Evidence-Based Sentencing and the Taint of Dangerousness
Today’s world is “all about the data.”1 In a variety of contexts, innovators have offered statistical models as a way to reduce or eliminate human error.2 The promise of quantitative optimization has even influenced our criminal justice system. About twenty states have developed or adopted models to predict a defendant’s risk of recidivating.3 Often used in bail determinations and pretrial diversion programs, these risk predictors are also increasingly used for sentencing.4 In these cases, defendants may actually receive harsher sentences if they are calculated to have a higher risk of committing future crimes.5
Proponents argue that future risk, as predicted by past data, is nothing new to sentencing.6 Creating the Federal Sentencing Guidelines, for example, involved analyzing “some 40,000 convictions [and] a sample of 10,000 augmented presentence reports.”7 Modern risk models are just more accurate “tool[s]”8 that improve the calculations judges already make about defendants.9 Plus, consideration of the models’ risk “scores” is not even mandatory.10
This Essay, however, joins the call for greater scrutiny of “evidence-based sentencing,”11 and cautions against a particular form. In Pennsylvania, the state legislature has mandated the creation of a “risk assessment instrument” for determining “the relative risk that an offender will reoffend and be a threat to public safety.”12 The model incorporates not only factors related to criminal conduct, but also demographic traits: age, gender, and county.13 As scholars like Sonja Starr have pointed out, these types of factors raise concerns under the Equal Protection Clause;14 at a minimum, their inclusion could aggravate existing disparities in incarceration.15 Yet surprisingly, Pennsylvania’s decision has drawn scant public attention.16 Indeed, the hearings designed to elicit feedback were so poorly attended that the state sentencing commission extended the public comment period through the end of 2015.17 Given the high stakes, this lack of attention is quite troubling.
This Essay seeks to foster debate by highlighting an additional problem18: legislatively tailored risk models that single out “high risk” people could violate the Constitution’s prohibition on bills of attainder.19 Bills of attainder are “legislative enactment[s] which determine[] guilt and inflict[] punishment upon an identifiable person or group without a judicial trial.”20Historically, the term “bill of attainder” referred to a legislative death sentence.21 However, the Supreme Court has instructed that the Bill of Attainder Clauses are “not to be given a narrow historical reading.”22 The reason is that the prohibitions on state23 and federal24 bills of attainder are vital safeguards of the separation of powers,25 preventing legislatures from exercising the judicial power to “rul[e] upon the blameworthiness of, and levy[] appropriate punishment upon, specific persons.”26 Thus, “[l]egislative acts, no matter what their form, that apply either to named individuals or to easily ascertainable members of a group in such a way as to inflict punishment on them without a judicial trial are bills of attainder prohibited by the Constitution.”27
Although Pennsylvania’s risk predictor is not a traditional bill of attainder, there are important similarities. First, creating and adopting Pennsylvania’s model was a “legislative act.”28 The Pennsylvania General Assembly amended Title 42 of Pennsylvania’s Code of Law (Judiciary and Judicial Procedure) to require the development of a “sentence risk assessment instrument.”29 The instrument was then designed by the state sentencing commission, which unlike the Federal Sentencing Commission,30 is a legislative agency31 that includes “[t]wo members of the House of Representatives” and “[t]wo members of the Senate.”32 The Commission’s purpose, as explained in Commonwealth v. Sessoms, is “the furtherance of . . . the legislative power.”33
Second, one could argue that Pennsylvania’s risk predictor “inflicts punishment without a judicial trial.”34 Whether consideration at sentencing is mandatory or not, legislative pronouncements that certain individuals deserve35 greater punishment could significantly shape sentencing outcomes.36 In addition, the consequences of being labelled “a threat to public safety” extend well beyond the criminal justice system. Adopting a risk model “endors[es] a message: that [the state] considers certain groups of people dangerous.”37 Historically,38 certain bills of attainder—called “bills of pains and penalties”39—did exactly this, imposing “brand[s] of infamy”40 on their targets to create “broader social disability.”41 For example, such bills humiliated people by having them locked in the public stocks.42 In addition to public shame, the targeted could suffer concrete private disabilities, such as discrimination in employment and housing; the result could be “a kind of social death.”43 Today, condemnation and censure continue to be important touchstones for bills of attainder: “[w]hile the prohibition against bills of attainder has evolved far beyond the original context of capital sentences, it continues to focus on legislative enactments that ‘set[] a note of infamy’ on the persons to whom the statute applies.”44 Branding the possessors of “high-risk” traits could have such an effect.
Finally, and most importantly, statistical risk predictors enable a legislature to “single out its enemies—or the politically unpopular.”45 By selecting the right traits from among the many that are correlated with recidivism, a legislature could deliberately taint a particular group.46 Modern risk models include a wide range of “static” factors that individuals cannot change or escape47—from criminal and substance-abuse history to demographic information to the criminality of one’s associates and friends.48 Even factors like whether one’s parents have been arrested or jailed have been found to predict recidivism;49 building these types of criteria into a risk model could entrench community-wide or generational taints of dangerousness. For instance, Pennsylvania’s model includes three demographic traits, concluding that “offenders were more likely to recidivate if they were young, . . . male, [and] from an urban county.”50 These traits, alone, may not specify a sufficiently narrow group,51 but the use of these types of factors opens the door to alarming possibilities. As the Justice Department warned, evidence-based sentencing “will become much more concerning over time as other far reaching sociological and personal information . . . are incorporated into risk tools.”52 Pennsylvania has taken another major step in the direction of Big Data.53
Alexander Hamilton cautioned that “[n]othing is more common than for a free people, in times of heat and violence, to gratify momentary passions, by letting into the government principles and precedents which afterwards prove fatal to themselves [including] banishment by acts of the legislature.”54 There is no doubt that crime reduction and efficient resource allocation are worthy goals of the criminal justice system. This Essay, however, is a reminder that, in our pursuit of solutions, we cannot lose sight of important constitutional principles.55 Thus, in Pennsylvania and beyond, we should move into the “Risk Assessment Era” with care and, crucially, deliberation. This is especially true in states that have adopted or are adopting risk models, but also in states that may join the trend someday; citizens should investigate these reforms, attend public hearings, and voice their opinions. For even worse than sacrificing our values would be losing them through inattention and indifference.
Gregory Cui is a member of the Yale Law School J.D. Class of 2017. He is deeply grateful to Professor Daniel Markovits, Professor Jed Rubenfeld, Michael Clemente, Charlie Bridge, Elizabeth Ingriselli, Joseph Masterman, David Simins, Tal Eisenzweig (his outstanding Lead Editor), McKaye Neumeister, Megan McGlynn, William Stone, Rachel Wilf, and Alina Lindblom for their thoughtful comments and suggestions.
Preferred Citation: Gregory Cui, Evidence-Based Sentencing and the Taint of Dangerousness, 125 Yale L.J. F. 315 (2016), http://www.yalelawjournal.org/forum/evidence-based-sentencing-and-the-taint-of-dangerousness.
See Colette Martin, Results of IBM’s CIO Study—Data Is King, Forbes (July 9, 2011, 3:43 PM), http://www.forbes.com/sites/work-in-progress/2011/07/09/results-of-ibms-cio-study-data-is-king/ [http://perma.cc/WBM3-J6EM].
See, e.g., Louis Columbus, 84% of Enterprises See Big Data Analytics Changing Their Industries’ Competitive Landscapes in the Next Year, Forbes (Oct. 19, 2014, 9:03 PM), http://www.forbes.com/sites/louiscolumbus/2014/10/19/84-of-enterprises-see-big-data-analytics-changing-their-industries-competitive-landscapes-in-the-next-year/#2c2424803250 [http://perma.cc/Q4BZ-W38P]; 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/LGS9-392X]; Alex Knapp, Scientists Beat the House at Roulette with Chaos Theory, Forbes (Oct. 27, 2012, 2:49 PM), http://www.forbes.com/sites/alexknapp/2012/10/27/scientists-beat-the-house-at-roulette-with-chaos-theory/#3801adb95217 [http://perma.cc/569Z-L7HS]; Ben Mathis-Lilley, Baseball Expert from Moneyball Hired to Run Worst Team in NFL, Slate (Jan. 5, 2016, 6:06 PM), http://www.slate.com/blogs/the_slatest/2016/01/05/paul_depodesta_hired_to_run_cleveland_browns.html [http://perma.cc/JG6Y-D5MG].
See, e.g., Jordan M. Hyatt et al., Follow the Evidence: Integrate Risk Assessment into Sentencing, 23 Fed. Sent’g Rep. 266, 266 (2011) (“The introduction of standardized risk assessments into sentencing would hardly represent a sea change.”). That future risk is relevant to sentencing, of course, does not end the matter: there are still limits on which determinants of risk can and should influence sentencing and, as this paper explains, on who decides which persons are “high risk.”
U.S. Sentencing Guidelines Manual ch. 1, pt. A, concluding note (U.S. Sentencing Comm’n 2015), http://www.ussc.gov/sites/default/files/pdf/guidelines-manual/2015/CHAPTER_1.pdf [http://perma.cc/N55K-6XZC].
Hyatt et al., supra note 6, at 266 (“With varying degrees of formality, judges already consider risk at sentencing.”); see also J.C. Oleson, Risk in Sentencing: Constitutionally Suspect Variables and Evidence-Based Sentencing, 64 S.M.U. L. Rev. 1329, 1340 (2011) (“Sentencing blindly, these judges will either over-sentence . . . or under-sentence and release dangerous criminals into communities, thereby creating new victims of crime.”).
42 Pa. Consol. Stat. § 2154.7 (2009) (providing that “[t]he risk assessment instrument may be used”); see Christian Alexandersen, Should Prison Sentences Be Partially Based on Possible, Future Crimes? Pennsylvania Thinks So, PennLive (Aug. 31, 2015, 2:06 PM), http://www.pennlive.com/midstate/index.ssf/2015/08/should_prison_sentences_be_par.html [http://perma.cc/9JPY-QFCB].
Pa. Comm’n on Sentencing, Special Report: Impact of Removing Demographic Factors 19 (2015), http://pcs.la.psu.edu/publications-and-research/research-and-evaluation-reports/risk-assessment/phase-ii-reports/special-report-impact-of-removing-demographic-factors/view [http://perma.cc/J9N7-P8YC].
See Sonja B. Starr, The New Profiling: Why Punishing Based on Poverty and Identity Is Unconstitutional and Wrong, 27 Fed. Sent’g Rep. 229, 230-31 (2015), http://fsr.ucpress.edu/content/27/4/229.full-text.pdf+html [http://perma.cc/7KAM-52V7] [hereinafter The New Profiling]; see also Sonja B. Starr, Opinion, Sentencing, By the Numbers, N.Y. Times (Aug. 10, 2014), http://www.nytimes.com/2014/08/11/opinion/sentencing-by-the-numbers.html [http://perma.cc/MV5R-EVND] (“I doubt many policy makers would publicly defend the claim that people should be imprisoned longer because they are poor, for instance. Such judgments are less transparent when they are embedded in a risk score. But they are no more defensible.”). In addition, the Pennsylvania model defines recidivism in terms of subsequent arrest, rather than conviction, drawing into the equation disparities arising from policing practices. See Interim Report 5: Developing Categories of Risk, Pa. Comm’n on Sentencing 4 (2012), http://pcs.la.psu.edu/publications-and-research/research-and-evaluation-reports/risk-assessment/phase-i-reports/interim-report-5-developing-categories-of-risk/view [http://perma.cc/SN4C-69N5].
See Bernard E. Harcourt, Risk as a Proxy for Race: The Dangers of Risk Assessment, 27 Fed. Sent’g Rep. 237, 238-40 (2015), http://fsr.ucpress.edu/content/27/4/237.full-text.pdf [http://perma.cc/LYV3-MX7A]; see also Devlin Barrett, Holder Cautions on Risk of Bias in Big Data Use in Criminal Justice, Wall St. J.: Law Blog (Aug. 1, 2014, 2:59 PM), http://blogs.wsj.com/law/2014/08/01/holder-cautions-on-risk-of-bias-in-big-data-use-in-criminal-justice [http://perma.cc/RT7V-S7Z3] (quoting then-Attorney General Eric Holder’s warning that “basing sentencing decisions on static factors and immutable characteristics . . . may exacerbate unwarranted and unjust disparities that are already far too common in our criminal justice system and in our society”).
Anna Maria Barry-Jester et al., Should Prison Sentences Be Based on Crimes That Haven’t Been Committed Yet?, FiveThirtyEight Politics (Aug. 4, 2015, 7:15 AM), http://fivethirtyeight.com/features/prison-reform-risk-assessment/ [http://perma.cc/Z6FK-H9UA].
This Essay is not directed at benevolent uses of risk assessment instruments—for example, as means of diverting low-risk individuals from unnecessarily harsh punishments. As Akhil Amar points out, “a law giving [a person] a special benefit would probably not violate the non-attainder principle.” Akhil Reed Amar, Attainder and Amendment 2: Romer’s Rightness, 95 Mich. L. Rev. 203, 213 (1996).
Akhil Reed Amar, America’s Constitution: A Biography 124 (2012) (“The bans on attainders and ex post facto laws had deep roots in rule-of-law ideology. As we have seen, the basic tripartite structure of the federal government reflected a strong commitment to the ideal that legislation, at least if punitive, should be general and prospective.”).
See Commonwealth v. Sessoms, 532 A.2d 775, 780 (Pa. 1987) (comparing the Commission to standing legislative committees and subcommittees); see also 42 Pa. Cons. Stat. § 2151.2(a) (2015) (describing it as “an agency of the General Assembly”). The distinction between legislative and executive agencies is important, as several courts have found the Bill of Attainder Clause inapplicable to the latter. See, e.g., Scheerer v. U.S. Att’y Gen., 513 F.3d 1244, 1253 n.9 (11th Cir. 2008), cert. denied sub nom. Scheerer v. Mukasey, 555 U.S. 825 (2008) (“We have never held that the Constitution’s Bill of Attainder Clause is applicable to Executive Branch regulations, and other courts have suggested to the contrary.” (internal citation omitted)).
See Cummings v. Missouri, 71 U.S. (4 Wall.) 277, 323 (1867). Note that “without a judicial trial” does not mean that the bill of attainder’s target is never subjected to trial. Rather, the question is whether the legislative act “assume[s] the guilt and adjudge[s] the punishment” beforehand. See id. at 325 (holding that a state constitutional amendment requiring certain officeholders to swear an oath of loyalty was a bill of attainder, even though the defendant was later indicted, convicted, and sentenced for serving as a clergyman without having taken the oath).
Proponents of these models sometimes argue that deterring future crime is not a punitive end. It is true that there is a distinction between retributivist, “just deserts” theories of punishment and understandings built upon the concept of deterrence (both general and specific). But as the Court in Brown explained, “[i]t would be archaic to limit the definition of ‘punishment’ to ‘retribution.’ Punishment serves several purposes: retributive, rehabilitative, deterrent – and preventive. One of the reasons society imprisons those convicted of crimes is to keep them from inflicting future harm, but that does not make imprisonment any the less punishment.” United States v. Brown, 381 U.S. 437, 458 (1965).
See Sonja B. Starr, Evidence-Based Sentencing and the Scientific Rationalization of Discrimination, 66 Stan. L. Rev. 803, 866 (2014) (“In many legal, policy, and other contexts, scholars have observed that judges and other decisionmakers [sic] often defer both to scientific models that they do not really understand and to ‘expert’ viewpoints.”); see also The New Profiling, supra note 14, at 233 (“Unless every judge simply ignores this estimate, we should worry . . . .”).
The Court has explained that “[t]he infamous history of bills of attainder is a useful starting point” for determining whether a law imposes “punishment.” Nixon v. Adm’r of Gen. Servs., 433 U.S. 425, 473 (1977). In addition, suspected bills of attainder must undergo two additional inquiries. See id. at 475 (“But our inquiry is not ended by the determination that the Act imposes no punishment traditionally judged to be prohibited . . . .”). Under the “functional approach,” a punitive purpose may be reasonably inferred from a lack of legitimate “non-punitive legislative purposes” for the type and severity of burdens imposed. Id. at 475-76. Under the “motivational” test, the legislative record, itself, can “evince[] a congressional intent to punish.” Id. at 478.
Darrell A. H. Miller, The Stain of Slavery: Notes Toward an Attainder Theory of the Thirteenth Amendment, 38 U. Toledo L. Rev. 1011, 1029 (2007) (“As Justice Story’s comments show, the Framers would have understood attainder to encompass both legal disabilities, such as the prohibition against owning property and the broader social disability, what he calls disgrace.”); see Amar, supra note 19, at 212-13 (“[I]f the purpose and social meaning of our ineligibility law is to stigmatize or degrade a named person—to ‘taint’ or ‘stain’ him, or to label him as less worthy or deserving of less respect or trust than his fellow citizens—then it should be treated . . . as a bill of pains and penalties that offends the nonattainder principle.”); see also Amar, supra note 25, at 125 (“In the federal Constitution, the spirit animating the ban on bills of attainder extended to all laws heaping scorn or punishment upon specifically named individuals.”).
Foretich, 351 F.3d at 1219-20 (“‘For when it is now clear beyond all dispute, that the criminal is no longer fit to live upon the earth, but is to be exterminated as a monster and a bane to human society, the law sets a note of infamy upon him, puts him out of it’s [sic] protection, and takes no farther care of him than barely to see him executed. He is then called attaint, attinctus, stained, or blackened. He is no longer of any credit or reputation . . . .’” (quoting 4 William Blackstone, Commentaries *380)).
Using traits, instead of names, to identify persons is part of the history of bills of attainder: “[i]t was not uncommon for English acts of attainder to inflict their deprivations upon relatively large groups of people, sometimes by description rather than name.” United States v. Brown, 381 U.S. 437, 461 (1965).
See Interim Report 1: Review of Factors Used in Risk Assessment Instruments, Pa. Comm’n on Sentencing 2, 4 (2011), http://pcs.la.psu.edu/publications-and-research/research-and-evaluation-reports/risk-assessment/phase-i-reports/interim-report-1-review-of-factors-used-in-risk-assessment-instruments/view [http://perma.cc/ACA6-SRY5] (“[S]tatic risk factors . . . are stable over time and not amendable [sic] to rehabilitative efforts. . . . Static predictors . . . were the most frequently cited risk factors.”); Jonathan J. Wroblewski, U.S. Dep’t of Justice, Criminal Div., Letter to U.S. Sent’g Comm’n 7 (July 29, 2014), http://www.justice.gov/sites/default/files/criminal/legacy/2014/08/01/2014annual-letter-final-072814.pdf [http://perma.cc/8V2Q-2LX9] (“[M]ost current risk assessments—and in particular the PCRA, which is specifically mentioned in the pending federal legislation—determine risk levels based on static, historical offender characteristics such as education level, employment history, family circumstances and demographic information.”). The original parole prediction instrument, known as the Burgess method, included nationality/race as one of the factors. See Harcourt, supra note 15, at 238.
Interim Report 1, supra note 41; see also Melissa Hamilton, Risk-Needs Assessment: Constitutional and Ethical Challenges, 52 Am. Crim. L. Rev. 231, app. at 290 (2015) (describing the fourth-generation tool “COMPAS,” which considers “criminal associates,” “financial problems,” “family criminality,” and even “social environment”); Nathan James, Cong. Research Serv., R44087, Risk and Needs Assessment in the Criminal Justice System 7 (2015) (identifying “antisocial associates” as one of the “central eight” risk and needs factors).
Phase II Interim Report 1: Development of a Risk Assessment Scale by Offense Gravity Score for All Offenders, Pa. Comm’n on Sentencing 8 (2015), http://pcs.la.psu.edu/publications-and-research/research-and-evaluation-reports/risk-assessment/phase-ii-reports/Interim-Rpt-1-Phase-2/view [http://perma.cc/UJ25-YGW3].
See, e.g., Jordan Robertson, How Big Data Could Help Identify the Next Felon—Or Blame the Wrong Guy, Bloomberg Bus. (Aug. 8, 2013), http://www.bloomberg.com/news/2013-08-14/how-big-data-could-help-identify-the-next-felon-or-blame-the-wrong-guy.html [http://perma.cc/M3GQ-MKHQ].
Cf. Richard G. Kopf, Like the Ostrich that Buries Its Head in the Sand, Mr. Holder Is Wrong About Data-Driven Sentencing, Hercules and the Umpire: The Role of the Federal Trial Judge (Aug. 10, 2014), http://wednesdaywiththedecentlyprofane.me/2014/08/10/like-the-ostrich-that-buries-its-head-in-the-sand-mr-holder-is-wrong-about-data-driven-sentencing [http://perma.cc/HZU2-JXUJ] (“If race, gender or age are predictive as validated by good empirical analysis, and we truly care about public safety while at the same time depopulating our prisons, why wouldn’t a rational[] sentencing system freely use race, gender or age as predictor of future criminality?” (emphasis added)).