RISK, RACE, AND RECIDIVISM: PREDICTIVE BIAS AND DISPARATE IMPACT*

• Date: 01 November 2016
• Author: CHRISTOPHER T. LOWENKAMP,JENNIFER L. SKEEM
• Published date: 01 November 2016
• DOI: http://doi.org/10.1111/1745-9125.12123

RISK, RACE, AND RECIDIVISM: PREDICTIVE BIAS

AND DISPARATE IMPACT∗

JENNIFER L. SKEEM1and CHRISTOPHER T. LOWENKAMP2

1School of Social Welfare and Goldman School of Public Policy, University of

California—Berkeley

2Administrative Office of the U.S. Courts, Probation and Pretrial Services Office

KEYWORDS: risk assessment, race, test bias, disparities, sentencing

One way to unwind mass incarceration without compromising public safety is to

use risk assessment instruments in sentencing and corrections. Although these instru-

ments figure prominently in current reforms, critics argue that benefits in crime control

will be offset by an adverse effect on racial minorities. Based on a sample of 34,794

federal offenders, we examine the relationships among race, risk assessment [the Post

Conviction Risk Assessment (PCRA)], and future arrest. First, application of well-

established principles of psychological science revealed little evidence of test bias for

the PCRA—the instrument strongly predicts arrest for both Black and White offend-

ers, and a given score has essentially the same meaning—that is, the same probability

of recidivism—across groups. Second, Black offenders obtain higher average PCRA

scores than do White offenders (d=.34; 13.5 percent nonoverlap in groups’ scores),

so some applications could create disparate impact. Third, most (66 percent) of the

racial difference in PCRA scores is attributable to criminal history—which is already

embedded in sentencing guidelines. Finally, criminal history is not a proxy for race, but

instead it mediates the relationship between race and future arrest. Data are more help-

ful than rhetoric if the goal is to improve practice at this opportune moment in history.

During the last few years, increased awareness of the economic and human toll of

mass incarceration in the United States has launched a reform movement in sentencing

and corrections (Lawrence, 2013). This remarkably bipartisan movement (Arnold and

Arnold, 2015) is shifting public discourse about criminal justice “away from the question

of how best to punish, to how best to achieve long-term public safety” (Subramanian,

Moreno, and Broomhead, 2014: 2).

One way to begin unwinding mass incarceration without compromising public safety

is to use risk assessment instruments in sentencing and corrections. These research-based

instruments estimate an offender’s likelihood of reoffending based on various risk factors

(e.g., young age and prior arrests)—and they figure prominently in current reforms

∗The views expressed in this article are those of the authors alone and do not reflect the official

position of the Administrative Office of the U.S. Courts. Prof. Lowenkamp specifically advises

against using the PCRA to inform front-end sentencing decisions or back-end decisions about

release without first conducting research on its use in these contexts, given that the PCRA was not

designed for those purposes.

Direct correspondence to Jennifer L. Skeem, University of California—Berkeley, 120 Haviland

Hall #7400, Berkeley, CA 94720-7400 (e-mail: jenskeem@berkeley.edu).

C2016 American Society of Criminology doi: 10.1111/1745-9125.12123

CRIMINOLOGY Volume 54 Number 4 680–712 2016 680

RISK, RACE, AND RECIDIVISM 681

(Monahan and Skeem, 2016). Across the United States, statutes and regulations increas-

ingly require that risk assessments inform decisions about the imprisonment of higher

risk offenders, the (supervised) release of lower risk offenders, and the prioritization of

treatment services to reduce offenders’ risk (National Conference of State Legislators,

2015; see also American Law Institute, 2014). By implementing risk assessment at

sentencing, Virginia diverted 25 percent of nonviolent offenders from prison without

raising the crime rate (Kleiman, Ostrom, and Cheesman, 2007).

Despite such promising results, controversy has begun to swirl around the use of risk

assessment in sentencing. The principal concern is that benefits in crime control will be

offset by costs in social justice—that is, a disparate and adverse effect on racial minorities

and the poor. Although race is omitted from these instruments, critics assert that risk

factors that are sometimes included (e.g., marital history and employment status) are

“proxies” for minority race and poverty (Harcourt, 2015; Silver and Miller, 2002; Starr,

2014). In the view of Former Attorney General Eric Holder (2014), risk assessment

may exacerbate unwarranted and unjust disparities that are already far too common

in our criminal justice system and in our society. Criminal sentences must be based on

the facts, the law, the actual crimes committed, the circumstances surrounding each

individual case, and the defendant’s history of criminal conduct. They should not be

based on unchangeable factors that a person cannot control, or on the possibility of

a future crime that has not taken place. (paras. 23 and 24)

These concerns are legitimate and important—but untested. In fact, Holder (2014)

specifically urged that this issue be studied. The main issue is whether the use of risk

assessment in sentencing affects racial disparities in imprisonment, given that young

Black men are six times more likely to be imprisoned than young White men are (Carson,

2015). Risk assessment could exacerbate racial disparities, as Holder speculated. But risk

assessment could instead have no effect on—or even reduce disparities—as others have

predicted (Hoge, 2002: see also Gottfredson and Gottfredson, 1988).

It must be understood that concerns about racial disparities are more-or-less applicable

to all uses of risk assessment in sentencing and corrections. Although criticism focuses

on the use of risk assessment to inform front-end sentences that judges impose, the

same concerns are applicable to back-end sentencing decisions about release from

incarceration (earned release, parole, etc.). Regardless of the decision’s timing (front-

or back-end) or type (to release lower risk offenders or to detain higher risk offenders),

there could be a net effect of risk assessment on racial disparities in incarceration. Even

the well-established use of risk assessment to inform resource allocation in corrections

(Elek and Casey, 2015) can invoke concern. If higher risk offenders are subject to more

intensive community supervision and risk reduction services—and service refusal violates

the terms of release—they are more subject to social control than are their lower risk

counterparts.

Does risk assessment exacerbate, mitigate, or have no effect on racial disparities? The

answer to this question probably depends on factors that include the instrument chosen.

Sensationalistic headlines aside, “risk assessment” is not reducible to “race assessment”

(Sentencing Project News, 2015). Validated risk assessment instruments differ in their

purpose and in the risk factors they include (Monahan and Skeem, 2016)—and little is

known about their association with race.

682 SKEEM & LOWENKAMP

In the present study, we use a cohort of federal supervisees to test the nature and

strength of relationships among race, risk assessment scores, and recidivism. Because

existing disparities in punishment “primarily affect black Americans” (Tonry, 2012:

54), we focus on Black and White offenders. Our goal is to inform debate and provide

guidance for instrument selection and refinement. To contextualize this study, we first

highlight where risk assessment fits in corrections and sentencing, and then we unpack

controversy about particular types of risk factors.

RISK ASSESSMENT IN (COMMUNITY) CORRECTIONS

Risk assessment has been used to inform correctional decisions for nearly a century

(Administrative Office of the U.S. Courts, 2011). Early instruments were designed

to achieve efficient prediction; they generally involved scoring a set of risk markers,

weighting them by predictive strength, and combining them into a risk score that

could be used to rationalize the use of supervision resources (e.g., assigning higher risk

offenders to more intensive community supervision). Later instruments have often been

infused with the concept of risk reduction: They include variable risk factors as “needs”

to be addressed in supervision and treatment and are meant to scaffold principles of

evidence-based correctional services. These principles specify who should be treated

(those at higher risk of recidivism, given the “risk” principle) and what should be treated

(variable risk factors for crime, given the “need” principle).

Decades ago, scholars noted the potentially discriminatory effects of risk assessment

in justice settings (Petersilia and Turner, 1987) and illustrated how to remove “invid-

ious predictors” (Gottfredson and Jarjoura, 1996). Since then, little concern has been

expressed about such correctional applications. In fact, risk assessment plays a central

role in the proposed Sentencing Reform and Corrections Act of 2015, a bill before

Congress that requires that risk assessments be conducted to assign federal inmates to

appropriate recidivism reduction programs (e.g., work and education programs or drug

rehabilitation). Inmates who comply with these programs can earn early release (for up

to 25 percent of their remaining sentence).

WHERE RISK ASSESSMENT FITS IN PUNISHMENT THEORY

Front-end applications of risk assessment attract the greatest controversy. Since the

mid-1970s, sentencing in the United States has largely been a backward-looking exercise

focused on an offender’s moral blameworthiness for the conviction offense, in keeping

with retributive theories of punishment (Monahan and Skeem, 2016). During the last

few years, sentencing reform has reflected a resurgence of interest in incorporating

forward-looking assessments of an offender’s risk of future crime, in keeping with

utilitarian or crime control theories of punishment.

Currently, risk assessment is considered—and in our view should be considered—

within bounds set by moral concerns about culpability (Monahan and Skeem, 2014).

This is consistent with the leading model of criminal punishment (Frase, 2004)—a hybrid

of retributive and utilitarian theories called “limiting retributivism” (Morris, 1974). As

operationalized in the Model Penal Code (American Law Institute, 2014), sentencing

takes place “within a range of severity proportionate to the gravity of offenses, [and]

the blameworthiness of offenders.” Within this range, a sentence is chosen to promote