![Sample our Social Sciences journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110822/TOC-Subject-Sample-2021-Social-Sciences.jpg"})
Abstract
A basic principle in sentencing offenders is proportionality. However, proportionality judgments are often left to the discretion of the judge, raising familiar concerns of arbitrariness and bias. This paper considers the case for systematizing judgments of proportionality in sentencing by means of an algorithm. The aim of such an algorithm would be to predict what a judge in that jurisdiction would regard as a proportionate sentence in a particular case. A predictive algorithm of this kind would not necessarily undermine justice in individual cases, is consistent with a particularistic account of moral judgment, and is attractive even in the face of uncertainty as to the legitimate purposes of punishment.
I am grateful to audiences at the Centre for the Study of Law and Society at Berkeley Law School, King’s College London, Osgoode Hall Law School, and the Centre for Ethics at the University of Toronto for comments on earlier drafts. I received invaluable written feedback from Aziz Huq as well as from an anonymous referee for Criminal Justice Ethics. I am also indebted to Ioana Dragalin for editorial assistance.
Notes
[Disclosure Statement: No potential conflict of interest was reported by the author.]
1 Unwarranted sentencing disparity was at the heart of a split judgment in a recent Alberta Court of Appeal judgment. See R v Ryan, 2015 A.B.C.A. 286 (Can. Alta. C.A.).
2 See American Law Institute, Model Penal Code: Sentencing (Tentative Draft), § 6B.09 (“Evidence-Based Sentencing; Offender Treatment Needs And Risk of Reoffending”), §§ (2) and (3).
3 There is a large body of literature on risk-assessment devices, both supporting and critical; see, for example, Harcourt, Against Prediction; Starr, “Evidence-Based Sentencing”; Hannah-Moffat, “Actuarial Sentencing”; Chanenson and Hyatt, The Use of Risk.
4 Others have made similar proposals. See, for instance, Adi Leibovitch’s argument in favor of “curving” sentencing discretion by providing judges with statistical information about how similar cases are treated in different courts in order to ensure that harmony between sentences is imposed in specialized and generalist courts (“Punishing on a Curve”). In a similar spirit, Laqueur and Copus propose a “synthetic crowdsourcing” approach to resolving inconsistency in parole hearings (see “Synthetic Crowdsourcing”). More generally, Bagaric and Wolf raise many of the points considered here in their discussion of computerized sentencing (see “Sentencing By Computer”).
5 Similar tools are already in existence. For instance, a Canadian firm has developed a searchable database to predict sentencing ranges based on decided cases that share similar features to a given case. See http://www.rangefindr.ca/
6 As a result, my proposal is not an instance of what Harcourt would regard as an “actuarial” method, as he defines the term. For Harcourt, “actuarial methods” are aimed at predicting “past, present or future criminal behavior of a particular person” (Against Prediction, 16). My proposal is focused on predicting judicial opinion, not criminal behavior.
7 See e.g. Morris, The Future of Imprisonment.
8 See Frase, “Limiting Retributivism,” 93.
9 More broadly in the philosophy of criminal law, the distinction reflects a divide between more Kantian approaches to punishment, which insist on punishment to vindicate purely abstract rights; and more liberal approaches (associated most prominently with H.L.A. Hart), which insist that punishment is legitimate only insofar as it furthers valuable social objectives.
10 See R v Latimer, [2001] 1 S.C.R. 3 (Can.).
11 Ethics Without Principles, 7. Moral particularism is typically, although not inevitably, associated with holism.
12 For a helpful discussion of holism in the context of moral particularism, see Little, “Moral Generalities Revisited.” The significance of holism in this context has not gone wholly unnoticed; see Tata, “The Application of Judicial Intelligence.”
13 It may be worth noting here that supervenience does not imply that there are patterns or general principles linking the facts of a case to proportionality judgments. It only entails that, as Margaret Olivia Little puts it, “[t]wo situations … cannot differ in some moral respect without differing in some nonmoral respect” (“Moral Generalities Revisited,” 280–81); see also Lance and Little, “From Particularism to Defeasibility”; Dancy, Ethics Without Principles, 85–88.
14 Of course, these features might themselves make an issue of the criminal’s identity—for instance, that the accused was a family member. But that does not violate anonymity, for the same punishment will be given to anyone who commits a similar crime under similar circumstances, where those circumstances include standing in a familial relation to the victim.
15 Some of the risk assessment devices in the bail context operate in this manner. For instance, the Arnold Foundation’s risk assessment tool for bail is based on a simple regression.
16 My thanks to Ryan Liss for emphasizing this point to me.
17 I am, obviously, not proposing any particular algorithm here. That said, it would be quite surprising—and rather disturbing—if it were to turn out that there are no meaningful predictive correlations between fact patterns and sentences imposed.
18 Making the algorithm’s predictions merely advisory would help offset some of the widely noted problems with prosecutorial charging discretion in structured sentencing regimes, as judges would not become boxed in by how a prosecutor has chosen to structure an indictment.
19 See Tonry, Sentencing Matters. However, others have argued that the evidence in favor of presumptive over advisory regimes is less clear cut, as some advisory regimes have compliance rates similar to presumptive regimes. See Hunt and Connelly, “Advisory Guidelines” and Reitz, “Comparing Sentencing Guidelines.” Reitz observes that “[s]ome advisory guidelines have proven to be irrelevancies,” although in other jurisdictions, “advisory guidelines have won a stature not so different from their presumptive counterparts” (196).
20 For simplicity’s sake, I shall be focusing on custodial sanctions. Most of the attention on proportionality and parity in sentencing has focused on custodial sanctions as well, although they comprise the minority of criminal sentences.
21 This distinguishes my approach from others, such as that of Bagaric and Wolf, who would prefer to encode substantive sentencing principles into the algorithm directly:
We recommend that a constant, unvarying suite of factors that inform penalty, including aggravating and mitigating considerations that increase or decrease penalty respectively, and specifications of the weight that attach to each of those factors in certain circumstances, should be built into the computer algorithm. (“Sentencing By Computer,” 33)
22 My thanks to Aziz Huq, who has helped me sort my thoughts in this section, and to whom I owe the representations in and .
23 One could render this requirement as convergence upon the same ex ante chance of falling somewhere in the given distribution, with the understanding that the endpoints of the distribution are cabined within some fixed range.
24 See Tata, “The Application of Judicial Intelligence” and “The Struggle for Sentencing Reform,” 247–48 (Scotland). Doob and Park, writing in 1987, presciently observed that the success of sentencing reform depended upon judicial attitudes: “[i]f … judges feel comfortable sentencing in the absence of systematic information about current practice, then no information system of any kind which provides this information is likely to be of use to them” (“Computerized Sentencing Information,” 72). As Tata notes, judicial indifference ultimately doomed efforts to rationalize sentencing in Canada: see “The Application of Judicial Intelligence,” 208–212.
25 See Coroners and Justice Act 2009, §§118–132 (Eng.), where the powers and responsibilities of the Sentencing Council are set forth. Under § 125(1), courts are required to adhere to the sentencing guidelines “unless the court is satisfied that it would be contrary to the interest of justice to do so.” The existing definitive guidelines are available on the Sentencing Council’s website: https://www.sentencingcouncil.org.uk/
26 See Penal Law (Amendment No 113) 2012, 2337 LSI 170 (Israel), § 40C; the enumerated factors are listed in §§ 40(d-e) and 40(k). I am indebted to Julian Roberts for bringing this point to my attention; for discussion, see Roberts and Gazal-Ayal, “Statutory Sentencing Reform.”
27 Criminal Code, § 718. The Supreme Court of Canada has added retribution to this list of purposes. See R v CAM, [1996] 1 S.C.R. 500 (Can.).
28 See Roberts, “Structuring Sentencing in Canada,” 330.
29 In addition to the Canadian and Israeli sentencing provisions discussed above, the Model Penal Code describes the “general purposes” of its sentencing provisions as ensuring that sentences are “in all cases within a range of severity proportionate to the gravity of offenses, the harms done to crime victims, and the blameworthiness of offenders.” See Model Penal Code, § 1.02(2)(a)(i).
30 Criminal Code, §718.1. Similarly, the Israeli Penal Law describes proportionality as “the guiding principle in sentencing.” Israel Penal Law, 5737–1977, § 40(b).
31 This is a theme in Elizabeth Hinton’s recent book, From the War on Poverty.
32 Sonja B. Starr, for instance, has argued this point in the context of risk-assessment devices for pre-trial detention; see “Evidence-Based Sentencing.”
33 See Yang, “Free at Last?,” which finds evidence of greater racial disparities in the sentencing patterns of federal judges appointed after the Supreme Court made the Federal Sentencing Guidelines advisory, in United States v. Booker, 543 U.S. 220 (2005), than in the sentencing patterns of judges appointed earlier. Researchers have found that judges suffer the same kinds of implicit biases, including racial biases, as lay people. See Rachlinski et al., “Does Unconscious Racial Bias.”
34 There is a significant body of psychological literature on this subject; for an overview, see Englich, “Heuristic strategies”; Goodman-Delahunty and Sporer, “Unconscious influences in sentencing.”
35 See R v Gladue, [1999] 1 S.C.R. 688 (Can.).
36 See R v Ipeelee, [2012] 1 S.C.R. 433 62 (Can.).
37 See Kleinberg et al., “Human Decisions,” 29–32.
38 See Kleinberg et al., “Human Decisions,” Table 7.
39 See Huq, “Racial Equity”; Berk et al. “Fairness in Criminal Justice.”
40 See Huq, “Racial Equity.”
41 They would have to be alert to “broken leg” cases, that is, cases of improbable occurrences that, when they do occur, undermine even otherwise robust actuarial predictions. See Grove and Meehl, “Comparative Efficiency,” 307–8.
42 See State v. Loomis, 881 N.W.2d 749 (Wis. 2016).
43 For discussion of some of the novel due process concerns raised by technological advances of this kind, see Citron, “Technological Due Process.”
44 See Grove and Meehl, “Comparative Efficiency”; Dana, Dawes and Peterson, “Belief in the unstructured interview,” which reports that unstructured interviews actually impair judgment; Bertrand and Mullainathan, “Are Emily and Greg,” which finds decisions about which job candidates to call back for an interview were biased against applicants with stereotypically African-American-sounding names relative to applicants with stereotypically white-sounding names, such that an applicant with an African-American name needed approximately eight additional years of experience to receive as many callbacks as an applicant with a white-sounding name).
![Supercharge Your Next Research Paper: Research and write your next paper with Jenni AI.]({"type":"image" , "src" : "/sda/112443/Skyscraper-socialsciences.png"})