matter of balancing the communal interest in safety against the private interest in liberty, the degree of risk that justifies restraint should be constant for defendants and non-defendants--unless defendants have a reduced liberty interest. Nothing in Salerno suggests that they do.

It might nonetheless still seem that Salerno effectively grants unique preventive authority in the pretrial arena because the Court has elsewhere prohibited detention for dangerousness alone. In its civil commitment cases, the Court has limited detention to people who, because of a mental disorder, "are unable to control their dangerousness." (147) The absence of this lack-of-control criterion in Salerno, the argument goes, means that the rules for preventive restraint are laxer for defendants than for others.

On closer analysis, however, it is not clear that the Court's civil commitment cases actually do prohibit detention for dangerousness alone. To begin with, the lack-of-control criterion is a dubious one. (148) More concretely, the Court's rejection of confinement for dangerousness in Foucha v. Louisiana did not rest on a lack-of-control criterion at all. (149) Rather, the Court held that the continued confinement of Mr. Foucha (an insanity acquittee who was no longer mentally disordered) violated substantive due process because the detention regime was not adequately tailored to its public-safety goals. The majority reached this conclusion by comparing the Louisiana regime against the pretrial regime upheld in Salerno. (150) Whereas detention pursuant to the Salerno regime was "strictly limited in duration," (151) Mr. Foucha was subject to "indefinite" commitment. (152) And whereas the Salerno regime included both substantive and procedural mechanisms to limit detention to the acutely dangerous, the Louisiana regime did not. (153) Justice O'Connor, who provided the crucial fifth vote in Foucha, specifically noted that the Court's holding did not foreclose more careful efforts at detention for dangerousness. (154)

Foucha thus strongly supports the conclusion that the reason the Court authorized detention for dangerousness in Salerno was simply that the detention regime at issue was "carefully limited." (155) It was not because defendants are a special case.

2. Pretrial Search Doctrine

The Supreme Court's Fourth Amendment jurisprudence governing pretrial searches might also seem to authorize special infringements on defendants' liberty and privacy. The Court has held that during and subsequent to a lawful arrest the state can invade a person's privacy in ways that it otherwise could not. Without a warrant, police can search an arrestee's person (156) and often her car as well, (157) conduct a protective sweep of a home where an arrest is made, (158) compel an arrestee to answer questions about her identity, (159) inventory her belongings, (160) extract her DNA for analysis, (161) and subject her to other radical deprivations of personal privacy, including strip searches, if she is booked into jail. (162) This case law might seem to suggest that arrestees and defendants have uniformly reduced privacy rights, and therefore a lesser right against those preventive restraints that are likely to be classified as searches (like GPS monitoring). Furthermore, if defendants have uniformly reduced privacy rights, they might also have uniformly reduced liberty interests, and therefore a lesser right against preventive restraints likely to be classified as seizures.

Two recent cases appear to support this reduced-privacy-rights view. In both Maryland v. King and Riley v. California, the Court stated explicitly that arrestees have "diminished expectations of privacy." (163) In King, which upheld the warrantless collection of DNA from arrestees, the Court relied in part on these diminished expectations to waive the normal Fourth Amendment warrant requirement and deploy the freewheeling "reasonableness" analysis that it has applied to searches of probationers and parolees. (164)

The King Court's reliance on arrestees' "diminished expectations of privacy" might imply that pretrial defendants share the status of children, public-school students, government employees, probationers, parolees, and detainees--groups that the Court has held to have broadly diminished expectations of privacy. (165) For these groups, Fourth Amendment analysis has a different starting point. The government need not meet the standards it normally must meet to justify a warrantless search, because these groups cannot assert the same right to privacy in the first place. (166) King suggests that the status of being a defendant affects Fourth Amendment analysis in much the same way. (167)

There are good reasons, however, to reject that reading of King. First, the rationales justifying the limited privacy rights of other groups do not apply to pretrial defendants. And second, while the Court has yet to directly confront the question of whether a pending charge affects the structure of Fourth Amendment analysis, (168) its jurisprudence already suggests a negative answer.

The Court's reasons for asserting a priori limitations on the privacy rights of certain groups are not always clear, but one core idea is that the state bears custodial responsibility for members of these groups. (169) This custodial rationale applies most cleanly to people in government detention centers, psychiatric hospitals, schools, government workplaces, and prisons. (170) It might also apply, albeit less plainly, to children in delinquency proceedings, as well as to parolees and probationers. (171) With respect to the latter two groups, the Court has further intimated that people still serving a sentence "on a continuum of possible punishments" have limited rights because a reduction in rights status is part of the punishment deserved. (172)

Neither the custodial rationale nor the punishment rationale applies to pretrial defendants--at least not for purposes of determining when preventive restraint is justified. The punishment rationale is obviously inapt because pretrial defendants have not been convicted. Although the custody rationale may kick in after pretrial restraints have been lawfully imposed, it is inapposite when the question is whether restraint is justified in the first place.

The Court's pretrial search jurisprudence itself, moreover, contradicts the notion that defendants have diminished rights against state intrusion simply because they are facing criminal charges. Rather, a limiting principle unifies the doctrine: only the demands of the adjudicative process can justify special infringements of defendants' privacy. The Court itself has not been clear on this point, but the principle explains the outcomes of its cases. The Court has upheld those warrantless pretrial intrusions that it deems essential to the safe and effective administration of criminal proceedings. Such intrusions include those necessary to ensure safety on the scene of arrest, (173) preserve evidence of the crime from imminent destruction, (174) ensure accountability for impounded possessions, (175) safely manage jail facilities, (176) and accurately identify defendants. (177) By contrast, the Court has rejected warrantless pretrial intrusions not necessary to ensure the safety and integrity of criminal proceedings. (178)

On this reading of the Court's pretrial search jurisprudence, defendants retain the same right to privacy as anyone else. But, because defendants are the subjects of criminal prosecution, their privacy rights must sometimes yield to the needs of the adjudicative process. For that reason, the outcome of Fourth Amendment analysis might differ, in this limited respect, for defendants and non-defendants. But the justification for any special limitations on defendants' privacy is that the state's heightened needs during the pretrial process demand such limitations, not that defendants have lesser a priori rights.

The familiar conceptual distinction between rights infringements and violations provides a helpful framework. (179) A right is infringed "when the interests that give rise to it are harmed." (180) If the infringement is justified, however, there is no rights violation. "A violation is a wrongful or unjustified infringement." (181) Using that vocabulary, any invasion of a defendant's privacy infringes the defendant's Fourth Amendment rights to exactly the same extent as it would infringe a non-defendant's. But when the invasion is justified by the demands of the adjudicative process, it does not violate a defendant's rights. It would violate the non-defendant's rights, because it would lack justification. Certain rights infringements that would violate non-defendants' rights are thus permissible for defendants. But this is not because defendants have lesser Fourth Amendment rights than other people to begin with. Rather, it is because the demands of the adjudicative process justify some special infringements of the Fourth Amendment rights we all share.

Despite their talk of diminished expectations, both King and Riley ultimately support the principle that only the demands of the adjudicative process justify special infringements of defendants' privacy. In order to uphold the challenged DNA collection regime, the King majority went to great lengths to cast it as necessary to guarantee the accurate identification of arrestees. (182) Conversely, in Riley, the Court held that the Fourth Amendment prohibits the warrantless search of arrestees' cell phones that are not necessary to protect the safety and integrity of ongoing proceedings. (183)

The majority opinion in King rested on the premise that the warrantless DNA collection at issue was reasonable, and thus constitutional, because of the state's need to correctly identify arrestees. While Justice Kennedy spent four paragraphs discussing arrestees' privacy expectations and the minimal intrusiveness of a buccal swab, he spent twenty-one arguing that the Maryland DNA Collection Act served "the need for law enforcement officers in a safe and accurate way to process and identify the persons and possessions they must take into custody." (184) Such identification is important, he explained, not only for basic administrative purposes, but also to determine whether the accused poses special risks to custodial staff (if he is detained) and whether he is likely to abscond or commit pretrial crime (if released). (185) "In sum," Justice Kennedy concluded, "there can be little reason to question" the government's legitimate interest in identification, and DNA collection is "no more than an extension of methods of identification long used in dealing with persons under arrest." (186)

This premise was patently false, as Justice Scalia argued in dissent, (187) and as many commentators have noted since. (188) Certainly the state has a strong interest in identifying arrestees, but DNA collection does nothing to advance that interest. The obvious purpose of Maryland's collection regime was, instead, to compile evidence that might prove useful in solving unrelated crimes. (189)

But the Court's contortions are revealing. Why would it go to such lengths to construct its decision around the state's need for accurate identification? The answer leads back to the unifying principle. If the Court had acknowledged the real purpose of the DNA collection regime, it would have run up against the rule that "suspicionless searches are never allowed if their principal end is ordinary crime-solving." (190) It would then have had three alternatives: (1) roll back the rule against suspicionless investigatory searches, (2) hold that the rule exempts defendants, or (3) hold the DNA searches unconstitutional. The Court's tortured efforts to avoid this choice suggest that it was unwilling to take any of these three paths.

In other words, the King Court's identification rationale may "tax[] the credulity of the credulous," (191) but it reflects the Court's refusal to hold that an arrestee, simply by virtue of having been arrested, becomes subject to warrantless searches unrelated to his prosecution. As Justice Scalia asked and answered: "Why not just come out and say that any suspicionless search of an arrestee is allowed if it will be useful to solve crimes? The Court does not say that because most Members of the Court do not believe it." (192) The Court relied on the identification rationale because some procedural-needs rationale was necessary to uphold the search. (193)

Riley, on the other hand, was a case without a procedural-needs fig leaf. In each of two cases consolidated before the Court, law enforcement officers had searched an arrestee's cell phone without a warrant. (194) The Court held that these

warrantless searches violated the Fourth Amendment. As in King, the Court invoked arrestees' allegedly diminished expectations of privacy. But the holding centrally relied on the Court's conclusion that the two risks that justify warrantless searches incident to arrest--"harm to officers and destruction of evidence"--do not necessitate immediate searches of cell phones. (193) Because such searches are not justified by the needs of the adjudicative process, they are impermissible.

King and Riley thus affirm the unifying principle that the constitutionality of a warrantless pretrial search turns on whether it is necessary to ensure the integrity of the ongoing proceeding. On this interpretation, the Court's "diminished expectations of privacy" language is a distraction. An arrestee's expectations of privacy "necessarily are of a diminished scope," because of the special needs of the adjudicative process. (196) If you are arrested, you expect officers to disarm you, booking personnel to extract identifying information, and jail personnel to inventory your belongings--because these intrusions are necessary for the system to work. But it is that necessity that justifies the intrusions, rather than your subjective expectations. (197) The Court would clarify matters by recognizing as much. (198)

If this interpretation is correct, the Court's pretrial search doctrine does provide special authority for pretrial restraint to prevent flight, harm to witnesses, or other obstructions of justice--but it does not provide special authority for restraint to prevent unrelated future crime. In some cases, monitoring or restraint may be required to prevent obstructive harms. But lawful arrest does not itself authorize invasions of privacy for other purposes. (199) And preventing unrelated crime is not a special need of the adjudicative process. (200) On the most coherent interpretation, then, the Supreme Court's pretrial search doctrine does not support the conclusion that the Fourth Amendment standard for preventive restraint is different for defendants than for anyone else. (201)

B. No Clear Moral Distinction

Those who grant that there is no doctrinal basis for permitting greater preventive restraint of defendants than of equally dangerous non-defendants might still assert a moral basis. (202) There is, after all, an obvious difference between the two groups: defendants have been charged with a crime. A few scholars have suggested that a pending charge supported by probable cause constitutes an adequate "moral predicate" for special restraint. (203) One might also argue that the state is in a different moral position vis-a-vis a defendant than others who are equally dangerous, because it has already taken the defendant into custody. Releasing a person who subsequently commits a crime seems like an act, whereas failing to detain a person who subsequently commits a crime seems like an omission. The following subsections evaluate each of these two potential moral distinctions.

1. "Moral Predicate" Theories

From the outset, it is important to distinguish between the evidentiary and moral significance of a criminal charge. A charge may be powerful evidence of risk. It may be the case that, on average, people charged with recent criminal acts are more dangerous than people who are not, or that the people who pose the most acute, identifiable threat are those facing certain lands of charges. Risk assessment purports to take this kind of evidence into account. A high-risk defendant in the PSA study had an 8.6% chance of rearrest on a violent charge, given the fact and nature of his current charge. (204) Beyond what statistical assessment can capture, the alleged facts in a particular case might suggest heightened risk relative to statistically similar peers. To be clear, nothing in the Article contests the state's authority to consider the evidentiary significance of pending allegations for purposes of a risk analysis. (205) But the fact that a criminal charge may be powerful evidence of risk does not help to answer this Article's core question, which is whether the threshold of risk at which restraint is justified is different for defendants than for others. As mere evidence of risk, a criminal charge provides no basis for distinguishing between people who are equally risky. (206)

A pending charge might also have moral significance. The judicial determination of probable cause denotes that there is "probable cause to believe the suspect has committed a crime." (207) And crimes, for the most part, entail volitional wrongful acts. (208) So a charge supported by probable cause reflects some probability that the accused has committed a wrongful act for which she bears responsibility. This probability of guilt carries moral weight. Especially if the charge is serious, it calls the accused's moral standing into question.

The fact that a criminal charge might have some moral significance does not, however, automatically justify special deprivations of liberty imposed by the state. Proving that proposition would require additional work. The argument could take several forms.

First, one could argue that at least some defendants deserve less protection against preventive interference than non-accused people by virtue of their probable guilt. Following this line of argument, some might contend that it is intrinsically good--or at least not intrinsically bad--for defendants to suffer, given the probability that they have committed blameworthy criminal acts. (209) A related argument would be that defendants forfeit their right against preventive interference. (210) Albert Alschuler suggests something along these lines. He proposes that preventive detention may be justified when a court can conclude--"based on imperfect evidence that is also the best available--that a defendant has abused his freedom by committing a serious crime." (211) Alschuler's formulation of this principle implies a moral judgment, but this land of forfeiture could also operate as a strict liability regime. (212)

Another argument, adapted from H.L.A. Hart's vision of the criminal law, might be that our legal system simply makes preventive restraint one consequence of a criminal charge. (213) Hart argued that, in order to maximize both individual choice and public welfare, the law should prohibit the state from interfering with individual liberty unless a person has chosen to break the law. In turn, the state can attach consequences to that choice. As long as people have adequate notice of the consequences of law-breaking and nonetheless choose to incur them, those consequences can fairly be enforced. (214) Under this fair-consequence theory, we might conceptualize pretrial preventive restraint as a consequence voluntarily incurred.

A final variant of the moral-predicate argument, suggested by the work of R.A. Duff (but notably rejected by Duff himself), is that some defendants might owe a civic "duty of reassurance" that justifies restraint to prevent non-case-related crime. (215) On Duff's view, each member of a polity owes the others a duty of "civic trust," and therefore also owes them reassurance if her own civic trustworthiness is called into question. (216) Duff reasons that in some cases the law can justifiably command such reassurance, and that this logic might justify the restraint of defendants who have demonstrated a propensity to abscond or to obstruct justice. (217) The same argument might be deployed to justify pretrial restraint of dangerous defendants. If the duty of reassurance is solely a function of apparent risk, then a defendant owes no greater duty than an equally dangerous non-defendant. But if the duty arises from a person's culpable choice to engage in behavior that demonstrates risk, then a defendant might have a greater duty of reassurance than a non-defendant who poses an equal risk but is not accused of recent criminal conduct. (218)

The central problem with each of these moral-predicate theories is that they justify pretrial deprivations of liberty by pointing to a defendant's responsibility for the charged offense. But to invoke a defendant's guilt as justification for pretrial restraint threatens fundamental due process values, which tend to run under the heading of the "presumption of innocence." Defendants, after all, have only been accused. Many are not guilty. Fewer than seventy percent of felony arrests nationwide lead to conviction. (219) And the protection of accused people against false condemnation and punishment is a core commitment of the criminal justice system. Although this due process objection is not unanswerable, it does represent a formidable challenge for moral-predicate theories. It warrants careful elaboration.

First off--and contrary to intuition--the presumption of innocence itself does not precisely capture the due process objection. Rhetorically, the presumption looms large. The Supreme Court, for instance, has mused that the presumption of innocence "is the undoubted law, axiomatic and elementary, and its enforcement lies at the foundation of the administration of our criminal law." (220) The Court has also clarified, however, that the presumption is only "a doctrine that allocates the burden of proof in criminal trials." (221) It has no specific application to pretrial defendants. (222) Some scholars lament this characterization as unduly narrowing the presumption's ambit. (223) On the other hand, as Richard Lippke has explained, it is unclear exactly how the presumption of innocence could or should operate in the pretrial arena. (224)

The due process objection is more precisely stated in terms of the values the presumption of innocence reflects. As the Supreme Court explained in In re Winship, (225) the presumption reflects a commitment to protect individuals against official condemnation unless their guilt has been firmly established. (226) The stakes are high. An official judgment of guilt is often life-altering, carrying the possibility of confinement and the certainty of stigma. (227) "Accordingly, a society that values the good name and freedom of every individual should not condemn a man for commission of a crime when there is reasonable doubt about his guilt." (228) Due process demands this protection not just for criminal conviction, but also for adjudications of guilt in juvenile delinquency proceedings that trigger (purportedly) non-punitive consequences only. (229)

The central objection to the various moral-predicate theories, then, is that due process prohibits depriving a person of liberty on the basis of an official judgment of guilt except on proof beyond a reasonable doubt. More specifically, one might argue that any deprivation imposed by virtue of guilt should be classified as punishment, (230) and is therefore flatly impermissible during the pretrial period. (231) Alternately, one might argue that due process prohibits pretrial deprivations by virtue of guilt even if they do not constitute punishment, because they involve official condemnation and the loss of liberty. After all, probable cause does not provide sufficient basis for a judgment of guilt; probable cause "means less than evidence which would justify condemnation." (232) Either way, as Winship held, due process prohibits the state from subjecting a person "to the stigma of a finding that he violated a criminal law and to the possibility of institutional confinement on proof insufficient to convict him." (233)

This due process objection to moral-predicate theories is not unassailable. Winship dealt with final determinations of guilt in delinquency proceedings, so it is not entirely clear that its holding extends to pretrial determinations of guilt for purposes of short-term restraint. Nor is it entirely clear that any deprivation imposed by virtue of guilt should be classified as punishment. (234) Moreover, even if current doctrine is a hurdle to moral-predicate models, the question at issue is normative: should the state have greater authority to preventively restrain defendants than equally dangerous non-defendants ? A moral-predicate proponent could intelligibly argue that probable guilt for a charged offense constitutes a moral basis to treat a defendant differently than an equally dangerous non-defendant, and that in the pretrial context (as opposed to the trial context), the stakes are not so high as to require that guilt be proven beyond a reasonable doubt before it is invoked to justify special restraints.

To advocate pretrial determinations of guilt at a less demanding standard of proof is no simple matter, however. The stakes may be as high in this context as in some delinquency proceedings or misdemeanor trials. The moral-predicate advocate might respond that we already permit pretrial deprivations on the basis of probable guilt: probable cause suffices for arrest. But this argument overlooks the fact that a probable-cause determination serves a different function than the guilt determination that a moral-predicate model would require. (235)

Even presuming that the probable guilt of a defendant could qualify as an independent moral justification for preventive restraint, there remains the daunting task of specifying what land of guilt determination should authorize what kind of restraint. What act, proven at what degree of confidence, with what procedures, justifies what restraint, and does it authorize restraint automatically or only if additional criteria are met? Each of these determinations is complex. Do all offenses constitute moral predicates for preventive restraint, or just serious offenses that entail significant culpability? What standard of proof must the state meet to demonstrate guilt for the charged offense in this context? Is defense counsel guaranteed, and what other procedural rights does the defense have to contest guilt? When guilt is sufficiently established, what degree of restraint does each offense trigger--are shoplifting and murder equal moral predicates, or does murder authorize greater restraint than shoplifting even if the shoplifter is more dangerous? (236) What other criteria must enter the calculus? (237) No existing pretrial regime has addressed and resolved these questions. (238)

No one, currently, is making a serious case for a moral-predicate regime. Any successful attempt to do so would require explaining why guilt justifies preventive restraint (i.e., defending a specific moral-predicate theory); rebutting the due process objection that any official judgment of guilt carries special stigma, such that it should only be made with near-certainty; and elaborating an affirmative vision of a moral-predicate regime that grapples with specifics. Unless and until someone does those things, the Winship principle should govern: the state may not invoke a person's guilt as moral justification for depriving her of liberty unless it has been proven beyond a reasonable doubt. (239)

2. The Causal-Responsibility Argument

The second possible line of moral argument focuses not on the defendant, but on the state. The argument is that the state has a greater duty to restrain dangerous defendants than equally dangerous non-defendants. Ronald Allen and Larry Laudan take this position, reasoning that "citizens have a right to be protected from criminal victimization," such that when the state has a dangerous defendant "in its custody," it has a duty to continue to restrain him. (240) If it releases him instead, "then the state bears a direct responsibility for such harm as that individual wreaks." (241)

This might be termed the causal-responsibility argument. The state seems to do something by releasing a defendant who goes on to commit a crime. (242) When it foresees crimes by equally dangerous non-defendants but does not intervene, on the other hand, it merely allows the crimes to happen. It is a foundational, though contested, principle of moral philosophy "that people have a greater responsibility, in general, for what they do than for what they merely allow or fail to prevent." (243) And with this greater responsibility comes a heightened duty of prevention.

The problem with this argument is that it assumes that the state has the legal or moral authority to hold a defendant, such that release is a meaningful choice. This is what makes release seem like an "act," something the state "does," and for which it bears special responsibility. If the premise is wrong--if the state must release a defendant--then it bears no moral responsibility for release and events that follow. The premise (that the state has authority to hold the defendant) thus begs the relevant question, which is whether the state has such authority in the first place. More specifically, it begs the question of why the state would have the authority to hold a defendant if it lacks authority to impose custody on an equally dangerous non-defendant. In other words, the causal-responsibility argument does not offer a justification for special state preventive authority in the pretrial arena. It is, instead, a conclusion that follows if the state has such authority.

Dissecting the intuition behind the causal-responsibility view is nonetheless valuable. First, the causal-responsibility argument might reflect loss aversion. We think of defendants as already in custody, so release seems like an affirmative action that requires justification, whereas detention feels like maintenance of the status quo. (244) Accordingly, we are quick to blame the judicial officer who ordered release whenever it contributes to harm. (245) But these are just psychological facts. They provide no normative justification for the causal-responsibility view.

A second, more promising intuition behind the causal-responsibility argument is that a defendant's arrest puts the state on notice of his dangerousness. (246) The fact that gives rise to a special state duty of crime prevention is not merely that the defendant has already been taken into custody. Rather, it is the nature of arrest.

To see this more clearly, imagine other custodial scenarios: a nurse is quarantined after working with Ebola patients; a recalcitrant witness is detained to compel her testimony; a jury member is sequestered during deliberations. Imagine that all are released, and each commits a domestic assault the following week. Assuming that neither the nurse, witness, nor juror appeared unduly dangerous while they were in custody, the fact that the state recently released them does not make it responsible for their crimes. Nor can the state be blamed for releasing them; it had no authority (legal or moral) to do otherwise. These scenarios illustrate the shallowness of the act-omission distinction. The state "acted" by releasing each person who subsequently committed a crime, and yet this act alone carries no particular moral significance.

The only difference in the pretrial context is the nature of the initial custody. The reason for the state's apparent causal responsibility for pretrial crime, then, is not the fact of release per se (the "act"), but rather the fact that arrest is thought to put the state on notice of a person likely to commit future crimes. On this view, defendants are different from equally dangerous non-defendants because they represent foreseeable harm. Given the state's general duty to prevent crime, the argument goes, it must take the opportunity to intervene once put on notice. Squandering such an opportunity is tantamount to a breach of duty (247) A more nuanced version of the causal-responsibility argument might thus run as follows: when arrest puts the state on notice of future crime risk, the state has a duty to prevent the future crime, and is morally responsible for failing to do so.

While not implausible, this notice-and-opportunity variant on the causalresponsibility argument still does not answer the question this Article sets out to resolve. Presume that it is correct, such that the state does have a heightened duty to prevent crime by defendants. The question is whether the state has authority to restrain a defendant if it could not permissibly restrain an equally dangerous person at large. And a heightened duty of crime prevention does not automatically translate into authority to restrain. The state always has some general duty to prevent crime, after all, and that general background duty does not automatically confer authority to preventively restrain non-accused people. If that is because pure preventive restraint is categorically morally prohibited, a heightened duty of crime prevention still does not overcome the moral prohibition. (248) A heightened duty might require the state to take other preventive measures, like providing supportive services that reduce the likelihood the defendant will turn to crime, minimizing the opportunities for crime, or enhancing deterrent penalties to discourage crime. But a heightened duty to prevent crime cannot overcome a moral prohibition on a particular mechanism of prevention.

If, on the other hand, the reason the state cannot permissibly restrain the equally dangerous non-defendant flows instead from an instrumentalist calculus (i.e., the costs of such restraint would outweigh its benefits), the notice-andopportunity argument has much greater traction. Notice of risk and an immediate opportunity for prevention might reduce the costs of pretrial restraint sufficiently to make it cost-justified. In an instrumentalist world, that would confer the authority to restrain defendants. And if the state has authority to restrain, a heightened duty might require it to do so.

That is to say, the notice-and-opportunity variant on the causal-responsibility argument is really an instrumentalist argument. It is not an independent moral argument. It is much more plausible than the shallow version of the causal-responsibility (act-omission) argument we first considered. But it depends on a set of instrumentalist conclusions. Specifically, the state may have both the authority and the duty to restrain defendants at lower thresholds of risk than non-defendants, if and only if (a.) the permissibility of restraint turns on a practical calculus, and (b) that calculus differs for defendants and non-defendants who pose equal risk. Assuming (a), the next Section considers whether (b) is true. (249)

C. No Clear Practical Distinction

There is a bevy of practical differences between defendants and equally dangerous non-defendants that might seem relevant to decision-making about preventive restraint. In cost-benefit terms, (250) these differences make the preventive restraint of defendants seem less costly. On closer analysis, however, the cost differential is not so apparent. The discussion that follows is admittedly speculative, both because the costs and benefits of pretrial preventive restraint are incompletely understood and because, at present, there is no analogous regime for non-defendants that would offer a direct comparison. That said, the analysis suggests that there is no clear cost difference between the preventive restraint of defendants and non-defendants who pose equal risk.

1. Benefits and Costs of Preventive Restraint

The central benefit of a preventive regime is that it averts harm. To quantify this benefit, it is necessary to specify the likelihood and severity of the harms it prevents. Even then, the nature of this benefit--the averted costs of crime--remains extremely complex. (251) The most immediate costs of crime are borne by the victim. (252) These individual costs may be physical, psychological, or financial. (253) Crime also has broader social costs, including the costs of treating victims; the costs of identifying, prosecuting, and punishing the perpetrator; the costs of aggravated fear, increased prevention measures, and disinvestment in crime-ridden neighborhoods; and the costs to the perpetrator's community if he is incarcerated. (254) Finally, the failure to prevent crime may cost the state political legitimacy, especially if it is perceived as indifferent to crime in already-marginalized communities. (255) The benefit of a preventive regime depends on the degree to which it averts these harms.

On the other side of the ledger, the costs of preventive restraint are equally complex. (256) The most profound cost is, of course, the loss of liberty to the person restrained, and its cascading social and economic effects. (257) Then there are the financial costs of operating a preventive regime and the lost tax revenue from the persons it keeps from working. (258) Preventive detention, furthermore, carries its own set of social costs, including unintended criminogenic effects, (259) the exacerbation of race and class inequalities, (260) and the psychological and relational costs of branding certain people or groups as dangers to the broader community. Lastly, preventive restraint entails a number of risks that any cost-benefit calculation should take into account, such as the possibility that prevention will serve as a veneer for punishment, (261) that the preventive regime will expand without good cause, (262) or that the use of predictive instruments will distort decision-making. (263)

Under an instrumentalist account, it is possible to make meaningful tradeoffs between these costs and benefits. Thus, at some threshold of likelihood and severity, a possible future crime can justify preventive interference. The question here is whether this threshold is different for defendants than for anyone else--that is, whether the cost-benefit calculus differs for defendants and non-defendants who pose an equal risk. Because the risk is equal, restraining either group produces equal benefit. The costs, however, might diverge.

2. Diminished Costs

There are three principal reasons that preventive restraint might seem less costly in the pretrial realm. (264) First, pretrial restraint is structurally limited to the period between arrest and adjudication, which minimizes concerns about uncontrolled expansion. Second, deprivations of liberty might cost less to defendants than to non-defendants. Third, arrest might provide a distinctively efficient means of identifying dangerous people and imposing restraint.

a. Bounded Restraint

The first diminished-cost argument is that pretrial restraint for dangerousness is less costly because it is bounded in time. (265) Preventive regimes tend to grow. Political incentives encourage institutional actors to expand the criteria for eligibility. (266) Once a person is caught in the preventive net, moreover, there are often powerful incentives to continue restraint and none to end it. This concern is attenuated in the pretrial context, the argument goes, because any restraint imposed is structurally limited. It cannot begin without a criminal charge supported by probable cause; it must end with dismissal or adjudication. A preventive regime outside the pretrial context could be designed with equivalent time limits, but, given the political liabilities of releasing a person who has already been designated as dangerous, enforcement of the limits would never be assured. The fact that the pretrial preventive restraint is structurally constricted makes it less alarming.

Although the boundedness of pretrial preventive restraint is a point in its favor, it constitutes an extremely limited reduction in cost. Given the unpredictability of pretrial proceedings, which can last months or even years, an explicitly time-bounded preventive regime not tied to the progress of a criminal case might actually provide greater protection against unnecessarily prolonged restraint. Regardless, the time-boundedness of pretrial restraint merely assuages one worry about a preventive regime. It does not diminish any of the actual harm that it inflicts. Finally, if a preventive regime for non-defendants did expand, such expansion would likely produce some additional benefit in terms of crime prevention. Direct cost comparison to a pretrial regime would no longer be possible. Thus, what initially seems like a significant cost difference between the restraint of defendants and equally dangerous non-defendants becomes, on reflection, an uncertain difference at best.

b. Less Cost in Liberty

A second argument is that, given the likelihood of conviction, restraints on pretrial defendants are relatively less disruptive to the individuals restrained than they would otherwise be. In an oft-cited national dataset, approximately sixty-six percent of felony defendants charged in state courts were eventually convicted and sentenced to a term of incarceration or probation. (267) One might argue that this eventual incursion on liberty lessens the impact of pretrial restraint.

This is a hard argument to assess. We have no satisfactory methodology for valuing liberty in the first place, let alone valuing it on the eve of a criminal sentence. (268) It is possible that such incremental deprivations are indeed less costly. On the other hand, for the person whose liberty is dwindling, and for her family, her last weeks of freedom may have a comparatively greater subjective value. In the absence of a compelling case, there is no reason to assume that a person's liberty is worth less just because she will soon have less of it.

A variation on this argument is that pretrial restraint costs less because it can be credited against the eventual sentences of those convicted. That is, detention has no cost to the guilty. The deprivation of liberty is just borrowed from their future punishment. (269)

The most obvious problem with this argument is its limited relevance: it applies only to those who receive a custodial sentence that is at least as long as the time they spent detained pretrial. Within that pool of defendants, moreover, the argument does not apply to people who receive a custodial sentence only as a result of being detained--that is, people who would not have received a custodial sentence if they had been released pretrial. For such people, the time spent in pretrial detention is not merely time they would have spent incarcerated anyway. Recent research (and anecdotal experience) suggests that this may be the case for a substantial percentage of misdemeanor defendants who serve custodial sentences. (270) And most misdemeanor defendants do not serve custodial sentences. (271) For these reasons, the "credited-against-punishment" argument likely includes no more than a small fraction of those arrested on misdemeanor charges.

Among felony defendants, the most recent national data show that approximately twenty-four percent were ultimately sentenced to prison. (272) Given that felony cases represent only around a quarter of all criminal cases, (273) the "credited-against-punishment" argument applies to around a quarter of a quarter of all defendants--or about six percent.

Furthermore, the only pretrial restraint that is credited against punishment in current practice is full-scale detention. (274) Time spent on pretrial supervision or GPS monitoring is not generally deducted from the eventual sentence imposed. (275) The "credited-against-punishment" argument therefore does not justify these lesser restraints. At best, then, this argument might justify pretrial detention for approximately six percent of all defendants.

There are additional problems with this best-case scenario. First, there is no foolproof way to know ahead of time which defendants will fall within the relevant six percent. While discounting their pretrial liberty loss might reduce the costs of pretrial restraint in the aggregate ex post, it cannot justify the imposition of restraint in an individual case ex ante. More fundamentally, claiming that pretrial detention is costless because it counts as punishment is problematic. Pretrial restraint is only constitutional if it is not punishment. It is at least arguable that it must therefore be justifiable without equating it with punishment. (276)

c. Notice and Opportunity

The last diminished-cost argument is that arrest gives the state notice of dangerous people and presents an immediate opportunity for intervention. It saves the costs of identifying dangerous people in the population at large (which we might term "search costs") and taking them into custody ("arrest costs"). Recall that this argument also forms the basis for the idea that the state has a heightened duty to prevent pretrial crime. (277)

The notice-and-opportunity argument has powerful appeal. It is indisputable that arrest can provide notice of risk and presents an immediate opportunity for restraint. Certain arrests may signal particular danger. And imposing any form of restraint on a dangerous person already in custody requires fewer resources than restraining a person who has not yet been located or processed into the system.

But the difference in search and arrest costs between defendants and non-defendants who pose the same risk may not be as substantial as one might imagine. Big data will increasingly allow governments to identify high-risk members of the general public with ease. And arrests, in the grand scheme of things, are not expensive.

To begin with search costs: If it is not the case already, many U.S. jurisdictions will soon have the capacity to identify non-defendants who pose precisely the same statistical risk as an average arrestee, or an average "high-risk" arrestee, with no more effort than it takes to identify high-risk arrestees themselves. (278) Implementing pretrial risk assessment already requires statistical analysis of large administrative datasets. Jurisdictions could conduct the same analysis for non-defendants. The databases, moreover, are only getting bigger. Governments are integrating their data systems so that it is possible to analyze consolidated data relating to individuals' employment, earnings, and past contacts with the DMV, criminal justice system, public housing and benefits system, child welfare system, public hospitals, as well as mental health and addiction treatment systems--in short, all past contacts with a pervasive state. (279) Given the scope of this data, predictive technologies could hypothetically enable fairly rigorous risk assessment of the entire population, or of non-defendant subgroups, with targeted interventions for those who score highest. (280) This is already becoming a reality. (281) As governmental data management improves, any real cost difference between the identification of statistically high-risk defendants and non-defendants who pose an equal risk will continue to collapse.

One might argue that limiting risk assessment to arrestees nonetheless promotes efficiency, because arrestees likely include a disproportionate number of dangerous people. That is, a pending criminal charge can function as a filter to facilitate efficient risk profiling. In order to identify equally dangerous people in the population at large, by comparison, the state would have to build and analyze much larger datasets. (282) Since these datasets will be built in any case, however, the relative efficiencies of each preventive system are likely to converge. Furthermore, using arrest as a profiling mechanism also has significant downsides. The concentration of policing in poor and minority communities means that members of such communities are arrested for drug and low-level crimes at rates disproportionate to the share of crime they commit. (283) Targeting this group for preventive restraint risks exacerbating these existing inequalities. In sum, using a pending criminal charge as a filter to identify dangerous people does not clearly produce net benefits.

The cost of arrest is a more obvious difference between the preventive restraint of defendants and the hypothetical equivalent restraint of non-defendants. Arrest provides an immediate opportunity for intervention, because defendants have already been subjected "to the physical dominion of the law." (284) They need not be physically taken into custody for purposes of restraint for dangerousness. (285)

The question is how much this alters the cost-benefit calculus. The answer depends, in turn, on how significant this cost differential is relative to the total costs of the preventive regime. Although the answer will necessarily be speculative, existing estimates of major relevant costs facilitate at least a rough assessment.

Suppose that the state wishes to detain ten people who each pose a ten percent risk of committing an aggravated assault. (286) According to the only recent effort to quantify the full cost of detention (on the basis of a national dataset), the cost of detention for the felony pretrial period would be $40,300 for the average state defendant. (287) Ten such detentions would cost $403,000. On the other side of the balance, the expected benefit of the ten detentions is to prevent one aggravated assault. A recent synthesis of prior research estimates the total average cost of an aggravated assault to be $118, 812. (288)

The cost of detention changes slightly if the ten detainees are non-defendants, because the state first has to locate them and take them into custody. These costs should amount to about as much as a simple arrest, which available estimates price between $150 and $880. (289) To err on the side of a greater cost differential between defendants and non-defendants, we will posit that a preventive arrest would cost $900. Thus, the additional cost of arresting the non-defendants would be $9,000, bringing the total cost of the ten detentions to $412,000.

Given the scale of the total costs, the cost of arrest is not very significant. The costs of detaining the ten non-defendants ($412,000) and the ten defendants ($403,000) both substantially outweigh the benefit ($119,812). There is a small margin where the $9,000 cost differential would tip the cost-benefit calculus--right around the degree of risk that makes the costs and benefits equal. Detention is not cost-justified for the ten non-defendants unless they each pose at least a 34.39 percent risk. Detention is cost-justified at a negligibly smaller percentage of risk, 33.64 percent, for the ten defendants. (290) The cost of arrest makes a difference for people who pose between a 33.64 percent and a 34.39 percent risk. In practice, these cost estimates are not sufficiently precise to be sure that this is exactly the right range of impact. Even if it is, our risk assessment tools are not precise enough to reliably identify people who fall into it.

This example suggests that, in an instrumentalist framework, the marginal cost difference between the equivalent restraint of defendants and non-defendants will be relatively small. It will flip the result of the cost-benefit analysis for a very select group. The state may indeed have a heightened duty to restrain defendants right on the borderline, and averting pretrial crime by this group may have particular benefits. (291) But few people will qualify, and we probably cannot tell who they are.

The takeaway is that, although this cost difference is not meaningless, it does not constitute a categorical justification for a regime of preventive restraint that we would never permit outside the pretrial context. For the vast majority of people--defendants and non-defendants--the fact that it is marginally cheaper to restrain those who have already been arrested will not affect the result of the cost-benefit analysis. (292) In all but a few cases, this cost differential provides no justification for preventive restraint of defendants that we would not permit for equally dangerous members of the general public.

Finally, even if the costs of preventive restraint are diminished in some ways pretrial, they are heightened in others. A growing body of rigorous empirical scholarship shows that pretrial detention causally increases the likelihood of conviction and a carceral sentence. (293) The increase is due almost exclusively to an increase in guilty pleas. In other words, some number of defendants plead guilty only because they are detained. Furthermore, scholars speculate that this phenomenon represents a significant source of wrongful convictions. (294) This special cost of preventive detention should be taken into account.

In sum, while several costs of preventive detention might be diminished in the pretrial context, the difference is not as significant as it might initially appear. This discussion is far from conclusive. But in the end, there is no clear evidence that the outcome of the cost-benefit analysis for preventive restraint differs for defendants versus non-defendants who pose equal risk.

III. PARITY OF PREVENTIVE AUTHORITY

Thus far, this Article has argued that, for purposes of preventive restraint, there is no clear constitutional, moral, or practical distinction between equally dangerous defendants and non-defendants. No doctrine holds that defendants have a diminished constitutional right against preventive interference. There is no clear moral basis authorizing greater preventive restraint of defendants. And the practical advantage of restraining a defendant versus an equally dangerous non-defendant is questionable at best.

This conclusion provides the first part of an answer to the question posed in the Introduction: what degree of risk justifies pretrial restraint for dangerousness? Since there is no basis to conclude otherwise, the answer should be "whatever degree of risk would justify equivalent restraint of a non-accused person." If a defendant is no more dangerous than many non-defendants, and that level of risk would not authorize restraint of a non-defendant, it should not authorize restraint of the defendant either. This principle can be termed "parity of preventive authority" or, for short, the "parity principle."

A. The Parity Principle in Action

The parity principle does not itself prohibit preventive detention. Rather, its implications depend on whether there is some threshold of risk at which preventive restraint of a non-accused person is permissible. If there is no such threshold, then there is no justification for preventive restraint of defendants either. But if there is a risk threshold at which pure preventive restraint is permissible, then pretrial preventive restraint is likewise permissible at that threshold of risk.

1. What Risk Justifies Pure Preventive Restraint?

The question of whether pure preventive restraint is permissible for non-defendants, and at what degree of risk, can be cast in either positive or normative terms. The positive question is whether current law authorizes pure preventive restraint, and, if so, at what risk threshold. The normative question is whether the law should authorize such restraint and, if so, at what risk threshold. Both questions are complex. A comprehensive treatment of either is beyond the scope of this Article. Nonetheless, a brief discussion is useful.

In positive terms, current U.S. law authorizes a wide array of purely preventive restraint. Preventive detention regimes include, for example, the civil commitment of sexually violent predators, immigration detention, the detention of suspected terrorists, and the involuntary commitment of pregnant drug users. (295) Short of detention, purely preventive restraints include temporary restraining orders, the No-Fly List, offender registration regimes, peace bonds, and restrictions on professional licensure for people with past convictions. (296) These regimes reflect a judgment that it is permissible for the state to restrain responsible agents solely to prevent future crime. Many of these regimes have been highly controversial, but courts have stopped short of finding any wholly unconstitutional. (297) In positive terms, then, the parity principle permits pretrial preventive restraint at whatever degree of risk authorizes analogous restraint within these purely preventive regimes.

Ascertaining where this risk threshold lies for any given restraint, however, is no easy matter. Current preventive law is extremely amorphous. The criteria for the No-Fly list are a mystery. Offender registration statutes and categorical licensure bars simply presume dangerousness on the basis of past conviction. The dangerousness standards in civil commitment statutes vary widely by state, as do standards for the commitment of sexual predators. (298) Most require some showing that the person to be committed is otherwise likely to cause harm to herself or others, but the definition of harm varies. (299) No statute specifies what numerical probability of a given harm in a given timespan warrants restraint. It is possible that a comprehensive survey of these fields might distill more robust common standards, but none are readily apparent.

The more useful iteration of the risk-threshold question might be the normative one: What probability of what harm over what timespan should authorize a given restraint? Is a five percent likelihood of committing an assault over the course of three months a sufficient basis for GPS monitoring? For detention? How many people should we be willing to restrain for six months to prevent one armed robbery?

Some jurists and scholars wholly reject pure preventive restraint, at least when it takes the form of full custodial detention. They believe that the detention of people who are responsible agents solely to prevent future intentional harm violates a fundamental tenet of any liberal legal order by treating restrained persons as less than autonomous moral beings. (300) To the extent that this is correct, purely preventive detention regimes should be eliminated. The parity principle would prohibit pretrial preventive detention as well.

Others, including myself, hold that restraint of responsible agents, including detention, may be warranted to prevent future crime at some threshold of risk. (301) To the extent that this is true, the parity principle permits pretrial preventive restraint at that threshold as well. The remainder of this Article presumes that this is the case.

The appropriate risk threshold for any given preventive restraint cannot be resolved here. (302) But existing preventive laws do share a common emphasis on preventing bodily harm. Those statutes that offer some specificity about what risk justifies pretrial preventive detention invoke the risk of violent crime. (303) Involuntary commitment laws tend to require a showing that the person to be committed is otherwise likely to cause physical harm. (304) Quarantine is justified by the risk of physical harm (contagion), and restraints designed to prevent sex crimes and terrorism target particular genres of violence. It therefore seems reasonable to propose that nothing less than a substantial likelihood of serious violent crime within a six-month span can justify onerous restraints on liberty. (305) The question of what counts as "substantial" remains. Is five, thirty, or sixty percent enough? (306) This is a debate that reformers ought to have, and it is possible that the answer will vary across jurisdictions.

2. Policy Implications

The policy implications of the parity principle depend on what degree of risk justifies pure preventive restraint. Presuming that such restraint is sometimes permissible, but never on the basis of anything less than a substantial risk of serious violent harm in a six-month span, then pretrial restraint of defendants for general dangerousness is permissible if, and only if, the evidence supports that degree of risk. To implement this risk threshold, jurisdictions undertaking reform should take three practical steps.

First, new statutes or policy guidance should specify that pretrial restraint for dangerousness is warranted only when a defendant presents a substantial risk of serious violent crime in the pretrial phase, and there are no less restrictive alternatives that would render the risk less than substantial. Ideally such laws would specify the numerical probability constituting a "substantial" risk. If agreement cannot be reached, the law could instead require any judge that finds there is a substantial risk to state what she believes the probability to be. Such transparency would promote debate and facilitate the organic development of a numerical definition in particular jurisdictions.

Articulating this risk threshold would clarify current pretrial standards. Six of the nineteen state constitutional provisions that authorize preventive detention condition it on a risk of violence. (307) But ten condition it on a vaguely articulated "danger" or the need to ensure "safety," and three do not articulate a se-verity-of-harm threshold at all. (308) State statutory law varies tremendously, but rarely provides an explicit severity-of-harm threshold. (309) The Bail Reform Act authorizes pretrial restraint to protect "the community" against any criminal activity at all. (310) As for the likelihood of harm, most laws mandate restraint if it is necessary to "adequately protect" or "reasonably assure" the safety of the community. (311) These standards are too vague to provide practical guidance.

Second, pretrial risk assessment tools should aspire to measure and clearly communicate the probability that a defendant will commit serious violent crime before trial. This means that risk assessment tools should stop measuring crime risk in terms of the likelihood of arrest for anything. "Any arrest" is an overbroad proxy for harm. Some eleven million people are arrested each year; their charges range from unpaid traffic fines to murder. (312) One-third of arrests lead to dismissal or acquittal. (313) And members of poor communities of color are disproportionately arrested for low-level crimes. (314) Moreover, those at highest risk of rearrest are not the people at highest risk of rearrest for violent crime, and vice versa. (315) If the goal is to avert violence, focusing on the likelihood of any future arrest targets the wrong individuals and fails to target the right ones.

Pretrial risk assessment tools should instead measure crime risk in terms of the likelihood of rearrest for a serious violent crime in the pretrial phase. This measure does not avoid all difficulties. The harm is the actual commission of violent crime. Many people are wrongfully arrested, and many people who commit violent crimes escape arrest. So, arrest for a serious violent crime is still both over- and under-inclusive as a proxy for the commission of violent crime itself. At the moment, however, it is the best measure available; conviction is thought to be too under-inclusive to be useful for these purposes. (316) Classifying by likelihood of pretrial arrest for a violent crime is wholly possible, as a number of recent studies and risk assessment tools demonstrate. (317)

Additionally, rather than make an implicit normative judgment that a given degree of risk is "low" or "high," pretrial risk assessment tools should instead communicate the statistical likelihood that a particular defendant will be rearrested for violent crime in the pretrial phase. Although any validation study of a tool that measures this outcome should produce this information, most instruments do not emphasize or communicate it to decision-makers. (318) A side benefit of this approach may be that it is less stigmatizing to restrain someone on the basis of an eight percent chance of rearrest for violent crime than on the basis that she is "dangerous" or poses a "threat." The former at least makes clear that the restraining authority thinks it highly unlikely that the person restrained will actually commit the feared harm.

Third, jurisdictions should ensure that any intervention to mitigate future crime risk is the least restrictive means to render the risk less than substantial. If GPS monitoring reduces the risk below this threshold, for example, detention would not be justified. Any intervention, moreover, should be cost-justified at the margin, relative to alternatives. (319) As Sonja Starr has noted in the post-conviction context, the real policy question is not "who's riskiest?" but "what will most effectively mitigate the risk?" (320)

As of now, we know almost nothing about the elasticity of serious, violent pretrial crime to interventions short of detention. But this information will be critical to understand as pretrial reform advances. It is worth remembering, too, that not all risk-management techniques need be intrusive. Prediction is not intrusive per se, and it need not lead to restraint. (321) The state might seek to respond to risk forecasts in a way that affirms individuals' agency, rather than disregards it. A jurisdiction might offer increased support and incentives to high-risk defendants, for example, instead of supervision. (322) Given that those most likely to commit future crimes may also be most likely to be future crime victims, a response that addresses the needs of defendants rather than risk alone may have significant advantages. (323)

In sum, the parity principle extends the presumptive threshold for pure preventive restraint--nothing less than a substantial risk of serious violent crime in a six-month span--to defendants and non-defendants alike. Accordingly, jurisdictions pursuing pretrial reform should (1) clarify that nothing less than a substantial risk of violence in the pretrial phase warrants restraint for general dangerousness; (2) ensure that risk assessment tools measure and communicate the likelihood of rearrest for serious violent crime; and (3) evaluate potential interventions (not all of which need be restrictive) in terms of their efficacy at reducing serious violent crime. These improvements in risk analysis and management are well within our grasp.

3. A World with Parity

What would a world governed by the parity principle look like? The most significant change would be a heightening of the standards for pretrial preventive restraint. Fewer defendants would be detained and supervised pretrial. It is possible, but unlikely, that the proposed framework could also encourage new preventive restraint outside the pretrial context.

The parity principle would not authorize the state to sweep up large groups of people who are rendered statistically dangerous by circumstances beyond their control, like gender, race, poverty, and age. Those variables simply do not have adequate predictive power. The most powerful predictors of violent crime are past arrest and conviction for violent crime. Thus, only people with a record of past violence will meet the relevant risk thresholds. A recent arrest for violence, moreover, is more predictive than an old one. So, in practice, it may well be that the only people who meet the risk standard for detention are those who have been credibly charged with a recent, serious crime. (324) In that sense, the parity principle does not foreclose a unique regime of preventive detention in the pretrial realm. But regardless, pretrial detention would meet the criteria of the parity principle: the state could justifiably detain equally dangerous non-defendants if it knew of any.

It might also be the case that, although both accused and non-accused people meet the relevant risk threshold, the state would elect to restrain defendants only. Pursuant to the parity principle, preventive restraint of a defendant is permissible if the state could justifiably impose the same restraint on an equally dangerous non-defendant. The parity principle does not require that the state actually treat both identically. It requires parity of preventive authority, not parity of preventive treatment.

A caveat is in order here. So far, I have assumed consensus not just on the parity principle, but also on the substantive judgment that nothing less than a substantial risk of serious violent harm can justify pure preventive restraint. As long as this is the baseline standard, the parity principle requires pretrial standards to level up to meet it. But if society decided that pure preventive restraint is justified at much lower levels of risk, the parity principle might have no effect on pretrial practice. It might even authorize a leveling-down. It is also possible that the notion of "parity" might encourage governments to extend current pretrial restraint standards to non-defendants. These are real concerns. On the other hand, a risk standard that applies outside the pretrial context might be the best bulwark against unwarranted preventive restraint. The more universal the standard, the less likely it is to be distorted by the perception that the only people it affects are an undeserving Other. (325)

B. The Parity Principle as Benchmark

This Article does not assert the parity principle conclusively. The argument in Part II is contestable at many points along the way. Even for skeptical readers, however, the parity principle can serve a useful analytical purpose. The ultimate question, with which this Article began, is what degree of crime risk justifies pretrial restraint for dangerousness. The parity principle provides a helpful benchmark. To answer the ultimate question, the analysis can begin with whatever degree of risk--if any--would justify restraint of a non-accused person. Those who believe that there is good reason to diverge from this standard in the pretrial realm should bear the burden of articulating what distinction justifies divergence, why it does, and how much divergence it warrants.

Some readers, for example, might think this Article has understated the efficiency of leveraging the infrastructure of arrest and prosecution to operate a preventive regime. Such readers must consider, first, whether preventive restraint can be justified on wholly instrumentalist grounds. If not, then a marginal difference in the instrumentalist calculus is irrelevant. If so, the question becomes how much the "true" cost differential alters the cost-benefit calculus. Unless it flips the outcome for a broad swath of people--which seems unlikely--it remains the case that the state is justified in preventively restraining defendants only at or just below the risk threshold where we would authorize preventive restraint of a non-accused person. We should still, in this case, endeavor to determine when pretrial restraint for dangerousness is justified by asking what degree of risk would be sufficient to restrain a non-defendant.

CONCLUSION

The third generation of bail reform has made tremendous strides in limiting money-bail practices that result in the systemic and unnecessary detention of the poor. Still, reformers would do well to keep in mind that the new risk-based model for pretrial policy is no panacea. A pretrial system centered on danger presents problems of its own.

Most fundamentally, there is no clear constitutional, moral, or practical basis for subjecting defendants to preventive restraint that would be unjustified for equally high-risk members of the population at large. This means that we should think carefully about what degree of risk warrants purely preventive restraint: what likelihood of what outcome over what timespan? (326) This is an extremely uncomfortable judgment to make. But it is necessitated by our new statistical methodology. The choice can be implicit or explicit, but it cannot be avoided.

Presuming that we proceed on this instrumentalist footing, the reform model presents a further set of practical problems. Suppose we decide that a thirty percent chance of arrest for a serious violent crime in the next three months justifies three months of detention. There is, first, the problem of predictive power. None of the existing pretrial risk assessment tools can forecast with this degree of precision. (327) Such precision may be attainable, but the tools are not there yet. Second, there is a racial equality problem. A recent ProPublica study sparked outrage by demonstrating that a prominent risk assessment tool produced much higher rates of "false" high-risk classifications for black defendants than for white, and much higher rates of "false" low-risk classifications for white defendants than for black. (328) Yet this is the inevitable result of any statistical algorithm developed on the basis of a mixed-race population where there is a higher incidence of both the primary risk factors (past criminal justice contacts) and the measured outcome (rearrest) among black defendants. (329) Actuarial risk assessment also presents thorny transparency problems and the possibility of perverse framing effects. The degree to which we can mitigate these problems remains uncertain.

The challenges of statistically informed policy do not mean that we should abandon actuarial forecasting altogether. Subjective risk assessment is no better--and probably worse. To the extent that we authorize or require predictions of future harm, actuarial tools hold great promise. We should, however, proceed with caution. And caution should begin with clarity about the grounds of preventive intrusions.

APPENDIX
TABLE 3. PRETRIAL RAIS--RISK FACTORS (330)

     FLPRAI                         PTRA

 1.  Age at admission;          1.  # Felony convictions;
 2.  Current most serious       2.  Prior FTAs;
     charge;                    3.  Pending felonies or
 3.  Is current charge              misdemeanors;
     907.041;                   4.  Current offense type;
 4.  Employment status at       5.  Offense class;
     admission;                 6.  Age at interview;
 5.  Marital status;            7.  Highest education;
 6.  Have a telephone or cell   8.  Employment status;
     phone;                     9.  Residence;
 7.  Time at current           10.  Current drug
     residence;                     problems;
 8.  History of substance      11.  Current alcohol
     abuse and/or mental            problems;
     health problems;          12.  Citizenship status;
 9.  Previous FTAs;            13.  Foreign ties
10.  Previous adult
     felonies;
11.  Previous adult
     misdemeanors

     VPRAI                          ORAS/IRAS-PAT
 1.  Charge type (M/F);         1.  Age at first arrest;
 2.  Pending charge(s);         2.  # FTA warrants past
 3.  Criminal history;              24 months;
 4.  Two or more FTAs;          3.  Three or more prior jail
 5.  Two or more violent            incarcerations;
     convictions;               4.  Employed at time of
 6.  Length of current              arrest;
     residence less than one    5.  Residential stability;
     year/residence             6.  Illegal drug use during
     verified;                      past six months;
 7.  Not employed two years     7.  Severe drug use
     or primary caregiver/          problem
     employed or primary
     caregiver;
 8.  History of drug abuse

     FLPRAI                         CPAT

 1.  Age at admission;          1.  Having a home or cell
 2.  Current most serious           phone;
     charge;                    2.  Owning or renting one's
 3.  Is current charge              residence;
     907.041;                   3.  Contributing to
 4.  Employment status at           residential payments;
     admission;                 4.  Past or current
 5.  Marital status;                problems with alcohol;
 6.  Have a telephone or cell   5.  Past or current mental
     phone;                         health treatment;
 7.  Time at current            6.  Age at first arrest;
     residence;                 7.  Past jail sentence;
 8.  History of substance       8.  Past prison sentence;
     abuse and/or mental        9.  Having active
     health problems;               warrants;
 9.  Previous FTAs;            10.  Having other pending
10.  Previous adult                 cases;
     felonies;                 11.  Currently on
11.  Previous adult                 supervision;
     misdemeanors              12.  History of revoked bond
                                    or supervision
     VPRAI                          PSA
 1.  Charge type (M/F);         1.  Current violent offense;
 2.  Pending charge(s);         2.  Pending charge at time of
 3.  Criminal history;              offense;
 4.  Two or more FTAs;          3.  Prior misdemeanor
 5.  Two or more violent            conviction;
     convictions;               4.  Prior felony conviction;
 6.  Length of current          5.  Prior violent conviction;
     residence less than one    6.  Prior FTA pretrial in past
     year/residence                 two years;
     verified;                  7.  Prior FTA pretrial older
 7.  Not employed two years         than two years;
     or primary caregiver/      8.  Prior sentence to
     employed or primary            incarceration;
     caregiver;                 9.  Age at current arrest
 8.  History of drug abuse


(1.) See infra notes 121-122 and accompanying text.

(2.) Todd D. Minton& Zhen Zeng, Jail Inmates at Midyear 2014, BUREAU JUST. STAT. 1 (June 2015), http://www.bjs.gov/content/pub/pdf/jim14.pdf [http://perma.cc/TJ6g-EYN7].

(3.) Brian A. Reaves, Felony Defendants in Large Urban Counties, 2009 - Statistical Tables, BUREAU JUST. STAT. 1, 15 (Dec. 2013), http://www.bjs.gov/content/pub/pdf/fdlucog.pdf [http://perma.cc/5EJA-XGNQ] (reporting that nine out of ten detained felony defendants had bail set).

(4.) Recent studies report that misdemeanor pretrial detention rates in several large cities range from twenty-five to more than fifty percent. Mary T. Phillips, Pretrial Detention and Case Outcomes, Part 1: Nonfelony Cases, N.Y.C. CRIM. JUST. AGENCY (2007), http://www.nycja.org/lwdcms/doc-view.php?moduIe=reports&module_id=66g&doc_name=doc [http://perma.cc/HL2K-ZLLM] (reporting that twenty-five percent of misdemeanor defendants are detained pretrial in New York City); Paul Heaton, Sandra Mayson & Megan Stevenson, The Downstream Consequences of Misdemeanor Pretrial Detention, 69 STAN. L. REV. 711, 733, 736 tbl.1 (2017) (reporting that fifty-three percent of Houston misdemeanor defendants were detained pretrial from 2008 to 2013); Charlie Gerstein, Note, Plea Bargaining and the Right to Counsel at Bail Hearings, 111 MICH. L. REV. 1513, 1525 n.81 (2013) (reporting that twenty-five percent of New York City misdemeanor defendants and fifty percent of Baltimore misdemeanor defendants are held on bail); Megan Stevenson, Distortion of Justice: How the Inability To Pay Bail Affects Case Outcomes 12 (Jan. 12, 2017) (unpublished manuscript) (on file with author) (reporting that, between 2006 and 2013, forty percent of defendants with bail set at five hundred dollars or less were detained in Philadelphia); see also id. at 11 (noting that twenty-eight percent of detained defendants only had misdemeanor charges).

(5.) See infra notes 83-122 and accompanying text.

(6.) See infra notes 83-122 and accompanying text.

(7.) Developing a National Model for Pretrial Risk Assessment, LAURA & JOHN ARNOLD FOUND. 5 (Nov. 2013) [hereinafter LJAF, Developing a National Model], http://www.arnoldfoundation.org/wp-content/uploads/2014/02/LJAF-research-summary_PSA-Court_4_1.pdf [http://perma.cc/64WV-EXQE].

(8.) See infra notes 93-94 and 121.

9. See, e.g., The Dangers of Bail Reform Across the U.S., U.S. BAIL REFORM NEWS, http://www.usbailreform.com/in-the-news [http://perma.cc/44SV-6EEC]; Have Bail Reforms Made Us Safer? Two Views, DAILY REC. (June 18, 2017), http://www.dailyrecord.com/story/opinion/2017/06/18/bail-reform-new-jersey/102990434 [http://perma.cc/QR67-66ZZ]. Opponents of bail reform are supporting a lawsuit recently filed by the mother of a murder victim against Chris Christie, alleging that New Jersey's bail reform is responsible for her son's killing by a defendant on pretrial release. See, e.g., Ariel Scotti, Dog the Bounty Hunter Joins Bail Reform Lawsuit against Chris Christie, N.Y. DAILY NEWS (Aug. 1, 2017), http://www.nydailynews.com/news/national/dog-bounty-hunter-joins-bail-lawsuit-chris-christie-article-1.3374458 [http://perma.cc/3PEB-UL7Y]; Christine Stuart, Murder Victims Mother Sues Chris Christie Over NJ Bail Reform, COURTHOUSE NEWS SERV. (Aug. 1, 2017), http://www.courthousenews.com/murder-victims-mother-sues-chris-christie-nj-bail-reform [http://perma.cc/37Y4-BRDX].

(10. ) Malcolm M. Feeley & Jonathan Simon, The New Penology: Notes on the Emerging Strategy of Corrections and Its Implications, 30 CRIMINOLOGY 449, 452, 455 (1992).

(11.) LJAF, Developing a National Model, supra note 7, at 1; see also Shima Baradaran & Frank L. McIntyre, Predicting Violence, 90 TEX. L. REV. 497, 547 (2012) (concluding on the basis of an empirical study "that judges are basing their [pretrial] decisions far more on predicted violence than on predicted flight"). The contemporary emphasis on danger is also reflected in the name of what has rapidly become the most prominent pretrial risk assessment tool: the Public Safety Assessment (PSA). See Public Safety Assessment, LAURA & JOHN ARNOLD FOUND., www.arnoldfoundation.org/initiative/criminal-justice/crime-prevention/public-safety-assessment [http://perma.cc/NAF4-DRYK].

(12.) For recent discussions of flight risk, see Lauryn P. Gouldin, Defining Flight Risk, U. CHI. L. REV. (forthcoming 2018); Lauryn P. Gouldin, Disentangling Flight Risk from Dangerousness, 2016 BYU L. REV. 837 (2016) [hereinafter Gouldin, Disentangling]; and Samuel R. Wiseman, Pretrial Detention and the Right To Be Monitored, 123 YALE L.J. 1344, 1351 (2014).

(13.) Is it better that ten men who will commit future crime go free than one who would otherwise commit no crime be detained, or the reverse? Blackstone himself endorsed much greater over-inclusiveness in preventive restraint than in punishment, but did not offer a precise ratio. See 4 WILLIAM BLACKSTONE, COMMENTARIES * 252-56.

(14.) Federal Pretrial Risk Assessment Instrument (PTRA), Version 2.0 (Mar. 1, 2010) [hereinafter Federal PTRA], http://www.pretrial.org/download/risk-assessment/Federal%20Pretrial%20Risk%20Assessment%20Instrument%20(2010).pdf [http://perma.cc/PT9S-LS6K]. The instrument does not specify the average length of the pretrial period in the dataset from which it was developed.

(15.) James Austin et al., Florida Pretrial Risk Assessment Instrument, JFA INST. 4, 11,13, http://www.pretrial.org/download/risk-assessment/FL%20Pretrial%20Risk%20Assessment%20Report%20(2012).pdf [http://perma.cc/BW42-XC9E].

(16.) Results from the First Six Months of the Public Safety Assessment--Court in Kentucky, LAURA & JOHN ARNOLD FOUND. 3 (July 1, 2014) [hereinafter LJAF, Results], http://www.arnoldfoundation.org/wp-content/uploads/2014/02/PSA-Court-Kentucky-6-Month-Report.pdf [http://perma.cc/S8GJ-ZKZ2].

(17.) See infra notes 57-61 and accompanying text.

18. See, e.g., Press Release, ACLU, Bail Reform Rules Must Uphold the Aims of Criminal Justice Reform, ACLU-NJ and Other Groups Tell NJ Supreme Court (June 3, 2016), http://www.aclu.org/news/bail-reform-rules-must-uphold-aims-criminal-justice-reform-aclu-nj-other-groups-tell-nj-supreme [http://perma.cc/6LKD-AHPE] (expressing concern about draft bail reform rules); Bail System Reform, ACLU N.J., http://www.aclu-nj.org/legislation/bail-system-reform1 [http://perma.cc/UR65-64CD] (opining that "[a] system based on actual risk to the community--not ability to pay--better protects public safety, reduces jail over-crowding, keeps families together, and saves taxpayer resources"); Job Announcement, ACLU, Staff Attorney--Bail Reform [CLRP-25], ACLUF, Trone Center for Justice and Equality, New York (Apr. 13, 2017), http://www.aclu.org/careers/staff-attorney-bail-reform-clrp-25-acluf-trone-center-justice-and-equality-new-york [http://perma.cc/KS2V-BVRB] (explaining that the national ACLU is seeking a staff attorney to "develop, pursue, and conduct litigation aimed at bail reform in America").

(19.) See, e.g., Jessica M. Eaglin, Constructing Recidivism Risk, 67 EMORY L.J. 59 (2017); Melissa Hamilton, Risk-Needs Assessment: Constitutional and Ethical Challenges, 52 AM. CRJM. L. REV. 231, 256-61 (2015); Anne Milgram et al., Pretrial Risk Assessment: Improving Public Safety and Fairness in Pretrial Decision Making, 27 FED. SENT'G REP. 216, 220 (2015); Julia Angwin et al., Machine Bias, PROPUBLICA (May 23, 2016), http://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing [http://perma.cc/FB8E-WSV2]; see also BERNARD E. HARCOURT, AGAINST PREDICTION: PROFILING, POLICING, AND PUNISHING IN AN ACTUARIAL AGE (2007) [hereinafter HARCOURT, AGAINST PREDICTION] (arguing against predictive law enforcement); Bernard E. Harcourt, Risk as a Proxy for Race: The Dangers of Risk Assessment, 27 FED. SENT'G REP. 237 (2015); Sonja B. Starr, Evidence-Based Sentencing and the Scientific Rationalization of Discrimination, 66 STAN. L. REV. 803, 803 (2014) (critiquing actuarial recidivism risk prediction instruments as "an explicit embrace of otherwise-condemned discrimination"); Megan Stevenson & Sandra G. Mayson, Pretrial Detention and Bail, in 3 REFORMING CRIMINAL JUSTICE : PRETRIAL AND TRIAL PROCESSES 21, 34-39 (Erik Luna ed., 2017); Sandra G. Mayson, Bias In, Bias Out: Criminal Justice Risk Assessment and the Myth of Race Neutrality (unpublished manuscript) (on file with author).

(20.) See infra note 121.

(21.) See infra note 122. By contrast, in New York, stakeholders continue to resist pretrial restraint for dangerousness entirely, galvanizing debate. See, e.g., Tina Luongo & Cherise Fanno Burdeen, Letters to the Editor, Setting Bail and Assessing Risk to Public Safety, N.Y. TIMES (Apr. 3, 2017), http://www.nytimes.com/2017/04/03/opinion/setting-bail-and-assessing-risk-to-public-safety.html [http://perma.cc/74LS-43CN].

(22.) See, e.g., Laurence H. Tribe, An Ounce of Detention: Preventive Justice in the World of John Mitchell, 56 VA. L. REV. 371, 378 (1970) ("To the limited extent that medical science furnishes techniques capable of objectively ascertaining the presence of dangerously incapacitating illness,... neither a comparable body of knowledge nor a comparable technology of prediction is yet available for dealing with criminal behavior generally."). But see Note, Preventive Detention Before Trial, 79 HARV. L. REV. 1489, 1506-07 (1966) (concluding that "given the current state of behavioral prediction," judges' discretion to detain should be restricted but not necessarily removed).

(23.) Tribe, supra note 22, at 382.

(24.) See Eaglin, supra note 19, at 87-88 (describing the algorithm design process).

(25.) See State v. C.W., 156 A.3d 1088, 1099-1100 (N.J. Super. Ct. App. Div. 2017) (citing N.J. STAT. ANN. [section] 2A:162-23(a)(2) (West 2017); N.J. Court Rule 3:4A(b)(5)).

(26.) Until a New Jersey-specific validation study is performed, we do not know exactly what that probability is, but if the premise of the PSA is correct--that it functions with comparable accuracy across jurisdictions--then a high-risk classification in New Jersey should approximately correspond to the risk it was shown to represent in the only published evaluation: a 23% chance of rearrest for anything in a six-month time span, or, for those flagged as high risk for violence, an 8.6% chance of rearrest on a violent charge. LJAF, Results, supra note 16, at 3.

(27.) The Chicago police department has perhaps come closest to this scenario by keeping a highly controversial "heat list" of high-risk individuals for special surveillance, threats, and incentives. See Matt Stroud, The Minority Report: Chicago's New Police Computer Predicts Crimes, but Is It Racist?, VERGE (Feb. 19, 2014, 9:31 AM), http://www.theverge.com/2014/2/19/5419854 [http://perma.cc/T678-AA68].

(28.) Note that this question does not arise for restraint to prevent flight or obstruction of justice, because those risks are unique to the pretrial process. There are pressing questions, however, about what restraints are permissible to mitigate the risks of flight and obstruction. See, e.g., R. A. Duff, Pre-Trial Detention and the Presumption of Innocence, in PREVENTION AND THE LIMITS OF THE CRIMINAL LAW 125-28 (Andrew Ashworth et al. eds., 2013) (arguing for limitations on pretrial restraints to prevent obstructive harms); Jocelyn Simonson, Bail Nullification, 115 MICH. L. REV. 585 (2017) (raising questions about the utility and constitutionality of money bail as a mechanism to ensure appearance); Wiseman, supra note 12, at 1350 (arguing that "non-dangerous defendants" have a right to electronic monitoring, in lieu of detention, to prevent flight).

(29.) 481 U.S. 739 (1987).

(30.) See, e.g., John N. Mitchell, Bail Reform and the Constitutionality of Pretrial Detention, 55 VA. L. REV. 1223 (1969) (arguing that pretrial preventive detention is constitutional); Tribe, supra note 22 (arguing against Mitchell); infra note 57 and accompanying text.

(31.) Tribe, supra note 22, at 405.

(32.) See, e.g., Stephen J. Morse, Blame and Danger: An Essay on Preventive Detention, 76 B.U. L. REV. 113, 119 n.15 (1996) (noting that it is not clear why preventive detention of a defendant "is more justifiable than for any other person posing an equal risk of similar harmdoing").

(33.) See, e.g., Duff, supra note 28, at 128 (explaining that, according to "traditional liberal" principles, "[r]esponsible agents ought to be left free to determine their own conduct... and are properly liable to coercion only if and when they embark on a criminal enterprise").

(34.) John E. Coons, Consistency, 75 CALIF. L. REV. 59, 59 (1987) ("Like cases should be treated alike: This form[u]la of Aristotle is widely accepted as a core element of egalitarian moral and social philosophy." (citing ARISTOTLE, ETHICA NICOMACHEA [section] 1131a-b (W.D. Ross trans., Clarendon Press 1925) (c. 384 B.C.E.))).

(35.) See Duff, supra note 28, at 120 (suggesting that pretrial detention incurs little outrage because "the defendant is seen as being in fact an offender, who awaits only the formal verdict of the court before receiving the punishment he deserves"); cf. Rachel A. Harmon, Federal Programs and the Real Costs of Policing, 90 N.Y.U. L. REV. 870, 903 (2015) ("One reason the coercion costs of policing are neglected is that many of them accrue to the targets of policing....").

(36.) See, e.g., ANDREW ASHWORTH & LUCIA ZEDNER, PREVENTIVE JUSTICE 152-67 (2014) (discussing forms of preventive detention in the United States and internationally); Adam Klein & Benjamin Wittes, Preventive Detention in American Theory and Practice, 2 HARV. NAT'L SECURITY J. 85, 87 (2011) (cataloging the "diverse statutes and regimes authorizing the preventive detention of individuals not convicted of a crime to prevent harms caused by that person").

(37.) See, e.g., David Cole, Out of the Shadows: Preventive Detention, Suspected Terrorists, and War, 97 CALIF. L. REV. 693, 695 (2009) ("In reality,... preventive detention is already an integral feature of the American legal landscape."); Christopher Slobogin, The Civilization of the Criminal Law, 58 VAND. L. REV. 121, 124(2005) (noting "the widespread promulgation of new incapacitation regimes" that condition detention on dangerousness alone, without the "traditional" prerequisites of a conviction or serious mental illness).

(38.) E.g., Stephen J. Morse, Preventive Confinement of Dangerous Offenders, 32 J. L. MED. & ETHICS 56, 56 (2004) (concluding "that pure preventive detention is more common than we usually assume, but that this practice violates fundamental assumptions concerning liberty under the American constitutional regime"); Tribe, supra note 22, at 371 (arguing that a proposal to detain "hard core recidivists" is misguided in light of "the dubious ability of pretrial preventive detention to contribute to the control of crime"); Alec Walen, A Unified Theory of Detention, with Application to Preventive Detention for Suspected Terrorists, 70 MD. L. REV. 871, 877 (2011) ("[A]n individual may not be deprived of his liberty unless the reasons for doing so respect his status as an autonomous person.").

(39.) E.g., Ronald J. Allen & Larry Laudan, Deadly Dilemmas III: Some Kind Words for Preventive Detention, 101 J. CRIM. L. & CRIMINOLOGY 781, 782 (2011) ("While we have no intention of defending all or even most forms of preventive detention in their concrete instantiations, we think that preventive detention is, under many circumstances, a legitimate and principled part of the criminal law."); Paul H. Robinson, Punishing Dangerousness: Cloaking Preventive Detention as Criminal Justice, 114 HARV. L. REV. 1429, 1456 (2001) (arguing that an explicit preventive regime would be preferable to current practice of "cloaking" prevention as punishment); Slobogin, supra note 37, at 122 (making the case for a pure preventive regime).

(40.) See infra notes 303-306 and accompanying text.

(41.) See, e.g., Haldane Robert Mayer, Preventive Detention and the Proposed Amendment to the Bail Reform Act of 1966, 11 WM. & MARY L. REV. 525, 529 (1969) ("Our pretrial bail laws have always had as their sole purpose the ensuring of the defendant's appearance at trial.").

(42.) Stack v. Boyle, 342 U.S. 1, 5 (1951) ("Since the function of bail is limited, the fixing of bail... must be based upon standards relevant to the purpose of assuring the presence of th[e] defendant."); Note, supra note 22, at 1489 ("In theory, the sole danger at which bail is aimed is the possibility of flight....").

(43.) See infra notes 54-56, 120, 307-311 and accompanying text.

(44.) See infra note 120 and accompanying text.

(45.) See infra notes 54-56, 307-311 and accompanying text.

(46.) See Act of Sept. 24, 1789, ch. 20, [section] 33, 1 Stat. 73, 91 (guaranteeing a right to bail in noncapital cases) (repealed 1984); JOHN S. GOLDKAMP, Two CLASSES OF ACCUSED: A STUDY OF BAIL AND DETENTION IN AMERICAN JUSTICE 55-59 (1979) (explaining "classic" state constitutional bail clause as excluding those charged with capital offenses); A. HIGHMORE, A DIGEST OF THE DOCTRINE OF BAIL: IN CIVIL AND CRIMINAL CASES vii, 194-96 (1783) (explaining that some are excluded from bail so that "the safety of the people should be preserved against the lawless depredations of atrocious offenders"); Mitchell, supra note 30, at 1225-26 & n.17 (interpreting colonial bail clauses).

(47.) See 4 WILLIAM BLACKSTONE, COMMENTARIES *294 ("[I]n... offences of a capital nature, no bail can be a security equivalent to the actual custody of the person. For what is there that a man may not be induced to forfeit, to save his own life?"); Tribe, supra note 22, at 401 (concluding that pretrial release was traditionally denied for offenses that "carried heavier penalties and therefore involved a greater temptation to flee," and to protect accused persons from vigilante justice).

(48.) I borrow the terminology of first-, second-, and third- generation or wave bail reform from Timothy R. Schnacke, Fundamentals of Bail: A Resource Guide for Pretrial Practitioners and a Framework for American Pretrial Reform, U.S. DEP'T JUST. 18 (Aug. 2014), http://static.nicic.gov/UserShared/2014-11-05_final_bail_fundamentals_september_8,_2014.pdf [http://perma.cc/R5TX-UWRK].

(49.) See, e.g., John S. Goldkamp, Danger and Detention: A Second Generation of Bail Reform, 76 J. CRIM. L. & CRIMINOLOGY 1, 2-3 (1985).

(50.) The movement culminated in the Bail Reform Act of 1966, which mandated that non-capital federal defendants be released with the least restrictive conditions that would "reasonably assure the appearance of the person" for trial. Pub. L. No. 89-465, [section] 3146(a), 80 Stat. 214, 214 (1966) (codified at 18 U.S.C. [section] 3146(a)) (repealed 1984).

(51.) Concerns about this practice were one motivation for first-wave reform. See, e.g., Hairston v. United States, 343 F.2d 313, 316 (D.C. Cir. 1965) (Bazelon, C.J., dissenting) ("Setting high bail to deny release discriminate[s] between the dangerous rich and the dangerous poor and masks the difficult problems of predicting future behavior..." (internal quotation marks and citation omitted)); BERNARD BOTEIN ET AL., NAT'L CONFERENCE ON BAIL & CRIMINAL JUSTICE, PROCEEDINGS AND INTERIM REPORT xxix (1965) ("A substantial body of opinion supports the view that setting high bail to detain dangerous offenders is unconstitutional.").

(52.) See H.R. REP. NO. 89-1541, at 6 (1966), as reprinted in 1966 US.C.CA.N. 2293, 2296 (noting that preventive detention for noncapital defendants was "beyond the scope of the present proposal and involves many difficult and complex problems which require deep study and analysis"); Marc Miller & Martin Guggenheim, Pretrial Detention and Punishment, 75 MINN. L. REV. 335, 344 (1990) (suggesting that the decline in the use of sub rosa detention may have encouraged legislators to develop a more "express ground for detention").

(53.) Goldkamp, supra note 49, at 1.

(54.) Id. at 15; Miller & Guggenheim, supra note 52, at 344-45.

(55.) Miller & Guggenheim, supra note 52, at 344-45.

(56.) District of Columbia Court Reform and Criminal Procedures Act of 1970, Pub. L. No. 91-358, 84 Stat. 473 (codified at D.C. Code Ann. [section][section] 23-1321 to -1323 (West 2017)); Bail Reform Act of 1984, Pub. L. No. 98-473, 98 Stat. 1976 (codified at 18 U.S.C. [section][section] 3141-50, 3062 (2012)).

(57.) See generally, e.g., Alan M. Dershowitz, Preventive Confinement: A Suggested Framework for Constitutional Analysis, 51 TEX. L. PvEV. 1277, 1278 (1973) (noting that "[l]eading scholars of the criminal law have... inveighed against preventive confinement"); Tribe, supra note 22, at 407 (concluding that preventive detention "violates the basic principle that an accusation of crime should not subject any man to imprisonment unless the government's need to prosecute him compels incarceration").

(58.) 481 U.S. 739 (1987).

(59.) Id. at 743.

(60.) Id. at 745-46; see U.S. CONST. amend. VIII ("Excessive bail shall not be required....").

(61.) United States v. Salerno, 794 F.2d 64, 71-72 (2d Cir. 1986), rev'd, 481 U.S. 739 (1987).

(62.) Salerno, 481 U.S. at 746-55.

(63.) Id. at 748, 753.

(64.) Id. at 747-48 (holding that the challenged detention regime did not categorically constitute impermissible punishment because "the incidents of detention" were not "excessive in relation to the regulatory goal Congress sought to achieve"); id. at 747 (explaining that detention regime applied only to those charged with "the most serious crimes" and included both procedural protections and a time limit); id. at 750-51 (finding that detention regime did not categorically violate substantive due process because it "narrowly focuses on a particularly acute problem in which the Government interests are overwhelming"); id. at 750 (explaining that regime operated "only on individuals who have been arrested for a specific category of extremely serious offenses" who posed a "demonstrable danger to the community," and that the regime was further limited by an array of procedural protections).

(65.) Id. at 751-52.

(66.) See, e.g., Laura I. Appleman, Justice in the Shadowlands: Pretrial Detention, Punishment, & the Sixth Amendment, 69 WASH. & LEE L. REV. 1297, 1349 (2012) ("Most practitioners and scholars have concluded that the Supreme Court's decision in Salerno pronounced the death knell for challenges to preventative detention."); Shima Baradaran, Restoring the Presumption of Innocence, 72 OHIO ST. L.J. 723, 725 (2011) ("While several legal scholars commented on bail and detention during the 1970s and 1980s, few contemporary legal scholars have analyzed the results of the changes in pretrial release standards and loss of the presumption of innocence."); Samuel R. Wiseman, Fixing Bail, 84 GEO. WASH. L. REV. 417, 420 (2016) ("Bail historically received wide scholarly attention, but this attention waned in the aftermath of United States v. Salerno...."); Chalmous G. Reemes, Case Note, United States v. Salerno: The Validation of Preventive Detention and the Denial of a Presumed Constitutional Right to Bail, 41 ARK. L. REV. 697,721(1988).

(67.) Salerno, 481 U.S. at 746-48; see also Bell v. Wolfish, 441 U.S. 520, 535 (1979) ("In evaluating the constitutionality of conditions or restrictions of pretrial detention... the proper inquiry is whether those conditions amount to punishment of the detainee.").

(68.) Salerno, 481 U.S. at 748-51.

(69.) Id. at 754.

(70.) Id. at 751-52.

(71.) See id. at 745 (noting that" [t]he fact that the Bail Reform Act might operate unconstitutionally under some conceivable set of circumstances is insufficient to render it wholly invalid"); id. at 751 (alluding to possibility that challenged procedures "might be insufficient in some particular circumstances"); see also Appleman, supra note 66, at 1349-51 (exploring how Salerno leaves open the possibility of bringing an as-applied challenge to the Bail Reform Act).

(72.) Salerno, 481 U.S. at 749.

(73.) Id. at 755.

74. Pretrial Release and Detention: The Bail Reform Act of 1984, BUREAU JUST. STAT. 1 (Feb. 1988), http://www.bjs.g0v/content/pub/pdf/prd-bra84.pdf [http://perma.cc/T5N8-F2AF].

(75.) Thomas H. Cohen, Pretrial Detention and Misconduct in Federal District Courts, 1995-2010, BUREAU JUST. STAT. 1, 2 (Feb. 2013), http://www.bjs.gov/content/pub/pdf/pdmfdc9510.pdf [http://perma.cc/3M9B-LTF8]. This percentage represents the number of defendants detained for the entire duration of a case.

(76.) Id.

(77.) See, e.g., United States v. Infelise, 934 F.2d 103, 104 (7th Cir. 1991) ("If judge and prosecutor are doing all they reasonably can be expected to do to move the case along, and the statutory criteria for pretrial detention are satisfied, then we do not think a defendant should be allowed to maintain a constitutional challenge to that detention."); United States v. Hare, 873 F.2d 796, 800 (5th Cir. 1989) ("We reject Hare's challenge under the excessive-bail clause of the Eighth Amendment as foreclosed by the Supreme Court's decision in United States v. Salerno."); United States v. Strong, 775 F.2d 504, 506 (3d Cir. 1985) (finding that "Congress intended to equate traffic in drugs with a danger to the community").

(78.) See Heaton, Mayson & Stevenson, supra note 4, at 782-83.

(79.) United States v. Salerno, 481 U.S. 739, 750 (1987).

(80.) Id.

(81.) See, e.g., State v. Anderson, 127 A.3d 100, 125 n.4 (Conn. 2015) (Palmer, J., dissenting) ("[I]t is undisputed that the trial court intentionally set a bond that far exceeded an amount that the defendant could pay solely to ensure that he would be incarcerated... due to his perceived dangerousness."); Baradaran & McIntyre, supra note 11, at 547 (concluding, on the basis of empirical analysis of defendants who remain detained on money bail, that "judges are basing their [bail] decisions far more on predicted violence than on predicted flight"); Wiseman, supra note 66, at 434 (noting that efforts to limit money bail "have met stiff, often successful resistance from the powerful bail bondsman lobby").

(82.) At midyear 2014 there were an estimated 467,500 people awaiting trial in local jails, up from 349,800 in 2000 and 298,100 in 1996. Darrell K. Gilliard & Allen J. Beck, Prison and Jail Inmates at Midyear 1996, BUREAU JUST. STAT. 7 (Jan. 1997), http://www.bjs.gov/content/pub/pdf/pjimy96.pdf [http://perma.cc/3NJA-PG2S]; Minton & Zeng, supra note 2, at 3. Between 1990 and 1994, 41% of pretrial releases were ROR and 24% were by cash bail. In 2002 and 2004, 23% of releases were ROR and 42% were by cash bail. Thomas H. Cohen & Brian A. Reaves, Pretrial Release of Felony Defendants in State Courts, U.S. DEP'T JUST. 2 (Nov. 2007), http://www.bjs.gov/content/pub/pdf/prfdsc.pdf [http://perma.cc/H2C7-LWX5]. As of 2015, financial conditions of release were imposed in 61% of criminal cases nationwide. Jessica Eaglin & Danyelle Solomon, Reducing Racial and Ethnic Disparities in Jails: Recommendations for Local Practice, BRENNAN CTR. FOR JUST. 19 (June 2015), http://www.brennancenter.org/sites/default/files/publications/Racial%20Disparities%20Report%20062515.pdf [http://perma.cc/84NR-YA54]. With respect to detention for inability to post bail, see Heaton, Mayson & Stevenson, supra note 4, at 736 tbl.1, which reports that 53% of misdemeanor defendants in Harris County, Texas are detained pretrial; Mary T. Phillips et al., Annual Report 201s, N.Y.C. CPJM. JUST. AGENCY 30 (Dec. 2014), http://www.nycja.org/lwdcms/doc-view.php?module=reports&module_id=1410&doc_narne=doc [http://perma.cc/G6V6-QJPX], which reports that 30% of felony defendants and 46% of non-felony defendants whose bail was $500 or less were detained until disposition; and Reaves, supra note 3, at 15, which reports that nine in ten felony defendants detained until disposition had bail set.

(83.) See, e.g., Christopher Moraff, U.S. Cities Are Looking for Alternatives to Cash Bail, NEXT CITY (Mar. 24, 2016), http://nextcity.org/daily/entry/cities-alternatives-cash-bail [http://perma.cc/R9LM-49DH]; Rational and Transparent Bail Decision Making: Moving From a Cash-Based to a Risk-Based Process, PRETRIAL JUST. INST. (Mar. 2012), http://www.pretrial.org/download/pji-reports/Rational%20and%20Transparent%20Bail%20Decision%20Making.pdf [http://perma.cc/G7MR-47CY].

(84.) See, e.g., Civil Rights Division, Dear Colleague Letter from Principal Deputy Assistant Att'y Gen. Vanita Gupta and Dir. Lisa Foster, U.S. Dep't Just. (Mar. 14, 2016) http://www.justice.gov/crt/file/832461/download [http://perma.cc/YM32-EZHH] (urging courts to "consider transitioning from a system based on secured monetary bail alone to one grounded in objective risk assessments by pretrial experts"); Pretrial Risk Assessment, PRETRIAL JUST. INST., http://pretrial.org/solutions/risk-assessment [http://perma.cc/G2TJ-6DZ8] ("Jurisdictions should have a pretrial services program or similar entity that conducts a risk assessment on all defendants in custody awaiting the initial appearance in court!').

(85.) E.g., John Monahan, A Jurisprudence of Risk Assessment: Forecasting Harm Among Prisoners, Predators, and Patients, 92 VA. L. REV. 391, 405-06 (2006) (defining "clinical" versus "actuarial" prediction); Barbara D. Underwood, Law and the Crystal Ball: Predicting Behavior with Statistical Inference and Individualized Judgment, 88 YALE L.J. 1408, 1423 (1979) ("A clinical decisionmaker is not committed in advance of decision to the factors that will be considered and the rule for combining them."). The clinical/actuarial dichotomy is not a clean one. Many "actuarial" assessments require determinations about traits like drug addiction, which require some subjective judgment.

(86.) For further explanation of this process, see generally Cynthia A. Mamalian, State of the Science of Pretrial Risk Assessment, PRETRIAL JUST. INST. (2011), http://www.pretrial.org/wpfb-file/pji-state-of-the-science-pretrial-risk-assessment-2011-pdf [http://perma.cc/R7ZW-6XG6]; and Pretrial Risk Assessment: Science Provides Guidance on Assessing Defendants, PRETRIAL JUST. INST. 3-4 (May 2015) [hereinafter Pretrial Risk Assessment], http://university.pretrial.org/viewdocument/issue-brief-pretrial-1 [http://perma.cc/67PJ-6SLX].

(87.) See, e.g., Mamalian, supra note 86, at 9.

(88.) That is, different factors may be "weighted" differently. See, e.g., Public Safety Assessment: Risk Factors and Formula, LAURA& JOHN ARNOLD FOUND. 3 (2016) [hereinafter LJAF, Public Safety], http://www.arnoldfoundation.org/wp-content/uploads/PSA-Risk-Factors-and-Formula.pdf [http://perma.cc/3CBE-5W37] (explaining weighting of different risk factors). Some instruments include a series of risk "scales" rather than binary yes-or-no factors. See, e.g., Thomas Blomberg et al., Validation of the COMPAS Risk Assessment Classification Instrument, CTR. FOR CRIMINOLOGY & PUB. POL'Y RES. 10 (2010), http://criminology.fsu.edu/wp-content/uploads/Validation-of-the-COMPAS-Risk-Assessment-Classification-Instrument.pdf [http://perma.cc/A4PL-DXGY] ("In total, the COMPAS includes 15 different factors, each of which is measured using multiple item scales.").

(89.) See, e.g., COMPAS Risk and Needs Assessment System: Selected Questions Posed by Inquiring Agencies, NORTHPOINTE INC. 9 (2012), http://www.northpointeinc.com/files/downloads/FAQ_Document.pdf [http://perma.cc/JR2A-5CZH] (discussing technical support for COMPAS software). The company behind the COMPAS, Northpointe Inc., seems to have rebranded itself as equivant and no longer advertises the COMPAS as a pretrial classification tool. North-pointe's former web address, www.northpointeinc.com (where it used to advertise the COMPAS suite software for multiple uses) now redirects visitors to equivant.com, which advertises the COMPAS classification system as a tool for jail/prison management only. See COMPAS Classification, EQUIVANT, http://www.equivant.com/solutions/inmate-classification [http://perma.cc/8PWZ-UP39] ("The COMPAS Classification software is a management support tool that can be interfaced with existing jail management systems to provide critical inmate management information to help support all aspects of a jail's classification operations.").

(90.) The PSA and COMPAS are the exceptions. They produce separate scores for flight and rearrest, and can also assess each person's risk of rearrest for a violent crime specifically. See Blomberg et al., supra note 88, at 18; LJAF, Developing a National Model, supra note 7, at 3-4.

(91.) Gouldin, Disentangling, supra note 12, at 842 (explaining that this aggregation "reinforces... judges' muddling of flight risk and dangerousness in the pretrial process").

(92.) Baradaran & McIntyre, supra note 11, at 547 (concluding on the basis of an empirical study that predictors of flight and of future violent crime "are almost completely uncorrelated"); Gouldin, Disentangling, supra note 12, at 893, 897 ("Risk assessment tools that generate a cumulative risk of pretrial failure have limited utility.").

(93.) Where Pretrial Improvements Are Happening, PRETRIAL JUST. INST. 4 (2017), http://university.pretrial.org/viewdocument/where-pretrial-improvements-are-hap-2 [http://perma.cc/G9FE-68AN] (reporting that the PSA in now in use throughout New Jersey, Kentucky, and Arizona and in thirty additional counties); id. at 14 (reporting that Delaware, Nevada, Washington, Hawaii, and three additional counties have recently received technical assistance to implement or improve pretrial risk assessment); sources cited infra note 96 (indicating that the federal courts and at least some counties in Colorado, Florida, Indiana, Ohio, and Virginia use a pretrial risk assessment tool); see also Jessica DaSilva, Hundreds of Jurisdictions Clamor for Pretrial Risk Test, BLOOMBERG BNA (Aug. 2, 2016), http://www.bna.com/hundreds-jurisdictions-clamor-n73014445751 [http://perma.cc/H9T4-RRMU].

(94.) Where Pretrial Improvements Are Happening, supra note 93 (describing pretrial reform, including shift toward risk assessment, in jurisdictions nationwide); Shaila Dewan, Judges Replacing Conjecture with Formula for Bail, N.Y. TIMES (June 26, 2015), http://www.nytimes.com/2015/06/27/us/turning-the-granting-of-bail-into-a-science.html [http://perma.cc/4HA5-BFGL]; Press Release, Laura & John Arnold Found., More Than 20 Cities and States Adopt Risk Assessment Tool To Help Judges Decide Which Defendants To Detain Prior to Trial (June 26, 2015), http://www.arnoldfoundation.org/more-than-20-cities-and-states-adopt-risk-assessment-tool-to-help-judges-decide-which-defendants-to-detain-prior-to-trial [http://perma.cc/P8QA-FHH8].

(95.) LJAF, Developing a National Model, supra note 7, at 4-5. There are at least six other tools in current use. They include the Federal Pretrial Risk Assessment (PTRA), Colorado Pretrial Risk Assessment Tool (CPAT), Florida Pretrial Risk Assessment Instrument (FL PRAI), Indiana Risk Assessment System-Pretrial Assessment Tool (IRAS-PAT), Ohio Risk Assessment System-Pretrial Assessment Tool (ORAS -PAT), Virginia Pretrial Risk Assessment Instrument (VPRAI), and the Correctional Offender Management Profile for Alternative Sanctions (COMPAS). See infra note 96 and Table 1.

(96.) This chart expands on a model created by the Pretrial Justice Institute, see Pretrial Risk Assessment, supra note 86, at 3, but draws directly from the following sources: Austin et al., supra note 15; Blomberg et al., supra note 88, at 15-16; The Colorado Pretrial Assessment Tool (CPAT), PRETRIAL JUST. INST. 15-18 (Oct. 19, 2012) [hereinafter Colorado Pretrial Assessment Tool], http://www.pretrial.org/download/risk-assessment/CO%20Pretrial%20Assessment%20Tool%20Report%20Rev%20-%20PJI%202012.pdf [http://perma.cc/6ZCR-F7AY]; Federal PTRA, supra note 14; Indiana Risk Assessment System, UNIV. OF CIN. at 1-1 to 1-2 (Apr. 23, 2010) [hereinafter Indiana Risk Assessment System], http://www.pretrial.org/download/risk-assessment/Indiana%2oRisk%20Assessment%20System%20(April%202010).pdf [http://perma.cc/K39L-QUAH]; Edward Latessa et al., Creation and Validation of the Ohio Risk Assessment System: Final Report, UNIV. OF CIN. 49-50 (July 2009) [hereinafter ORAS Final Report], http://www.pretrial.org/download/risk-assessment/Ohio%20Pretrial%20Risk%20Assessment%202009.pdf [http://perma.cc/JK6L-659Q]; LJAF, Developing a National Model, supra note 7, at 3-4; LJAF, Public Safety, supra note 88; LJAF, Results, supra note 16, at 3; MARIE VANNOSTRAND, VA. DEP'T OF CRIMINAL JUSTICE SERVS., ASSESSING RISK AMONG PRETRIAL DEFENDANTS IN VIRGINIA: THE VIRGINIA PRETRIAL RISK ASSESSMENT INSTRUMENT 6 (2003); Marie VanNostrand & Kenneth J. Rose, Pretrial Risk Assessment in Virginia, LUMINOSITY INC. 2 (May 1, 2009), http://www.pretrial.org/download/risk-assessment/VA%20Risk%20Report%202009.pdf [http://perma.cc/E68P-JC95]; Timothy P. Cadigan et al., The Re-Validation of the Federal Pretrial Services Risk Assessment (PTRA), 76 FED. PROB. 3 (2012); Timothy P. Cadigan & Christopher T Lowenkamp, Implementing Risk Assessment in the Federal Pretrial Services System, 75 FED. PROB. 30, 32 (2011); Edward J. Latessa et al., The Creation and Validation of the Ohio Risk Assessment System (ORAS), 74 FED. PROB. 16 (2010) [hereinafter ORAS Creation and Validation]; Christopher T. Lowenkamp & Jay Whetzel, The Development of an Actuarial Risk Assessment Instrument for U.S. Pretrial Services, 73 FED. PROB. 33 (2009); and Angwin et al., supra note 19.

(97.) At its core, a machine-learning risk assessment process is simply a more powerful and complex version of a scoring instrument. See generally Richard A. Berk et al., Forecasting Domestic Violence: A Machine Learning Approach To Help Inform Arraignment Decisions, 13 J. EMPIRICAL LEGAL STUD. 94 (2016) (finding that if magistrates used machine-learning methods, the rate of rearrest for those released after an arraignment for domestic violence would be cut nearly in half); Richard Berk & Jordan Hyatt, Machine Learning Forecasts of Risk To Inform Sentencing Decisions, 27 FED. SENT'G REP. 222 (2015) (summarizing the results of a machine-learning approach using "random forests").

(98.) Tools use anywhere between three and six risk classes. See, e.g., LJAF, Developing a National Model, supra note 7, at 4-5; Pretrial Risk Assessment, supra note 86, at 4. Tool developers must decide where, along the curve of rearrest probability, to draw the lines between risk classes. These are called "cut points." See, e.g., Eaglin, supra note 19, at 87-88 (explaining cut points, using the alternate phrase "cut-off points"); COMPAS Decile Cut Points Norming, ELECTRONIC PRIVACY INFO. CTR., https://epic.org/algorithmic-transparency/crim-justice/EPIC-16-06-23-WI-FOIA-201600805-DecileCutPointsNormingo20216.pdf [http://perma.cc/63PS-8YV6].

(99.) LJAF, Developing a National Model, supra note 7, at 4-5; Mamalian, supra note 86, at 10.

(100.) See, e.g., Cadigan et al., supra note 96.

(101.) In the machine-learning context, these rates of the predicted harm would be called "forecasting accuracy." See Berk & Hyatt, supra note 97, at 224. Note that if an algorithm is applied to a different population or outcomes are monitored over a different time period, the rate will likely change.

(102.) Federal PTRA, supra note 14, at 4; Cadigan & Lowenkamp, supra note 96, at 32.

(103.) Austin et al., supra note 15, at 4, 11 (showing rate of rearrest just over 15%).

(104.) Indiana Risk Assessment System, supra note 96, at 1-1. There is no public documentation of the measurement period.

(105.) VanNostrand & Rose, supra note 96, at 7-8, 12.

(106.) LJAF, Results, supra note 16, at 1, 3.

(107.) ORAS Final Report, supra note 96, at 14, 16; ORAS Creation and Validation, supra note 96, at 19. There is no available data on rearrest rates alone.

(108.) Colorado Pretrial Assessment Tool, supra note 96, at 9 & n.14, 10 & n.15,15-18, 18 n.23.

(109.) Blomberg et al., supra note 88, at 35.

(110.) Id. at 51.

(111.) LJAF, Results, supra note 16, at 3.

(112.) A team led by Richard Berk developed a machine-learned algorithm for use at the arraignment of people charged with domestic violence offenses to forecast rearrest specifically for domestic violence (particularly on a charge involving physical injury). See Berk et al., supra note 97, at 103-04.

(113.) Colorado Pretrial Assessment Tool, supra note 96, at 18 n.23 ("The public safety rate for the CPAT study... was defined very broadly as a filing for any new felony, misdemeanor, traffic, municipal, and petty offense....").

(114.) The information in the table above represents the extent of the definition of the "new arrest" outcome measured in the cited studies. Few of the studies, for instance, specify whether "new arrest" includes or excludes arrest for municipal or traffic offenses.

(115.) See, e.g., LJAF, Developing a National Model, supra note 7, at 5; Mamalian, supra note 86, at 2, 5.

(116.) CRIMINAL JUSTICE SECTION STANDARDS: PRETRIAL RELEASE Standard 10-5.3(b) (AM. BAR ASS'N 2002) ("Financial conditions of release should not be set to prevent future criminal conduct during the pretrial period or to protect the safety of the community or any person.").

(117.) See, e.g., LJAF, Developing a National Model, supra note 7, at 5; Mamalian, supra note 86, at 2; Cadigan et al., supra note 96, at 4-6 (explaining that low-risk defendants should not be "over-supervised," but that high-risk defendants should generally be detained).

(118.) See, e.g., Wiseman, supra note 66, at 422, 454-477 (arguing that judicial discretion presents a "principal-agent problem that must be addressed if we are to fix the bail system," and proposing mandatory bail guidelines that rely on actuarial risk assessment).

(119.) These states retain "the traditional state constitutional approach" to bail, which guarantees a right to bail except in capital or extremely serious cases when "the proof is evident, or the presumption great." 4 WAYNE R. LAFAVE ET AL., CRIMINAL PROCEDURE [section] 12.3(b), at 55 (4th ed. 2015) (internal quotation marks omitted). Seventeen states guarantee bail except in capital cases; the other six extend the exception to treason, murder, rape, or offenses punishable by life imprisonment. Id. at 55-56. Of the remaining twenty-seven states, nine prohibit "excessive bail" without stating whether bail may be denied altogether. Id. And eighteen already have constitutional provisions authorizing pretrial preventive detention, although the provisions range widely. Id.

(120.) See Goldkamp, supra note 49, at 24, 57-74 (chronicling "great variation" in state constitutional and statutory bail provisions); Gouldin, Disentangling, supra note 12, at 866 (noting that state statutes direct bail courts to consider various factors, but "do not indicate which factors are relevant to flight risk and which are believed to predict dangerousness"); id. at 882-85 (observing that "many statutes do a poor job of guiding judges about which risks are relevant to different pretrial decisions," and providing examples). New York is often said to exclude consideration of dangerousness, but it does authorize courts to issue protective orders as conditions of release, and other aspects of its pretrial law also suggest otherwise. See N.Y. CRIMINAL PROCEDURE LAW [section] 510.30(2)(a) (McKinney 2017) (describing that considerations for pretrial commitment without bail include past criminal record and firearm use); id. [section][section] 530.12-14 (authorizing orders of protection); id. [section] 530.20(2)(a) (prohibiting recognizance or bail when a defendant is charged with a class A felony or has two previous felony convictions). Many states either permit or require courts to consider dangerousness in setting money bail, which contravenes reform principles. See supra note 116.

(121.) In fact, reforms are moving forward so fast that it is difficult to keep up. For a helpful overview of the landscape, see Where Pretrial Improvements Are Happening, PRETRIAL JUST. INST. (July 2017), http://university.pretrial.org/HigherLogic/System/DownloadDocumentFile.ashx?DocumentFileKey=f3d14ge5-aooc-7022-3e6b-97doaf428010 [http://perma.cc/3WRN-HHM3]. Kentucky, Colorado, and New Jersey undertook comprehensive reform--including amendment of the New Jersey Constitution. See Public Safety and Offender Accountability Act, 2011 Ky. Acts 4; Pretrial Servs., Pretrial Reform in Kentucky, ADMIN. OFF. CTS. KY. CT. JUST. 5-8 (Jan. 2013), http://www.pretrial.org/download/infostop/Pretrial%20Reform%20in%20Kentucky%20Implementation%20Guide%202013.pdf [http://perma.cc/4FMW-TM3U]; Timothy R. Schnacke, Best Practices in Bond Setting: Colorado's New Pretrial Bail Law, CTR. FOR LEGAL & EVIDENCE-BASED PRACTICES (July 3, 2013), http://www.pretrial.org/download/lawpolicy/Best%20Practices%20in%20Bond%20Setting%20-%20Colorado.pdf [http://perma.cc/6LMK-PULB]; Matt Arco, Christie Signs Bail Reform Measure, Lauds Lawmakers for Bipartisanship, NJ.COM (Aug. 11, 2014, 3:34 PM), http://www.nj.com/politics/index.ssf/2014/08/christie_signs_bail_reform_measure_lauds_lawmakers_for_bipartisanship.html [http://perma.cc/X8PA-6WGA]. Alabama, Alaska, California, Connecticut, Delaware, Illinois, Indiana, Maryland, New Mexico, New York, Nevada, and Ohio are actively pursuing reform. See, e.g., Final Report, COMMISSION To REFORM MD.'S PRETRIAL SYS. (Dec. 19, 2014), http://goccp.maryland.gov/pretrial/documents/2014-pretrial-commission-final-report.pdf [http://perma.cc/H2DG-KYA2]; 13 Places that Saw Bail Reform in 2016, CLEVELAND.COM (Dec. 13, 2016), http://www.cleveland.com/metro/index.ssf/2016/12/13_places_that_saw_bail_reform.html [http://perma.cc/425X-6A7W]; Thomas Cole, Lawmakers Give Final OK to Bail Reform Measure, ALBUQUERQUE J. (Feb. 17, 2016, 12:09 PM), http://www.abqjournal.com/725486/lawmakers-give-final-ok-to-bail-reform-measure.html [http://perma.cc/J3EB-EA5H]; Connecticut Moves To Reform Cash Bail System for Low-Level Defendants, BETWEEN LINES (Feb. 3, 2016), http://www.btlonline.org/2016/seg/160212bf-btl-mcguire.html [http://perma.cc/66JW-7CH5]; Christopher Connelly, Texas Lawmakers, fudges Push for Major Bail Reform, Hous. PUB. MEDIA (Mar. 10, 2017, 12:31 PM), http://www.houstonpublicmedia.org/articIes/news/2017/03/10/191236/texaslawmakers -judges-push-for-major-bail-reform [http://perma.cc/L4TK-7RH7]; Sara Dorn, Developing 'Fair' Bail System Goal of Ohio Panel: Impact 2016: Justice for All, CLEVELAND.COM (May 18, 2016, 4:40 PM), http://www.cleveland.com/metro/index.ssf/2016/05/developing_fair_bail_system_go.html [http://perma.cc/KM74-6RNB]; Geoff Dornan, Feds Agree To Help Study and Pay for Nevada Bail Reform Project, NEV. APPEAL (Feb. 11,2016), http://www.nevadaappeal.com/news/government/feds-agree-to-help-study-and-pay-for-nevada-bail-reform-project [http://perma.cc/65BP-RF3L]; Editorial: Reforming Bail Bonds a Must for a Connecticut Second Chance Society, NEW HAVEN REG. (Feb. 6, 2016, 10:09 PM), http://www.nhregister.com/opinion/20160206/editorial-reformirig-bail-bonds-a-must-for-a-connecticut-second-chance-society [http://perma.cc/8MT6-FS48]; Kim Geiger, Bail Reform Act Signed in Illinois, GOVERNING STATES & LOCALITIES (June 13, 2017), http://www.governing.com/topics/public-justice-safety/Bail-Reform-Enacted-in-Illinois.html [http://perma.cc/ZR8G-PYUE]; James C. McKinley, Jr., State's Chief judge, Citing 'Injustice,' Lays Out Plans To Alter Bail System, N.Y. TIMES (Oct. 1, 2015), http://www.nytimes.com/2015/10/02/nyregion/jonathan-Iippman-bail-incarceration-new-york-state-chief-judge.html [http://perma.cc/XEL2-TB5H]; Press Release, Pretrial Justice Inst., PJI Commends Alaska's Criminal Justice Reforms (July 11, 2016), http://www.pretrial.org/pji-commends-alaskas-criminal-justice-reforms [http://perma.cc/2343-AMEV]; Smart Pretrial Reform in Delaware, PUB. WELFARE FOUND. (Apr. 22, 2015), http://www.publicwelfare.org/smart-pretrial-reform-in-delaware [http://perma.cc/P7Y8-3P5U].

(122.) Litigation is spurring considerable change. See Statement of Interest of the United States, Varden v. City of Clanton, No. 2:15-cv-34-MHT-WC (M.D. Ala. Feb. 13, 2015); Ending the American Money Bail System, EQUAL JUST. UNDER L. (2014), http://www.equaljusticeunderlaw.org/wp/current-cases/ending-the-american-money-bail-systern [http://perma.cc/RSW8-HY6X]; Moraff, supra note 83. At the national level, both Congress and the White House have gotten involved. See Pretrial Integrity and Safety Act of 2017, S. 1593, 115th Cong. (2017); No Money Bail Act of 2016, H.R. 4611, 114th Cong. (2016); Fact Sheet on White House and Justice Department Convening--A Cycle of Incarceration: Prison, Debt and Bail Practices, U.S. DEP'T JUST. (Dec. 3, 2015), http://www.justice.gov/opa/pr/fact-sheet-white-house-and-justice-department-convening-cycle-incarceration-prison-debt-and [http://perma.cc/GX2K-KGP4].

(123.) Many hold that the purely preventive restraint of responsible actors is never permissible, no matter the risk they pose, because such restraint denies a person's autonomy. See sources cited supra notes 33, 38; see also, e.g., Kimberly Kessler Ferzan, Beyond Crime and Commitment: Justifying Liberty Deprivations of the Dangerous and Responsible, 96 MINN. L. REV. 141, 141-42 (2011) (describing this orthodoxy). Other scholars accept that the purely preventive restraint of responsible actors may be permissible in some circumstances, but that substantial preventive intrusions could only be permissible to mitigate serious risk. See, e.g., Sandra G. Mayson, Collateral Consequences and the Preventive State, 91 NOTRE DAME L. REV. 301, 348 (2015) (arguing that any preventive restraint should be "an appropriately tailored means of preventing future harm, given its cost in liberty"); Christopher Slobogin, A Jurisprudence of Dangerousness, 98 Nw. U. L. REV. 1, 4 (2003) ("The proportionality principle requires that the degree of danger be roughly proportionate to the proposed government intervention.").

(124.) But see Rachel A. Harmon, Why Arrest?, 115 MICH. L. REV. 307 (2016) (arguing that custodial arrest as a means to initiate criminal proceedings may be justified far less frequently than people assume).

(125.) See, e.g., 18 U.S.C. [section] 3144 (2012) (providing for the arrest and detention of material witnesses when "it is shown that it may become impracticable to secure the presence of the person by subpoena").

(126.) See infra Section III.B.

(127.) These numbers depend on the jurisdiction and the tool. See supra Table 2.

(128.) Many statutes, of course, currently authorize broad pretrial preventive restraint. But statutory law is subject to revision--particularly now, as jurisdictions rewrite their bail laws. The more meaningful positive inquiry, therefore, is what the Constitution says in this context.

(129.) The Court's pretrial seizure case law is relevant too, and both Gerstein v. Pugh, 420 U.S. 103 (1975), and Manuel v. City of Joliet, 137 S. Ct. 911 (2017), can also be categorized as Fourth Amendment pretrial seizure cases. For simplicity, I limit the discussion to detention and search cases.

(130.) 420 U.S. at 114 ("[W]e hold that the Fourth Amendment requires a judicial determination of probable cause as a prerequisite to extended restraint of liberty following arrest."). If a judicial determination of probable cause were sufficient to justify any pretrial restraint, Salerno would have been an easy case indeed! Accord Albert W. Alschuler, Preventive Pretrial Detention and the Failure of Interest-Balancing Approaches to Due Process, 85 MICH. L. REV. 510, 560 (1986) ("Gerstein did not hold that defendants could be imprisoned to protect the community from crime simply because judges had found probable cause for their arrests.").

(131.) 137 S. Ct. at 913.

(132.) United States v. Salerno, 481 U.S. 739, 755 (1987) ("In our society liberty is the norm, and detention prior to trial or without trial is the carefully limited exception.").

(133.) Id.

(134.) Many thanks to Alec Walen and the editors of the Yale Law Journal for raising this possibility.

(135.) See, e.g., Kansas v. Hendricks, 521 U.S. 346 (1997).

(136.) Id. at 357.

(137.) Gerstein v. Pugh, 420 U.S. 103, 111 (1975) (defining probable cause as "facts and circumstances 'sufficient to warrant a prudent man in believing that the [suspect] had committed or was committing an offense'" (alteration in original) (quoting Beck v. Ohio, 379 U.S. 89, 91 (1964))). Gerstein does not explicitly specify that probable cause for an already adjudicated charge is inadequate to support a new arrest, but that is obvious from the language and context of the decision.

(138.) One of the Gerstein respondents was detained because he could not afford to post bail; the other was "denied bail because one of the charges against him carried a potential life sentence." Id. at 105. It is unclear why Manuel was detained. Manuel v. City of Joliet, 137 S. Ct. 911, 915 (2017).

(139.) 481 U.S. 739 (1987).

(140.) See supra notes 57-80 and accompanying text.

(141.) 481 U.S. at 741-55.

(142.) Id. at 750.

(143.) Id. at 748.

(144.) Cf. United States v. Scott, 450 F.3d 863, 870 (gth Cir. 2006) (noting that "the government's interest in preventing crime by anyone is legitimate and compelling" (citing United States v. Restrepo, 946 F.2d 654, 674 (9th Cir. 1991) (en banc) (Norris, J., dissenting))).

(145.) 481 U.S. at 749 (emphasis added).

(146.) Id. at 748-49 (citations omitted).

(147.) Kansas v. Hendricks, 521 U.S. 346, 358 (1997) (emphasis added); see also Kansas v. Crane, 534 U.S. 407, 412 (2002) (emphasizing the "constitutional importance" of the control criterion); cf. Foucha v. Louisiana, 504 U.S. 71 (1992) (holding that substantive due process prohibited the continued confinement of an insanity acquitee who remained dangerous but who no longer suffered from any mental disorder).

(148.) It is often impossible to distinguish between someone who cannot control her impulses and someone who does not. The Court itself acknowledged this in Kansas v. Crane. 534 U.S. at 412 ("The line between an irresistible impulse and an impulse not resisted is probably no sharper than that between twilight and dusk." (quoting the American Psychiatric Association, Statement on the Insanity Defense 11 (1982), reprinted in G. MELTON, J. PETRILA, N. POYTHRESS, & C. SLOBOGIN, PSYCHOLOGICAL EVALUATIONS FOR THE COURTS 200 (2d ed. 1997))). The Court therefore clarified that a civil commitment regime need not require proof "of total or complete lack of control.... It is enough to say that there must be proof of serious difficulty in controlling behavior." Id. at 411-13. But most situations where a person poses demonstrable danger will entail evidence of "serious difficulty in controlling behavior." Id. at 413.

(149.) 504 U.S. 71 (1992).

(150.) Id. at 80-83.

(151.) Id. at 81 ("[T]he duration of confinement under the Bail Reform Act of 1984 (Act) was strictly limited...."); id. at 82 ("It was emphasized in Salerno that the detention we found constitutionally permissible was strictly limited in duration." (citing Salerno, 481 U.S. at 747 (1987) ("The arrestee is entitled to a prompt detention hearing..., and the maximum length of pretrial detention is limited by the stringent time limitations of the Speedy Trial Act."))).

(152.) Id. at 82 ("Here, in contrast, the State asserts that... [Foucha] may be held indefinitely."); id. at 102 (Thomas, J., dissenting) ("[T]he Court suggests--and the concurrence states explicitly--that the constitutional flaw with this scheme is not that it provides for the confinement of sane insanity acquittees, but that it (allegedly) provides for their 'indefinite' confinement in a mental facility.").

(153.) Id. at 81 ("Unlike the sharply focused scheme at issue in Salerno, the Louisiana scheme of confinement is not carefully limited."); id. at 81-84 (comparing Salerno and Louisiana regimes).

(154.) Id. at 87 (O'Connor, J., concurring) ("I do not understand the Court to hold that Louisiana may never confine dangerous insanity acquittees after they regain mental health."); id. (opining that such confinement might be permissible if "the nature and duration of detention were tailored to reflect pressing public safety concerns related to the acquittee's continuing dangerousness.").

(155.) Id. at 84.

(156.) See, e.g., United States v. Robinson, 414 U.S. 218, 235 (1973) ("[I]n the case of a lawful custodial arrest a full search of the person is ... a 'reasonable' search under [the Fourth] Amendment."); Chimel v. California, 395 U.S. 752, 763 (1969) ("[I]t is reasonable for the arresting officer to search the person arrested in order to remove any weapons....").

(157.) Arizona v. Gant, 556 U.S. 332, 351 (2009) (holding that the Fourth Amendment permits such searches "only if the arrestee is within reaching distance of the passenger compartment at the time of the search or it is reasonable to believe the vehicle contains evidence of the offense of arrest").

(158.) Maryland v. Buie, 494 U.S. 325, 337(1990) ("The Fourth Amendment permits a properly limited protective sweep in conjunction with an in-home arrest when the searching officer possesses a reasonable belief... that the area to be swept harbors an individual posing a danger to those on the arrest scene").

(159.) See, e.g., Hiibel v. Sixth Judicial Dist. Court, 542 U.S. 177, 191 (2004) (holding that, in general, "[a]nswering a request to disclose a name is likely to be so insignificant in the scheme of things as to be incriminating only in unusual circumstances").

(160.) See, e.g., Colorado v. Bertine, 479 U.S. 367, 369 (1987) ("We are asked to decide whether the Fourth Amendment prohibits the State from proving [the criminal] charges with the evidence discovered during the inventory search of respondent's van. We hold that it does not."); United States v. Edwards, 415 U.S. 800, 807 (1974) (holding that "once an accused has been lawfully arrested and is in custody, the effects in his possession at the place of detention that were subject to search at the time and place of arrest may lawfully be searched and seized without a warrant").

(161.) Maryland v. King, 133 S. Ct. 1958, 1980 (2013) (holding that "taking and analyzing a cheek swab of the arrestee's DNA is... a legitimate police booking procedure that is reasonable under the Fourth Amendment").

(162.) See, e.g., Florence v. Bd. of Chosen Freeholders, 132 S. Ct. 1510, 1523 (2012) (holding that certain search procedures, which included strip searches, were constitutionally reasonable); Bell v. Wolfish, 441 U.S. 520, 558-60 (1979) (upholding the constitutionality of strip searches of inmates); Edwards, 415 U.S. at 803 ("[Sjearches and seizures that could be made on the spot at the time of arrest may legally be conducted later when the accused arrives at the place of detention.").

(163.) Riley v. California, 134 S. Ct. 2473, 2488 (2014) (quoting King, 133 S. Ct. at 1969,1979); King, 133 S. Ct. at 1969, 1979 (quoting Illinois v. McArthur, 531 U.S. 326, 330 (2001)).

(164.) 133 S. Ct. at 1969-70; see also Samson v. California, 547 U.S. 843, 850-57 (2006) (upholding a warrantless search of a parolee's person); United States v. Knights, 534 U.S. 112, 119-22 (2001) (upholding a warrantless search of a probationer's home).

(165.) See, e.g., Samson, 547 U.S. at 850-57 (discussing parolees); Knights, 534 U.S. at 119-22 (discussing probationers); Vernonia Sch. Dist. 47J v.Acton, 515 U.S. 646, 654-56 (1995) (discussing public-school students); Schall v. Martin, 467 U.S. 253, 265 (1984) (discussing juveniles); Bell, 441 U.S. at 546 (discussing detainees); see also Barry Friedman & Cynthia Benin Stein, Redefining What's "Reasonable": The Protections for Policing, 84 GEO. WASH. L. REV. 281, 349 (2016) (discussing cases where the Court justified "suspicion-based or suspicionless" searches of people in certain circumstances "primarily by arguing that people in schools, prisons, and government workplaces have reduced expectations of privacy").

(166.) See sources cited supra note 165.

(167.) 133 S. Ct. at 1978 ("The expectations of privacy of an individual taken into police custody 'necessarily [are] of a diminished scope.'") (quoting and citing Bell, 441 U.S. at 557); id. at 1979 (asserting that the Fourth Amendment reasonableness inquiry "considers" the "diminished expectations of privacy of the arrestee").

(168.) Nor has it received much attention from scholars or lower courts. The Ninth Circuit has confronted the question but without much clarity. Compare United States v. Scott, 450 F.3d 863, 872-74 (9th Cir. 2006) (noting that "[p]eople released pending trial, [in] contrast [to probationers], have suffered no judicial abridgment of their constitutional rights," and thus their "privacy and liberty interests [a]re far greater than a probationer's") with Scott, 450 F.3d at 885 (Bybee, J., dissenting) ("Scott's reasonable expectation of privacy may be somewhat greater than that of a probationer, parolee, or pre-sentence releasee, but it is less than that of an 'ordinary citizen.'"). Only one academic work squarely addresses this point. See Andrew J. Smith, Note, Unconstitutional Conditional Release: A Pyrrhic Victory for Arrestees' Privacy Rights Under United States v. Scott, 48 WM. & MARY L. REV. 2365, 2389 (2007) ("The lack of a clear definition of a pretrial arrestee's status is a lacuna in criminal law that must be filled."). The Court has established that detainees' rights are necessarily limited by virtue of being detained. Bell, 441 U.S. at 546. At the same time, pretrial detainees retain some rights that convicted detainees do not. Kingsley v. Hendrickson, 135 S. Ct. 2466, 2475 (2015) ("[P]retrial detainees (unlike convicted prisoners) cannot be punished at all....").

(169.) See Friedman & Stein, supra note 165, at 349-53.

(170.) Id.

(171.) See Schall v. Martin, 467 U.S. 253, 265 (1984) ("[Children] are assumed to be subject to the control of their parents, and if parental control falters, the State must play its part as parens patriae."). The quasi-custody rationale is more attenuated for people serving noncustodial sentences, but still plausible if release on probation or parole is a privilege the state grants in lieu of lawful detention, rather than a right. See, e.g., Morrissey v. Brewer, 408 U.S. 471, 477 (1972) ("The essence of parole is release from prison, before the completion of sentence, on the condition that the prisoner abide by certain rules during the balance of the sentence.").

(172.) See, e.g., Samson v. California, 547 U.S. 843, 848-49 (2006) ("[B]y virtue of their status alone, probationers 'do not enjoy "the absolute liberty to which every citizen is entitled,'" justifying the 'impos[ition] [of] reasonable conditions that deprive the offender of some freedoms enjoyed by law-abiding citizens.'" (citations omitted) (quoting United States v. Knights, 534 U.S. 112, 119 (2001))).

(173.) Maryland v. Buie, 494 U.S. 325, 327 (1990); Chimel v. California, 395 U.S. 752, 763 (1969).

(174.) Arizona v. Gant, 556 U.S. 332, 344 (2009); Chimel, 395 U.S. at 752.

(175.) Colorado v. Bertine, 479 U.S. 367, 373 (1987).

(176.) Florence v. Burlington, 132 S. Ct. 1510, 1518 (2012); Bell v. Wolfish, 441 U.S. 520, 546, 547 (1979).

(177.) Maryland v. King, 133 S. Ct. 1958, 1970 (2013).

(178.) See, e.g., Gant, 556 U.S. at 344, 351 (holding that warrantless automobile searches violate the Fourth Amendment when not necessary to protect officer safety or preserve evidence of the crime of arrest); Knowles v. Iowa, 525 U.S. 113, 118-19 (1998) (rejecting warrantless search pursuant to citation, "where the concern for officer safety is not present to the same extent and the concern for destruction or loss of evidence is not present at all"); Mincey v. Arizona, 437 U.S. 385, 390-93 (1978) (holding warrantless search of murder defendant's home that was not "justified by any emergency threatening life or limb" to violate the Fourth Amendment); id. at 391 ("It is one thing to say that one who is legally taken into police custody has a lessened right of privacy in his person. It is quite another to argue that he also has a lessened right of privacy in his entire house." (citations omitted)).

(179.) The distinction is usually credited to Judith Jarvis Thomson. See, e.g., JUDITH JARVIS THOMSON, THE REALM OF RIGHTS (1990); Judith Jarvis Thomson, A Defense of Abortion, 1 PHIL. & PUB. AFF. 47 (exploring the distinction between permissible and impermissible killing).

(180.) Mitchell N. Berman, Punishment and Justification, 118 ETHICS 258, 272 (2008).

(181.) Id. at 273.

(182.) King, 133 S. Ct. at 1970-77.

(183.) Riley v. California, 134 S. Ct. 2473, 2492-94 (2014) (identifying three permissible purposes: learning of imminent threats, obtaining evidence of the crime charged, and preventing its destruction).

(184.) King, 133 S. Ct. at 1970; see also id. at 1982 (Scalia, J., dissenting) ("The Court alludes at several points to the fact that King was an arrestee.... But the Court does not really rest on this principle, and for good reason.... Sensing (correctly) that it needs more, the Court elaborates at length the ways that the search here served the special purpose of 'identifying' King." (footnote omitted)).

(185.) Id. at 1971-74.

(186.) Id. at 1977 (quoting United States v. Kelly, 55 F.2d 67, 69 (2d Cir. 1932)).

(187.) Id. at 1980-90 (Scalia, J., dissenting); id. at 1986 ("[I]t is safe to say that if the Court's identification theory is not wrong, there is no such thing as error.").

(188.) See, e.g., BARRY FRIEDMAN, UNWARRANTED: POLICING WITHOUT PERMISSION 274 (2017) ("The decision in King is built on a lie.").

(189.) King, 133 S. Ct. at 1982-85 (Scalia, J., dissenting).

(190.) Id. at 1982.

(191.) Id. at 1980.

(192.) Id. at 1982 n.1.

(193.) This is true notwithstanding Justice Kennedy's assertion that the Court's "special needs cases... do not have a direct bearing on the issues presented in this case." Id. at 1978 (majority opinion).

(194.) Riley v. California, 134 S. Ct. 2473, 2480-82 (2014).

(195.) Id. at 2484-85; see also id. at 2485-88 (analyzing officer-safety and loss-of-evidence concerns).

(196.) King, 133 S. Ct. at 1978 (internal alterations, citation, and quotation marks omitted).

(197.) If subjective expectations could justify state action, the status quo would be self-justifying. Thus, the Court has acknowledged that the "reasonable expectations" prong of the traditional Katz Fourth Amendment analysis turns on what we think people should expect and tolerate, not what they actually do. See, e.g., Hudson v. Palmer, 468 U.S. 517, 525 n.7 (1984) ("[Constitutional rights are generally not defined by the subjective intent of those asserting the rights. The problems inherent in such a standard are self-evident." (citation omitted)); Katz v. United States, 389 U.S. 347, 361 (1967) (Harlan, J., concurring) (explaining that subjective expectation must "be one that society is prepared to recognize as 'reasonable'"); Orin S. Kerr, Katz Has Only One Step: The Irrelevance of Subjective Expectations, 82 U. CHI. L. REV. 113, 114-15 (2015) (arguing that the subjective-expectations prong of the Katz test is a "phantom doctrine" that the Court should "formally abolish").

(198.) Cf. Kerr, supra note 197 (making a similar argument about the Katz test); Antoine McNamara, Note, The "Special Needs" of Prison, Probation, and Parole, 82 N.Y.U. L. REV. 209 (2007) (making a similar argument with respect to penal contexts).

(199.) Cf. Chimel v. California, 395 U.S. 752, 766 n.12 (1969) ("[W]e can see no reason why, simply because some interference with an individual's privacy and freedom of movement has lawfully taken place, further intrusions should automatically be allowed....").

(200.) Accord United States v. Scott, 450 F.3d 863, 870 (9th Cir. 2006) ("Crime prevention is a quintessential general law enforcement purpose and therefore is the exact opposite of a special need.").

(201.) It is important to clarify that this is an argument about the doctrinal rule, not the Court's application of it. The Court has found intrusions to be necessary to the adjudicative process that patently are not--most obviously, the DNA search in King. Such decisions have had the effect of limiting defendants' liberty and privacy far more than the administrative demands of the process require. It is the rule itself, however, that should govern future cases.

(202.) I admittedly use the word "moral" here quite loosely, to denote normative arguments that invoke rights or duties (in contrast to the "practical," or instrumentalist, arguments explored in Section II.C). In other words, this Section addresses arguments that sound in deontological ethics, and the next Section addresses arguments that sound in consequentialist ethics.

(203.) Alschuler, supra note 130, at 533.

(204.) LJAF, Results, supra note 16, at 3.

(205.) On this point I part ways with Shima Baradaran Baughman, who has argued that "judges should not 'weigh' any of the evidence alleged against defendants before trial." Shima Baradaran, Restoring the Presumption of Innocence, 72 OHIO ST. L.J. 723, 772 (2011).

(206.) Accord Morse, supra note 32, at 119 n.15 ("The claim that the criminal charge is good evidence of dangerousness is an evidentiary argument rather than a principled reason to distinguish charged offenders.").

(207.) Gerstein v. Pugh, 420 U.S. 103,120 (1975) (holding that, after warrantless arrests, the Fourth Amendment requires a detached and neutral magistrate to determine that there is "probable cause to believe the suspect has committed a crime").

(208.) See, e.g., Robinson v. California, 370 U.S. 660, 666-67 (1962) (holding that the Eighth Amendment prohibits punishment for status (as opposed to punishment for volitional acts)).

(209.) See Mitchell N. Berman, Two Kinds of Retributivism, in PHILOSOPHICAL FOUNDATIONS OF CRIMINAL LAW 433, 437-38 (R.A. Duff& Stuart P. Green eds., 2011) (describing ambiguities latent in the concept of desert, and identifying the "dominant view" among retributivists as the notion that guilty people deserve to suffer); see also, e.g., Richard L. Lippke, Preventive Pre-Trial Detention Without Punishment, 20 RES PUBLICA 111, 122 (2014) (suggesting that preventive detention may be justified if the state can demonstrate that there is "substantial evidence" of guilt on a serious charge); Stephen J. Morse, Protecting Liberty and Autonomy: Desert/Disease Jurisprudence, 48 SAN DIEGO L. REV. 1077, 1124 (2011) (suggesting that pretrial detention "is justified by probable cause to believe that the accused has culpably committed a criminal offense").

(210.) Cf. Christopher Heath Wellman, The Rights Forfeiture Theory of Punishment, 122 ETHICS 371, 371 (2012).

(211.) Alschuler, supra note 130, at 557; id. at 511.

(212.) Thanks to Mitch Berman for this point. I have previously--and incorrectly--asserted that any judgment of forfeiture necessarily entails a judgment of desert. See Sandra G. Mayson, Collateral Consequences and the Preventive State, 91 NOTRE DAME L. PVEV. 301, 337 (2015) ("Only by blameworthy conduct can a person forfeit rights.").

(213.) See generally H.L.A. HART, PUNISHMENT AND RESPONSIBILITY: ESSAYS IN THE PHILOSOPHY OF LAW 23, 44 (2d ed. 2008) (describing criminal law as a "choosing system," the "method of social control which maximizes individual freedom within the coercive framework of law").

(214.) Id. at 37.

(215.) See Duff, supra note 28, at 130.

(216.) Id. at 122-23.

(217.) Id. at 124-27.

(218.) As noted above the line, Duff himself rejects this line of argument, at least with respect to detention. He does not address how it might apply to non-custodial restraints. Id. at 127-29, 131.

(219.) Reaves, supra note 3, at 22.

(220.) Coffin v. United States, 156 U.S. 432, 453 (1895).

(221.) Bell v. Wolfish, 441 U.S. 520, 533 (1979).

(222.) Id.

(223.) See, e.g., Shima Baradaran, Restoring the Presumption of Innocence, 72 OHIO ST. L.J. 723, 736-37 (2011).

(224.) RICHARD L. LIPPKE, TAMING THE PRESUMPTION OF INNOCENCE 129-54 (2016) (explaining the difficulties of applying the presumption in the pretrial context).

(225.) 397 U.S. 358(1970).

(226.) Id. at 361-65.

(227.) Id. at 363. In addition, "use of the reasonable-doubt standard is indispensable to command the respect and confidence of the community in applications of the criminal law." Id. at 364.

(228.) Id. at 363-64.

(229.) Id. at 367.

(230.) This argument rests on two premises. The first is that official condemnation is the defining feature of state punishment. See, e.g., Henry M. Hart, Jr., The Aims of the Criminal Law, 23 LAW&CONTEMP. PROBS. 401, 404 (1958) ("What distinguishes a criminal from a civil sanction... is the judgment of community condemnation which accompanies and justifies its imposition."); Mayson, supra note 212, at 318 & n.87 (proposing that "punishment is hard treatment inflicted as a putatively just consequence of blameworthy conduct" and collecting sources that support this view); Tribe, supra note 22, at 379 n.30 ("[I]t is the expression of community condemnation rather than any necessarily retributive purpose that characterizes a particular disability as punishment."). The second premise is that any official judgment of guilt entails official condemnation. If these two premises are true, it follows that any deprivation imposed by virtue of guilt constitutes punishment, and cannot constitutionally be imposed before trial. In other work, I have argued for a punishment test along these lines. See Mayson, supra note 212, at 345-46 (proposing that courts classify a deprivation of liberty as punishment whenever the state "claims primary authorization from a judgment of culpability," and otherwise classify deprivations as regulatory).

(231.) See Kingsley v. Hendrickson, 135 S. Ct. 2466, 2475 (2015) (noting that "pretrial detainees (unlike convicted prisoners) cannot be punished at all"); Bell v. Wolfish, 441 U.S. 520, 535 (1979) ("[U]nder the Due Process Clause, a detainee may not be punished prior to an adjudication of guilt in accordance with due process of law.").

(232.) Locke v. United States, 11 U.S. (7 Cranch) 339, 348 (1813); see also Harmon, supra note 124, at 310 (2016) (noting that probable cause "is almost by definition not enough proof to establish blameworthiness"). The function of the probable cause determination is not to establish blameworthiness, but rather "to safeguard citizens from rash and unreasonable interferences with privacy and from unfounded charges of crime." Brinegar v. United States, 338 U.S. 160, 176 (1949).

(233.) Winship, 397 U.S. at 367; cf. Kimberly Kessler Ferzan, Preventive Justice and the Presumption of Innocence, 8 CRIM. L. & PHIL. 505, 515, 523 (2014) (defending the right of states to restrain "culpable aggressors" who threaten future harm, but concluding that states should be required to prove the predicate criteria for culpability beyond a reasonable doubt).

(234.) Perhaps a deprivation should not be classified as "punishment" unless it has a purpose to censure or cause suffering. Cf. Mitchell N. Berman, The Justification of Punishment, in THE ROUTLEDGE COMPANION TO PHILOSOPHY OF LAW 143 (Andrei Marmor ed., 2012) (defining punishment as infliction of hard treatment "because of, and not despite" the suffering it will cause); Douglas Husak, Lifting the Cloak: Preventive Detention as Punishment, 48 SAN DIEGO L. REV. 1173, 1189 (2011) ("[A] sanction is not a punishment without a purpose to deprive and censure."). The Supreme Court, moreover, has not held that any deprivation imposed by virtue of guilt constitutes punishment. According to current doctrine, a pretrial deprivation is "punishment" if it is inflicted with "punitive" intent, or if it is irrational or excessive as a regulatory measure. Kingsley, 135 S. Ct. at 2473-74; United States v. Salerno, 481 U.S. 739, 746-47 (1987); Bell, 441 U.S. at 538-39.

(235.) The guilt of the accused is not the direct justification for arrest, or for other intrusions necessary to the administration of criminal proceedings. The direct justification for those intrusions is the state's legitimate interest in prosecution, which hinges partly on the likelihood of conviction. In other words, some likelihood of demonstrable guilt justifies the state in undertaking criminal proceedings, which may require some restriction of the accused's liberty. The probable-cause determination is addressed to the question of whether the evidence is adequate to justify the state in pursuing prosecution. To invoke the content and strength of the evidence against the accused as practical justification for prosecution (and the deprivations of liberty it entails) is one thing. To invoke the accused's guilt as moral justification for deprivations to prevent unrelated future crime is quite another. What constitutes an appropriate evidentiary standard for the former has no clear bearing on what constitutes an appropriate evidentiary standard for the latter.

(236.) This illustrates a limitation of the moral-predicate model. Some shoplifters are more dangerous than some murderers. Presuming that shoplifting is less of a moral predicate than murder, and authorizes less preventive restraint, the state must either forego effective restraint of dangerous shoplifters or impose greater restraint than is warranted. See Darin Clearwater, "If the Cloak Doesn't Fit, You Must Acquit": Retributivist Models of Preventive Detention and the Problem of Coextensiveness, 11 CRIM. L. & PHIL. 49 (2017); Robinson, supra note 39, at 1432 (arguing that "[s]egregation of the punishment and prevention functions offers a superior alternative").

(237.) Kimberly Ferzan and Alec Walen, for instance, hold that culpability can eliminate a person's right against preventive interference, but would also require the state to demonstrate that any restraint is reasonable and necessary to accomplish its preventive goals. See Ferzan, supra note 123, at 143-45; Alec Walen, A Punitive Precondition for Preventive Detention: Lost Status as a Foundation for a Lost Immunity, 48 SAN DIEGO L. REV. 1229, 1240 (2011).

(238.) Alschuler, supra note 130, at 559 (observing that, "[e]ven as supplemented by Gerstein, the [Bail Reform] Act authorizes imprisonment grounded almost entirely on a prediction of future misconduct," without a meaningful determination of probable guilt).

(239.) Once again, this principle does not preclude the state from invoking a person's riskiness as justification for depriving her of liberty nor from considering the strength of pending allegations in assessing that risk. What it prohibits is invoking guilt as an independent, or additional, justification--unless it is proven beyond a reasonable doubt. Nothing in this Article contests the notion that a conviction can serve as a moral predicate for preventive restraint. See, e.g., Husak, supra note 234, at 1186-87 (arguing that the state can pursue incapacitative ends through punishment); Walen, supra note 237, at 1240 (arguing that "one element of a justified punishment can be the temporary loss of the normal immunity to [long-term preventive detention]").

(240.) Larry Laudan & Ronald J. Alien, Deadly Dilemmas II: Bail and Crime, 85 CHI.-KENT L. REV. 23,39 (2010).

(241.) Id.

(242.) On this view, pretrial crime is akin to a "state-created danger." See, e.g., Morrow v. Balaski, 719 F.3d 160, 167 (3d Cir. 2013) ("[W]e have recognized that the Due Process Clause can impose an affirmative duty to protect if the state's own actions create the very danger that causes the plaintiff's injury.").

(243.) Samuel Scheffler, Doing and Allowing, 114 ETHICS 214 (2004) (noting that some consequentialists deny the doing/allowing distinction).

(244.) See S. REP. No. 98-225 (1983), as reprinted in 1984 U.S.C.C.A.N. 3182, 3329-30. The Senate Report for the Bail Reform Act of 1984 also referred to, for example, "the problem of how to change current bail laws to provide appropriate authority to deal with dangerous defendants seeking release,'' as if pretrial liberty were a privilege rather than the default. Id. at 6, as reprinted in 1984 U.S.C.C.A.N. 3182, 3189 (emphasis added).

(245.) See Wiseman, supra note 66, at 426-32 (explaining this phenomenon as part of "the principal-agent problem in bail determinations").

(246.) Many thanks to Paul Heaton, Michael Cahill, and Youngjae Lee for raising this argument.

(247.) This reasoning aligns with tort-law doctrines that impose a duty on certain actors, like landlords or therapists, to minimize third-party crime risk when they have notice of likely crime and a unique opportunity to prevent it. See, e.g., Kline v. 1500 Mass. Ave. Apartment Corp., 439 F.2d 477, 481 (D.C. Cir. 1970) (holding that a landlord's duty of care includes taking steps "to minimize the predictable risk to his tenants" from third-party crime, where he had notice of the risk and "the exclusive power to take preventive action"); Tarasoff v. Regents of the Univ. of Cal., 551 P.2d 334, 347 (Cal. 1976) (holding that therapists may have a duty of care to warn likely victims of foreseeable violence by their patients).

(248.) Cf. Youngjae Lee, Deontology, Political Morality, and the State, 8 OHIO ST. J. CRIM. L. 385 (2011) (arguing that core constraints on state punishment are categorical constraints imposed to limit the power of official blame, and thus not subject to instrumentalist adjustment).

(249.) Stephen Schulhofer has suggested a different heightened-duty argument. He posits that the pretrial phase might represent a "gap" where the threat of punishment is inadequate to prevent crime, such that the state has special authority (and responsibility) to engage in preventive restraint. Stephen J. Schulhofer, Two Systems of Social Protection: Comments on the Civil-Criminal Distinction, with Particular Reference to Sexually Violent Predator Laws, 7 J. CONTEMP. LEGAL ISSUES 69, 93-94 (1996). It is questionable, however, whether the pretrial phase represents such a gap. Schulhofer notes that some defendants are unlikely to be deterred by the threat of additional punishment from committing pretrial crime. Id. at 86-87. Relatedly, Christopher Slobogin has advocated for the notion of "undeterrability" to serve as a criterion for preventive detention. Slobogin, supra note 123, at 48 ("The undeterrability criterion better describes the 'gap' population that cannot be addressed by the criminal law--those people who are impervious to its dictates."). But there are many people for whom deterrence is ineffective, and Schulhofer does not explain why we should treat defendants differently from anyone else who is equally unlikely to heed the threat. In the end, he seems to conclude that the pretrial phase does not represent a "gap" in the competence of criminal-law deterrence at all. See Schulhofer, supra, at 96 (concluding that "[i]n the absence of mental illness sufficiently serious to preclude criminal responsibility," predictive confinement is impermissible).

(250.) For a defense of short-term pretrial preventive detention on cost-benefit grounds, see Walen, supra note 237, at 1238 (arguing that some defendants "may justifiably be detained for the sake of the general welfare because the burden on them is not too great, and we may ask reasonably small sacrifices of people for the sake of the general welfare").

(251.) There is a sizable economic literature that has endeavored, with increasing sophistication, to quantify these costs. See Aaron Chalfin, Economic Costs of Crime, in 2 THE ENCYCLOPEDIA OF CRIME AND PUNISHMENT 543, 544-48 (Wesley G. Jennings ed., 2016) (describing alternate methodologies to quantify "external" and "social" costs of crime and surveying existing cost estimates).

(252.) In the economics literature, these are called "external costs." Id. at 544-45.

(253.) Many crimes also have serious costs for the perpetrator. It is debatable whether these, if deserved, ought to count in the calculus. See, e.g., Berman, supra note 180, at 269-70 (explaining the retributivist position that "the state of affairs in which offenders experience the suffering they deserve is not bad").

(254.) Chalfin, supra note 251, at 545.

(255.) Cf. Harmon, supra note 35, at 895 (noting that the "potential benefits of criminal justice policy" include "reducing fear, improving citizen satisfaction, decreasing perceived disorder, and promoting legal compliance and cooperation with law enforcement").

(256.) As Rachel Harmon has noted, however, empirical analysis "of the costs of criminal justice policy continues to be anemic." Id.

(257.) Detention, in particular, may entail serious physical, psychological, and reputational harm. See, e.g., Barker v. Wingo, 407 U.S. 514, 532-33 (1972) (cataloguing the costs of detention to detainees); Appleman, supra note 66, at 1318-21 (describing abysmal jail conditions, as well as the effects of pretrial detention on families and on a defendant's case); Shima Baradaran Baughman, Costs of Pretrial Detention, 97 B.U. L. REV. 1, 15 (2017) (explaining that detention may entail "loss of freedom, income, and housing; childcare costs; loss and theft of property; strain on intimate relationships;" and "potential violent or sexual assault"). The sole recent attempt to quantify this loss of liberty is David S. Abrams & Chris Rohlfs, Optimal Bail and the Value of Freedom: Evidence from the Philadelphia Bail Experiment, 49 ECON. INQUIRY 750, 751, 766 tbl.7 (2011), which interprets bail-posting behavior as revealing defendants' valuation of their freedom, and concludes that average valuation was $1,050 for 90 days of liberty. But see David S. Abrams, The Imprisoner's Dilemma: A Cost-Benefit Approach to Incarceration, 98 IOWA L. REV. 905, 950 n.182 (2013) (acknowledging that this estimate is "likely to be a lower bound, since some offenders are likely credit constrained").

(258.) See, e.g., PRETRIAL JUSTICE INST., PRETRIAL JUSTICE: How MUCH DOES IT COST? 2 (2017) (calculating the total annual cost of pretrial jail beds to be $14 billion in the time period assessed).

(259.) See, e.g., CHRISTOPHER T. LOWENKAMP ET AL., ARNOLD FOUND., THE HIDDEN COSTS OF PRETRIAL DETENTION 4 (2013) (finding that longer pretrial detention increases the likelihood of future crime); Heaton, Mayson & Stevenson, supra note 4, at 767 (finding that pretrial detention of misdemeanor defendants substantially increases the likelihood of a new criminal charge within eighteen months).

(260.) See Angwin et al., supra note 19; Harcourt, supra note 19, at 240 ("[R]elying on prediction instruments to reduce mass incarceration will surely aggravate what is already an unacceptable racial disproportionality in our prisons."); Starr, supra note 19, at 836-37 (noting that the use of "demographic, socioeconomic, and family- and neighborhood-related characteristics" as indicators of risk will "further demographically concentrate" the impact of mass incarceration).

(261.) See Allen v. Illinois, 478 U.S. 364, 380 (1986) (Stevens, J., dissenting) (arguing that if a state can avoid criminal procedural protections by deeming a proceeding "civil," then "nothing would prevent a State from creating an entire corpus of 'dangerous person' statutes to shadow its criminal code"); cf. Allen & Laudan, supra note 39, at 796 n.47 ("The Supreme Court often pretends that jail time served while on bail is not 'punishment' but simply community protection; that distinction in this context seems strained at best.").

(262.) See Morse, supra note 209, at 1085 ("The incentive structure predisposes decisionmakers in cases involving danger to overpredict and thus to imprison or hospitalize longer than is necessary.").

(263.) Sonja Starr has shown that the use of an actuarial risk instrument may lead decisionmakers to weigh risk more heavily than they otherwise would. Starr, supra note 19, at 867-70 (describing a classroom experiment demonstrating this effect); see also HARCOURT, AGAINST PREDICTION, supra note 19, at 31-34, 173-92 (arguing that use of predictive instruments can distort "conceptions of justice"). Relatedly, risk-based decision-making may be particularly susceptible to framing effects. See Nicholas Scurich & Richard S. John, The Effect of Framing Actuarial Risk Probabilities on Involuntary Civil Commitment Decisions, 35 LAW & HUM. BEHAV. 83 (2011) (reporting that the majority of study participants deemed a twenty-six percent risk of violence to warrant civil commitment, but deemed civil commitment unwarranted if risk was expressed as a seventy-four percent chance of no violence). Lastly, and critically, criminal justice system actors, like judges and prosecutors, may lack the resources or technical sophistication necessary to identify flaws in predictive algorithms.

(264.) There could well be more, but these three emerged as themes in the literature and through my own reflection and conversations with colleagues.

(265.) Thanks to Richard Lippke for suggesting this point.

(266.) See supra note 262 and accompanying text.

(267.) Reaves, supra note 3, at 22.

(268.) See supra note 257 and accompanying text.

(269.) See Laudan & Allen, supra note 240, at 34 n.32 ("The loss to guilty defendants is... typically... zero as time served pre-trial is accounted for in their sentence.").

(270.) Using a large dataset from Harris County (including Houston, Texas), for example, Paul Heaton, Megan Stevenson, and I recently found that approximately seventeen percent of the detained misdemeanor defendants who pled guilty would not have been convicted at all had they been released rather than detained pretrial. Heaton, Mayson & Stevenson, supra note 4, at 771. Many such people plead guilty in exchange for a "time served" sentence, which means immediate release. Id. at 715 n.15 and accompanying text, 771 n.162 and accompanying text.

(271.) See, e.g., id. at 732-33 (noting that approximately ten percent of misdemeanor defendants in New York City and sixteen percent in Philadelphia receive a custodial sentence, though in Harris County the number is fifty-eight percent).

(272.) Reaves, supra note 3, at 24 (66% of felony defendants in the 75 largest urban jurisdictions in 2009 were convicted); id. at 29 (36% of those defendants--or 24% of the total--were sentenced to prison). An additional 37% of those convicted were sentenced to jail, id., but these sentences likely reflect "time served" sentences that were the result of pretrial detention rather than the sentence these defendants would have received had they been released pretrial. See id. at 31 (reporting mean and median jail sentences of five and four months, respectively).

(273.) To the author's knowledge, there is no reliable national estimate of what percentage of U.S. criminal cases are felonies (versus misdemeanors), but the best available evidence--a 2010 survey conducted by the National Center for State Courts--suggests that it is around a quarter. See Robert C. LaFountain et al., Examining the Work of State Courts: An Analysis 0/2010 State Court Caseloads, NAT'L CTR. FOR ST. CTS. 24 (2012), http://www.courtstatistics.org/other-pages/~/media/microsites/files/csp/data%2opdf/csp_dec.ashx [http://perma.cc/AF2X-GGV5] (reporting that misdemeanors represent more than three-quarters of state court caseloads in reporting states).

(274.) See Adam J. Kolber, Against Proportional Punishment, 66 VAND. L. REV. 1141, 1147 (2013) (explaining that pretrial detainees "who are subsequently convicted usually have their sentences shortened by the amount of time they spent in detention," and citing statutory provisions that mandate such "credit for time served").

(275.) E.g., 18 U.S.C. [section] 3585 (2012) (providing that "[a] defendant shall be given credit toward the service of a term of imprisonment for any time he has spent in official detention prior to the date the sentence commences," but providing no such credit for non-custodial pretrial restraint); 42 PA. CONS. STAT. ANN. [section] 9760 (West 2017) ("Credit shall include credit for time spent in custody prior to trial, during trial, pending sentence, and pending the resolution of an appeal."); Fraley v. U.S. Bureau of Prisons, 1 F.3d 924, 925 (9th Cir. 1993) ("[T]ime on probation does not qualify for credit."); Commonwealth v. Kyle, 874 A.2d 12, 22 (Pa. 2005) ("[W]e hold that time spent subject to electronic monitoring at home is not time spent in 'custody' for purposes of credit under Section 9760.").

(276.) The reality is that pretrial detention does currently function as pre-punishment. That should change. See Lippke, supra note 209, at 114 (suggesting criteria to ensure that detention is non-punitive). We might even rethink the practice of crediting pretrial detention as "time served." It fosters the illusion that pretrial detention has no cost in liberty, while reinforcing the impression that pretrial detainees are guilty people getting a head start on their sentences. What if, instead, the state was required to compensate all pretrial detainees for their loss of liberty? See Michael Louis Corrado, Punishment and the Wild Beast of Prey: The Problem of Preventive Detention, 86 J. CRIM. L. & CRIMINOLOGY 778, 814 (1996) (arguing that it would be "both fair and efficient" to compensate preventive detainees); Jeffrey Manns, Liberty Takings: A Framework for Compensating Pretrial Detainees, 26 CARDOZO L. PVEV. 1947, 1953 (2005).

(277.) See supra Section II.B.2.

(278.) The average arrestee is not acutely dangerous. The only two recent published studies that have measured the rate of arrest for violent crime among bailees have documented rates of 1.8-1.9%. Qudsia Siddiqi, Predicting the Likelihood of Pretrial Failure To Appear and/or Re-Arrest for a Violent Offense Among New York City Defendants: An Analysis of the 2001 Dataset (Final Report), N.Y. CRIM. JUST. AGENCY (2009), http://www.nycja.org/lwdcms/doc-view.php?module=reports&modulc_id=629&doc_name=doc [http://perma.cc/3B7Q-3V3F] (finding that, in a sample of 26,821 defendants released pretrial, the rate of rearrest for violent felony was 1.8% and the rate of rearrest for "violent offense" including misdemeanors was 3.0%); Baradaran & Mclntyre, supra note 11, at 527 (analyzing Bureau of Justice data on state-court felony cases). By "high-risk arrestee" I mean someone who would be classified as high risk by an existing pretrial risk assessment tool.

(279.) See, e.g., Integrated Data Systems (IDS), ACTIONABLE INTELLIGENCE FOR SOC. POL'Y, http://www.aisp.upenn.edu/integrated-data-systems [http://perma.cc/RSS5-CKZ5].

(280.) See generally Andrew Guthrie Ferguson, Big Data and Predictive Reasonable Suspicion, 163 U. PA. L. REV. 327, 331-32 (2015) (analyzing the interplay between big data and Fourth Amendment doctrine); Michael L. Rich, Machine Learning, Automated Suspicion Algorithms, and the Fourth Amendment, 164 U. PA. L. REV. 871, 878 (2016) (describing automated suspicion algorithm technology and analyzing it under existing Fourth Amendment doctrine).

(281.) See, e.g., Ferguson, supra note 280, at 385-86 (describing the use of predictive analytics to identify people at highest risk of committing or suffering gun violence in Chicago); Journalism for Social Change, Pennsylvania County Leads in Use of Big Data To Stem Child Abuse, Probes Ethics First, CHRON. OF SOC. CHANGE (May 26, 2016), http://chronicleofsocialchange.org/news-2/pennsylvania-county-leads-globe-uses-big-data-stem-child-abuse-not-without-probing-ethics-first [http://perma.cc/4228-EJNT] (describing a data initiative to identify children at highest risk of future abuse--and, implicitly, caregivers at highest risk of committing abuse).

(282.) The prospect of governments maintaining and analyzing comprehensive data dossiers on all of their citizens also raises privacy concerns and underlines the question of whether Fourth Amendment analysis should be different for defendants versus non-defendants. Certainly, the needs of criminal prosecution itself justify some special intrusions into defendants' privacy. If the intrusion is not necessitated by the prosecution, however, the standards for intrusion should not be more relaxed for defendants than for others. See supra Sections II.A and II.B. Thanks to Jessica Eaglin for raising this point.

(283.) See, e.g., Civil Rights Div., Investigation of the Baltimore City Police Department, U.S. DEP'T JUST. 55-61 (Aug. 10, 2016) http://www.civilrights.baltimorecity.gov/sites/default/files/20160810_DOJ%20BPD%20Report-FINAL.pdf [http://perma.cc/Q89K-CL66] (documenting dramatic racial disparities in the Baltimore City Police Department's arrest practices); Civil Rights Div., Investigation of the Ferguson Police Department, U.S. DEP'T JUST. 4 (Mar. 4, 2015) http://www.justice.gov/sites/default/files/opa/press-releases/attachments/2015/03/04/ferguson_police_department_report.pdf [http://perma.cc/8RRP-U86Q] ("Ferguson's approach to law enforcement both reflects and reinforces racial bias."); Frank Mclntyre & Shima Baradaran, Race, Prediction, and Pretrial Detention, 10 J. EMPIRICAL LEGAL STUD. 741, 759 (2013) ("Drug usage and sale rates among whites and blacks are often similar but systematically more blacks are arrested for drug possession and trafficking crimes than whites.").

(284.) United States v. Robinson, 414 U.S. 218, 238 n.1 (1973) (Powell, J., concurring) (quoting Charles v. United States, 278 F.2d 386, 388-389 (9th Cir. 1960)).

(285.) Note, however, that if we start using non-custodial means to initiate more criminal proceedings, this cost differential largely disappears. See Harmon, supra note 124 (suggesting that we should consider non-custodial options for a broader array of offenses).

(286.) Suppose this risk is calculated over the period of time equivalent to the average pretrial period for released defendants.

(287.) Baughman, supra note 257, at 18. Baughman's estimate admittedly excludes the component costs most difficult to quantify. She relies, furthermore, on the single existing estimate of the value of defendants' freedom (by Abrams & Rohlfs), which its authors acknowledge is likely to be an underestimate. Id. at 6 n.26, 18 fig.2; see also Abrams & Rohlfs, supra note 257, at 751 (noting that, because their methodology relies on bail-posting behavior as revealing the price defendants place on their own freedom, defendants' "credit constraints" may affect their estimate) ; David S. Abrams and David Rohlfs, Web Appendix for "Optimal Bail and the Value of Freedom: Evidence from the Philadelphia Bail Experiment," 3 (Aug. 2007), available under "Supporting Information" at http://onlinelibrary.wiley.com/doi/10.1111/j.1465-7295.2010.00288.x/abstract [http://perma.cc/B733-6S2V] (acknowledging that, because authors' data "do not allow for identifying the effects of credit constraints... our estimates may understate defendants' true valuations of freedom"); supra note 254 and accompanying text (describing the social costs of detention).

(288.) In 2016 dollars ($107,020 in 2008 dollars, as reported by the authors). This includes tangible and intangible costs, to both the victim and society. Kathryn E. McCollister et al., The Cost of Crime to Society: New Crime-Specific Estimates for Policy and Program Evaluation, 108 DRUG & ALCOHOL DEPENDENCE 98, 98-99 (2010) (synthesizing prior research and updating cost estimates for major crimes); see also Baughman, supra note 257, at 11 tbl.2 (reporting that cost estimates for assault range from $14, 715-$158, 250 in 2014 dollars, or $15,019-$161,126 in 2016 dollars); Paul Heaton, Hidden in Plain Sight: What Cost-of-Crime Research Can Tell Us About Investing in Police, 5 tbl.1 (RAND Corp., Occasional Paper No. 279, 2010) (reporting that the average of three prior estimates of the cost of "serious assault" was $87,238 in 2007 dollars, or $101,415 in 2016 dollars).

(289.) Harmon, supra note 124, at 319 (estimating the cost of arrest at $150 in officer time alone, i.e., excluding booking, transportation, or court costs); Christian Henrichson & Sarah Galgano, A Guide to Calculating Justice-System Marginal Costs, VERA INST. JUST. 19-20 (May 2013), https://storage.googleapis.com/vera-web-assets/downIoads/PubIications/a-guide-to-calcu-lating-justice-system-marginal-costs-1/legacy_downloads/marginal-costs-guide.pdf [http://perma.cc/P9EZ-8GZ6] (reporting estimates that range from $165 to $880 for arrests for non-violent crime).

(290.) To be cost-justified, detention that costs $412,000 must prevent 3.439 aggravated assaults; detention that costs $403,000 must prevent 3.364 aggravated assaults. For the detention of ten people to prevent 3.439 assaults, the ten must each present (on average) a 34.39% likelihood of committing an assault; to prevent 3.364, they must each present a 33.64% likelihood.

(291.) See supra Section II.B.2.

(292.) The less costly the restraint, the more of a difference it will make, and the greater the difference in minimum permissible risk thresholds will be.

(293.) Five studies published in the last two years, deploying quasi-experimental design, have documented this effect. Heaton, Mayson & Stevenson, supra note 4; Will Dobbie, Jacob Goldin & Crystal Yang, The Effects of Pre-Trial Detention on Conviction, Future Crime, and Employment: Evidence from Randomly Assigned Judges 3 (Nat'l Bureau of Econ. Research, Working Paper No. 22511, 2016), http://www.nber.org/papers/w22511.pdf [http://perma.cc/Q4KD-GTKZ]; Arpit Gupta, Christopher Hansman & Ethan Frenchman, The Heavy Costs of High Bail: Evidence from Judge Randomization 22 (Columbia Law & Econ. Working Paper No. 531, 2016), http://papers.ssrn.com/s0l3/papers.cfm?abstract_id=2774453 [http://perma.cc/2FGR-5WPB]; Emily Leslie & Nolan G. Pope, The Unintended Impact of Pretrial Detention on Case Outcomes: Evidence from NYC Arraignments 34-35 (Nov. 9, 2016) (unpublished manuscript), http://home.uchicago.edu/~npope/pretrial_paper.pdf [http://perma.cc/GV7S-H7GU]; Stevenson, supra note 4.

(294.) See, e.g., Samuel R. Gross & Barbara O'Brien, Frequency and Predictors of False Conviction: Why We Know So Little, and New Data on Capital Cases, 5 J. EMPIRICAL LEGAL STUD. 927, 930-31 (2008) (noting that "it is entirely possible that most wrongful convictions... are based on negotiated guilty pleas to comparatively light charges" to avoid "prolonged pretrial detention"); Alexandra Natapoff, Misdemeanors, 85 S. CAL. L. REV. 1313, 1316 (2012) ("[E]very year the criminal system punishes thousands of petty offenders who are not guilty").

(295.) See supra note 36 and accompanying text. Sex offender commitment statutes condition commitment on a "mental disorder" that renders a person substantially unable to refrain from harmful sexual behavior, but as commentators have noted, this conditioning is broad enough to reach most people at particularly high risk of committing serious crime. See Slobogin, supra note 37, at 123-26.

(296.) See generally Jennifer C. Daskal, Pre-Crime Restraints: The Explosion of Targeted, Non-Custodial Prevention, 99 CORNELL L. REV. 327, 327 (2014) (exploring "terrorism-related financial sanctions, the No Fly List, and the array of residential, employment, and related restrictions imposed on sex offenders"); Mayson, supra note 123 (examining the collateral consequences of criminal convictions as predictive risk regulations).

(297.) Cf. Morse, supra note 209, at 1121 (commenting that proposals to replace criminal justice with a regime of pure prevention "are surely coherent and many would be constitutional").

(298.) See, e.g., Sara Gordon, The Danger Zone: How the Dangerousness Standard in Civil Commitment Proceedings Harms People with Serious Mental Illness, 66 CASE W. RES. L. REV. 657, 673 (2016) ("Because many state statutes do not define 'danger,' the statutes themselves put the burden on clinicians to substitute their own judgment for what a finding of dangerousness should encompass." (citation omitted)); Fredrick E. Vars, Delineating Sexual Dangerousness, 50 Hous. L. REV. 855, 869-70 (2013) ("Notwithstanding three major United States Supreme Court cases addressing the constitutionality of sex offender commitment, the minimum probability of harm has not been squarely decided"); id. at 871 ("State statutes and judicial opinions have set the probability of recidivism bar at different heights.").

(299.) Christopher Slobogin, Dangerousness and Expertise Redux, 56 EMORY L.J. 275, 282 (2006) (noting that some civil and criminal commitment statutes define "danger" to include a likelihood of property damage or emotional harm).

(300.) See, e.g., Duff, supra note 28; sources cited supra note 38.

(301.) See, e.g., sources cited supra note 39.

(302.) Cf. Stephen J. Morse, Neuroprediction: New Technology, Old Problems, 8 BIOETHICA F. 128, 128 (2015) ("Deciding what rate and types of error are justifiable is a normative issue that can be resolved only by balancing the various interests implicated by the prediction, including the consequences to the subject and society and the cost of producing the prediction.").

(303.) See infra note 307.

(304.) See, e.g., MINN. STAT. ANN. [section] 253B.02 (West 2017) (limiting involuntary commitment to a person who "poses a substantial likelihood of physical harm to self or others"); State v. B.B., 245 P.3d 697, 701 (Or. Ct. App. 2010) ("To establish that a person is 'dangerous to self,' the state must present evidence that the person's mental disorder would cause him or her to engage in behavior that is likely to result in physical harm to himself or herself in the near term." (internal alterations and quotation marks omitted)).

(305.) A six-month period is a popular time period for assessing pretrial risk, because it is, at once, short enough to make data collection and assessment feasible and long enough to accommodate the average pretrial period for released defendants in many jurisdictions. See, e.g., LJAF, Results, supra note 16 (evaluating the success of the risk assessment tool in its first six months of use); Cohen & Reaves, supra note 82, at 7 ("Released defendants waited a median of 127 days from time of arrest until adjudication...."). One could argue that some property harms are more serious than some physical harm, such that the optimal risk standard would encompass serious non-bodily harms as well. The problem is the difficulty of determining what constitutes a sufficiently serious non-bodily harm.

(306.) Note that this kind of substantive risk standard is distinct from the standard of proof. See, e.g., Vars, supra note 298, at 872 (exploring the "relationship between standards of proof and recidivism risk thresholds").

(307.) CAL. CONST, art. I, [section] 12(b) ("great bodily harm to others"); FLA. CONST, art. I, [section] 14 ("physical harm to persons"); ILL. CONST, art. I, [section] 9 ("a real and present threat to the physical safety of any person"); OHIO CONST. art. I, [section] 9 ("serious physical harm"); VT. CONST, ch. II, [section] 40 ("substantial threat of physical violence"); WIS. CONST, art. I, [section] 8(2) ("serious bodily harm"); see also supra note 119.

(308.) See ARIZ. CONST. art. II, [section] 22 ("a substantial danger" and need to protect the "safety" of other persons and the community); COLO. CONST, art. II, [section] 19(1)(B) ("the public would be placed in significant peril"); LA. CONST, art. I, [section] 18 ("an imminent danger"); MICH. CONST, art. I, [section] 15 ("a danger"); MISS. CONST, art. III, [section] 29(3) ("a special danger"); Mo. CONST, art. I, [section] 32.2 ("a danger"); N.J. CONST, art. I, [section] 11 (need to protect the "safety" of other persons and the community); OKLA. CONST, art. II, [section] 8 (need to protect the "safety" of other persons and the community); UTAH CONST, art. I, [section] 8(1)(C) ("a substantial danger"); see also N.M. CONST. art. II, [section] 13 (no danger threshold specification); R.I. CONST, art. I, [section] 9 (same); TEX. CONST. art. I, [section][section] 11, 11a (same).

(309.) See Goldkamp, supra note 49, at 19, 27 (discussing the vagueness of "danger" definitions in state laws as of 1985); Gouldin, Disentangling, supra note 12, at 882-84.

(310.) 18 U.S.C. [section] 3142(b) (2012); S. REP. NO. 98-225, at 4-25 (1983); see also, e.g., United States v. Kelsey, 82 F. App'x 652, 654 (10th Cir. 2003) ("Mr. Kelsey has demonstrated an inability to stay away from drugs and drug-related activity, thereby making him a danger to society."); United States v. Strong, 775 F.2d 504, 506 (3d Cir. 1985) (finding that "Congress intended to equate traffic in drugs with a danger to the community").

(311.) See, e.g., ARIZ. CONST. art. II, [section] 22 ("reasonably assure the safety of the other person or the community"); FLA. CONST. art. I, [section] 14 ("reasonably protect the community"); N.J. CONST. art. I, [section] 11 ("reasonably... protect the safety of any other person or the community"); OKLA. CONST. art. II, [section] 8 ("assure the safety of the community or any person"); PA. CONST. art. I, [section] 14 ("reasonably assure the safety of any person and the community"); VT. CONST. II, [section] 40 ("reasonably prevent the physical violence"); WIS. CONST, art. I, [section] 8(3) (adequately protect members of the community"); 18 U.S.C.A. 53142(c) (West 2008) ("reasonably assure" safety).

(312.) Minton & Zeng, supra note 2, at 1.

(313.) Reaves, supra note 3, at 22.

(314.) See supra note 283.

(315.) Baradaran & McIntyre, supra note 11, at 528-29.

(316.) This is my understanding on the basis of conversations with statisticians in the field.

(317.) See Baradaran & Mclntyre, supra note 11 (conducting statistical risk assessment on a national pretrial dataset and identifying groups most likely to be rearrested for violent crime during pretrial release); Berk et al., supra note 97 (describing a machine-learned algorithm developed to forecast rearrest for domestic violence); LJAF, Public Safety, supra note 88 (explaining risk factors and formula for PSA risk scales, including violence-risk scale).

(318.) Rather, the primary metric offered to demonstrate the accuracy of a risk assessment instrument is the "area under the curve for the receiver operator characteristics" (AUC-ROC), a measure of classification accuracy (success at correctly identifying instances of each outcome). See, e.g., Cadigan et al., supra note 96, at 7-8 & n.3 (analyzing the "predictive ability" of the federal PTRA in terms of the AUC-ROC and explaining the basis for using this measure); Mona J.E. Danner et al., Race and Gender Neutral Pretrial Risk Assessment, Release Recommendations, and Supervision: VPRAI and PRAXIS Revised, LUMINOSITY, INC. 3 (Nov. 2016), http://luminosity-solutions.com/site/wp-content/uploads/2014/02/Race-and-Gender-Neutral-Pretrial-Risk-Assessment-November-2016.pdf [http://perma.cc/2KB6-C2CH] (calculating AUC-ROC for VPRAI and noting that AUC-ROC is "a common measure of risk assessment performance"). But for determining what intervention is appropriate for a given person or group, it is forecasting accuracy rather than the AUC-ROC that matters.

(319.) This means that the prevention benefit it provides relative to alternatives must outweigh its relative cost. Or, in more mathematical terms: If detention prevents 10% more crime than GPS monitoring, but costs 40% more, the question is not just whether the benefit of detention outweighs its cost. Instead, the question is whether the incremental (10%) prevention benefit outweighs the incremental (40%) increase in cost.

(320.) Starr, supra note 19, at 855-62.

(321.) See Morse, supra note 209, at 1125 ("The best hope for the future is that we discover preventive, nonintrusive techniques that will lower the risk of violent offenses for everyone and nonintrusive interventions that will reduce the risk of recidivism for offenders."); Underwood, supra note 85, at 1424-26 (noting that the cost of prediction depends on the response to it).

(322.) Richmond, California, for instance, operates a fellowship program for those at highest risk of killing or being killed; the program offers intensive mentorship, guidance, and a financial incentive for fellows to turn their lives around. See A.M. Wolf et al., Process Evaluation for the Office of Neighborhood Safety, NAT'L COUNCIL ON CRIME & DELINQUENCY (July 2015), http://www.nccdglobal.org/sites/default/files/publication_pdf/ons-process-evaluation.pdf [http://perma.cc/7ASR-6TD].

(323.) See Richard Berk, The Role of Race in Forecasts of Violent Crime, 1 RACE & Soc. PROBS. 1, 3 (2009).

(324.) Cf. Mitchell, supra note 30, at 1239 ("[A] finding of probable guilt of a violent crime is the best possible evidence of future dangerousness.... [N]othing more clearly forebodes future criminal activity than the commission of a crime in the immediate past.").

(325.) Cf. Cruzan v. Dir., Missouri Dep't of Health, 497 U.S. 261, 300 (1990) (Scalia, J., concurring) (noting that the Equal Protection Clause offers powerful protection against oppressive laws by "requir[ing] the democratic majority to accept for themselves and their loved ones what they impose on you and me").

(326.) There is a compelling case that none does, although this Article has not made it. See, e.g., Andrew von Hirsch, Prediction of Criminal Conduct and Preventive Confinement of Convicted Persons, 21 BUFF. L. REV. 717, 740 (1972) ("If a system of preventive incarceration is known systematically to generate mistaken confinements, then it is unacceptable in absolute terms because it violates the obligation of society to do individual justice.").

(327.) See supra Table 2 and accompanying discussion.

(328.) Angwin et al., supra note 19. "False" is in quotation marks because no outcome can refute a mere risk classification.

(329.) There are ways to engineer the algorithm to avoid this result, but they raise separate problems. I explore these problems and possible solutions in a work in progress. Mayson, supra note 19.

(330.) See sources cited supra note 96.
TABLE 1. SELECTED PRETRIAL RISK ASSESSMENT TOOL RISK FACTORS

                                PSA    COMPAS  FL PRAI  VPRAI  CPAT

Separate risk scores for
FTA/rearrest?                   Yes     Yes      No      No     No
Assesses risk of rearrest for   Yes     Yes      No      No     No
violent crime specifically?
Jurisdiction(s)                 Many     FL      FL      VA     CO
Requires Interview               No     Yes      Yes     Yes   Yes
Static/Dynamic/Both            Static   Both    Both    Both   Both
# Factors/Scales                 9       15      11       8     12
                               Risk Factors:Relate to...
Current Charge                   X       X       X        X
Prior Pending Charge(s)          X       ?                X     X
Prior Conviction(s)              X       X       X        X     X
Prior Violent Conviction(s)      X       X                X
Prior FTA(s)                     X       X       X        X     X
Residence                                X       X        X     X
Employment                               X       X        X
Drug/Alcohol Abuse                       X       X        X     X
Working Phone                            ?       X              X
Age                              X       ?       X              X
Active Warrant(s)                        ?                      X
Mental Health                            ?       X              X
Under Supervision                        ?                      X
Education                                X
Citizenship                              ?
Marital Status                           ?       X
Prior Custodial Sentence         X       ?                      X

                               PTRA  ORAS-PAT  IRAS-PAT

Separate risk scores for
FTA/rearrest?                   No      No        No
Assesses risk of rearrest for   No      No        No
violent crime specifically?
Jurisdiction(s)                Fed.     OH        IN
Requires Interview             Yes     Yes       Yes
Static/Dynamic/Both            Both    Both      Both
# Factors/Scales                14      7         7
                               Risk Factors:Relate to...
Current Charge                   X
Prior Pending Charge(s)          X
Prior Conviction(s)              X
Prior Violent Conviction(s)
Prior FTA(s)                     X       X         X
Residence                        X       X         X
Employment                       X       X         X
Drug/Alcohol Abuse               X       X         X
Working Phone
Age                              X       X         X
Active Warrant(s)
Mental Health
Under Supervision
Education                        X
Citizenship                      X
Marital Status
Prior Custodial Sentence                 X         X

TABLE 2. PREDICTIVE ACCURACY FOR HIGHEST-RISK GROUPS

Tool                           Outcome                   Period

                        Rearrest for Any Crime
Federal PTRA (102)    Any "new criminal arrest"     Pretrial period
FL PRAI (103)       Any "arrest for another crime"      6 months
IRAS-PAT (104)              Any new arrest                 ?
VPRAI (105)                 Any new arrest          Pretrial period
PSA (106)                   Any new arrest          Pretrial period,
                                                     up to 6 months
ORAS-PAT (107)          Any new arrest or FTA            1 year
CPAT (108)            Any "new criminal filing,"    Pretrial period,
                      incl. traffic or municipal      up to 1 year
COMPAS (109)                Any new arrest              6 months
                      Rearrest for Violent Crime
COMPAS (110)          New arrest - violent crime        6 months
PSA (111)             New arrest - violent crime    Pretrial period,
                                                     up to 6 months
Berk et al. (112)       New arrest - domestic           2 years
                         violence with injury

Tool                   Rate

Federal PTRA (102)     10%
FL PRAI (103)       Approx. 15%
IRAS-PAT (104)         17%
VPRAI (105)            29.5%
PSA (106)              23%

ORAS-PAT (107)         29.5%
CPAT (108)             42%

COMPAS (109)           42.1%

COMPAS (110)            8%
PSA (111)               8.6%

Berk et al. (112)      21%
COPYRIGHT 2018 Yale University, School of Law
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2018 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Mayson, Sandra G.
Publication:Yale Law Journal
Date:Jan 1, 2018
Words:36608
Previous Article:Uncovering the Codifier's Canon: How Codification Informs Interpretation.
Next Article:The Original Theory of Constitutionalism.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters