The Case Against Pretrial Risk Assessment Instruments, Pretrial Justice Institute, 2020

• Locations: United States of America • Topics: Bail/Pretrial Release

Download original document:
Brief thumbnail

Document text

This text is machine-read, and may contain errors. Check the original document to verify accuracy.

P.11
•

•

Pretrial Justice Institute

PRETRIAL
JUSTICE
INfflTUTE

November, 2020

The Case Against

Pretrial Risk Assessment
Instruments
Pretrial risk assessment instruments (RAIs) are actuarial tools

KEY TAKEAWAYS
n

Pretrial risk assessment instruments (RAIs) are constructed
from biased data, so the
RAIs perpetuate racism.

RAIs are not able to accurately
predict whether someone will
flee prosecution or commit a
violent crime.

RAIs label people as “risky”
even when their odds of
success are high.

RAI scores inform conditions
of release, but there is no
proven connection between
RAI scores, specific conditions, and pretrial success.

intended to estimate two key outcomes for those who are released
pending trial: the likelihood that someone will fail to appear for
court, and the likelihood that someone will be arrested for a new
crime before the disposition of their case. The use of RAIs is a topic
of great debate among criminal justice practitioners, community
advocates, formerly incarcerated people, policymakers, and academics. Some see them as an important tool for facilitating pretrial
release, while others argue that they perpetuate racial bias and
unfairly label people as “risky.”
In 2020, after an extensive period of reflection, the Pretrial Justice
Institute released a statement opposing RAIs. First and foremost,
that position was driven by a strengthening commitment to racial
equity and a conviction that the tools do not adequately address
the biases inherent in the system. Underscoring this new position,
though, was the understanding, based on research, that these tools
are not able to do what they claim to do—accurately predict the
behavior of people released pretrial and guide the setting of conditions to mitigate certain behaviors. RAIs simply add a veneer of
scientific objectivity and mathematical precision to what are really
very weak guesses about the future, based on information gathered
from within a structurally racist and unequal system of law, policy
and practice. This paper interrogates the role that RAIs are supposed
to play in advancing pretrial justice, and how they fall short.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 1

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

RAIs CANNOT RELIABLY PREDICT FLIGHT FROM PROSECUTION

The most basic tenet of pretrial justice is that people need to appear in court to face the charges
against them. The initial concern, centuries-old, was that people would flee the jurisdiction to
avoid prosecution, and the purpose of bail was to address this risk of flight. But RAIs are not
constructed to estimate the likelihood of flight.1 Instead, they are constructed to estimate the
likelihood of a “failure to appear”—and they typically treat past failure to appear, regardless of
the reason, as a valid predictor of future failure.
RAIs do not—cannot—distinguish between an actual flight from prosecution and missed court
appointments. The omission of “true flight” from RAIs reflects the use of readily available data
such as failures to appear or bond forfeitures, which do not make distinctions among types of
nonappearance, and the fact when “tools promise only to predict nonappearance broadly, they
can claim greater success than if the tools purported to predict the narrower and more serious
categories of risk.”2
RAIs do not distinguish between people who have the means to flee the jurisdiction to
avoid justice and people without resources who have problems coming to court due
to transportation or housing instability, child care or employment issues. The inclina-

tion and ability to flee prosecution is a relatively rare event in criminal court.3 Data show that
only three percent of all released people charged with felonies do not appear in court at all and
are not returned to court after a year.4 More plainly, most people lack the means to flee. Over
one-third of people (36%) who had one arrest in the preceding 12 months had annual incomes
below $10,000. That percentage increased to nearly half (49%) among people who are arrested
multiple times in a year—a population that represents more than one-fourth of people who are
jailed.5
People who are poor face greater obstacles to appearing in court than those who have greater
access to resources. Forgoing paid time, risking employment, finding childcare, and accessing
reliable transportation are just some of the barriers experienced by people with low-paying jobs
and few resources. This issue is compounded by higher rates of poverty among Black, Latinx,

RAIs simply add a veneer of scientific objectivity and
mathematical precision to what are really very weak
guesses about the future.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 2

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

and Indigenous people. Acknowledging and addressing these barriers could raise court appearance rates.
Even if RAIs could be revised to focus on the risk of willful flight, they cannot predict this behavior to a level of accuracy that justifies the restriction of liberty on individuals. Consider that the
predictive accuracy for the failure to appear model of the Public Safety Assessment, developed by
Arnold Ventures, is 0.644, where 0.5 is considered a coin flip and 1.0 is perfect prediction.6 While
some have argued that in the criminal justice arena, 0.64 to 0.7 is acceptable predictive power
for pretrial assessments,7 many disciplines consider a score of less than 0.7 to have poor discriminatory power.8 Rather than accept this lower standard of predictive power, it should be rejected
as a basis for depriving people of their right to pretrial liberty and presumption of innocence. The
mere existence of this information perpetuates the conflation of flight with nonappearance, and
encourages the corresponding use of incarceration or restrictive conditions.

II. RAIs CANNOT RELIABLY PREDICT VIOLENT CRIME
The second basic tenet of pretrial justice is protecting the public from danger. The concern
is that people will commit an act of violence in the community while awaiting trial. But RAIs
do not actually estimate the likelihood of this outcome. Instead, most existing pretrial RAIs
estimate the likelihood of any new arrest, regardless of whether it involves an accusation of
violence. According to the Court Statistics Project, in 2018, misdemeanors made up 77% of
criminal cases in state trial courts, and felonies made up 23%. Of those cases, a small percentage made up “person” cases, i.e., those cases involving bodily harm, including homicide. Person
cases made up 9% of misdemeanor cases, and 18% of felony cases.9
Because violence is rare, RAIs do not—cannot—reliably predict future arrest for violence. Being
charged with an act of violence while awaiting trial has always been a second “statistically rare
event” that cannot be accurately predicted. In Washington, DC, where 88 percent of all people
are released before trial, the percentage of people who remain arrest-free is 89%. Of those who
are released, only 1% are arrested for a violent crime.

Even if RAIs could be revised to focus on the risk of
willful flight, they cannot predict this behavior to a
level of accuracy that justifies the restriction of liberty
on individuals.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 3

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

When RAIs do attempt to identify people who are more likely to commit a crime of violence,
rates of rearrest for violence are quite low even among the “high risk” group. The Public Safety

Assessment, developed by Arnold Ventures, contains a flag suggesting someone is at a higher
risk for being arrested for what they label a “New Violent Criminal Activity.” However, more
than 9 times out of 10, people who receive that flag are not arrested for a violent offense
while on pretrial release.10 High risk designations are also biased against Black people.
A study of 175,000 people arrested in New York City found that looking at a hypothetical
classification of high risk and no actual re-arrest, 23% of Black defendants would have been
classified as high-risk and flagged for detention, compared with 17% of Hispanic defendants,
and only 10% of white defendants.11

III. FACTORS EMPLOYED BY RAIs REFLECT AND PERPETUATE STRUCTURAL RACISM
RAIs are presented as “objective” or “race neutral” because the items are statistically
validated as predictive of court appearance and new crime. However, when the data used for
that statistical analysis reflect bias against Black, Latinx, Indigenous and poor people, the
resulting tool is inherently biased as well.
Structural and individual biases in policing practices make it more likely that Black people
will be stopped, searched, subjected to force and arrested than white people for the same
behavior.12 RAI developers have attempted to mitigate bias using statistical techniques
that identify racial differences in specific items on the tool. These analyses have resulted in
incremental improvements by removing items that are overtly biased, like whether someone
owns or rents their home. Still, a tool that can accurately “predict” biased outcomes, such as
the likelihood of arrest, is more of a reflection of the system than it is of the person.13
Each RAI uses a different combination of factors and weighting of factors. The table on the
next page includes some commonly used factors and how each factor reflects bias.

When RAIs do attempt to identify people who are more
likely to commit a crime of violence, rates of rearrest for
violence are quite low even among the “high risk” group.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 4

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

Bias in Pretrial Risk Assessment Instruments
FACTOR

HOW IT IS BIASED

Age at first
arrest

Police are more likely to arrest Black people, even after controlling for factors such
as seriousness of the offense and prior record, according to a meta-analysis of 23
research studies looking at arrests between 1977 and 2004.14 Thirty percent of
Black men have experienced at least one arrest by age 18, compared to 22% of
white males. This differential increases with age.15

Current charge

A 2020 study from Harvard Law School found that one factor, racial/ethnic
differences in the initial charge, accounted for 70 percent of racial disparities in
sentence length; which were approximately 5 months longer for Black and Latinx
people than white people, among those sentenced to incarceration.16 A 2017
study of 48,000 misdemeanor and felony cases in Wisconsin found that white
people were 25 percent more likely to have their top charge dropped or reduced
by prosecutors than Black people.17

Employment

Employment as a factor is highly biased against Black people. Since 1954, when
the Bureau of Labor Statistics began tracking these numbers, the unemployment
rate for Black people has consistently been twice that of white people.18

Drug use

Black and white people use and sell drugs at approximately the same rates, but
Black people are 2.7 times more likely to be arrested for drug-related offenses.19
Moreover, Black and Latinx people are more likely to be incarcerated for drug-related offenses than white people.20

Pending charge
at time of arrest

A study of 10,753 cases in San Francisco over a three-year period revealed cases
involving Black people take longer to resolve. On average, cases for Black people
took 90 days to process, compared to 77.5 days for white people (making it more
likely for a Black person that a charge would be pending in the event of arrest).21

Prior convictions
(for violence)

People with felony convictions (which do not exactly match to arrests for violence)
account for 8% of all adults and 33% of the Black adult male population, and likely
reflect police sweeps and mass arrests executed disproportionately in neighborhoods of color.22 At the same time, Black people are more likely to be wrongfully
convicted than white people. Half of people exonerated for murder and 59% of
people exonerated for sexual assault are Black people; ninety percent of people
framed in large-scale police scandals are Black.23

Prior sentence to
incarceration

A study of men charged with felonies in urban U.S. counties calculated that the
cumulative effects of bias in the criminal legal system made the probability of
going to prison 26% higher for a Black man, and 30% for a Latino, than that of
a white man charged with a felony.24 A literature review on race and sentencing
found that Black and Latino people were much more likely to be disadvantaged in
the decision to incarcerate than whites, and this initial disparity was greater than
the subsequent decision around how long to incarcerate.25

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 5

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

IV. RAI SCORES PUT A MISLEADING EMPHASIS ON CALCULATED RISK
Constructors of RAIs have always reminded users that RAIs are actuarial tools and do not give
the court specific, individualized information about the person facing charges. They create
categories that reflect their similarity to others “like them.” As shown above, it is not possible
to reliably predict willful flight or arrest for a violent crime, so RAIs are designed to estimate
the likelihood of any missed court appearance or new arrest. Many RAIs even combine risk
of missing court and risk of any rearrest into a single risk of “pretrial failure.” Most
RAIs focus on the likelihood of failure rather than the likelihood of success, which creates a
misleading perception of risk, and combining risk of missing court and risk of any arrest does
not decrease that likelihood. As shown earlier, many factors are biased against Black and Latinx
people.
Within the label of “risk,” RAIs rate the probability of success or failure using various units.
Some have just three levels of risk, typically labeled “Low,” “Medium,” and “High,” while
others have four, five or six levels of risk. Either way, these boundaries are artificially drawn.
They can be cut-points set by the researchers who conducted the study that developed the
tool, or they can be negotiated by a policy team based on local risk tolerance. Cut-points may
also be based on interests of court systems that are unrelated to the rights of an accused individual, such as case flow needs or jail population management. One group may set the boundaries of each risk level one way, and another group may set them another way.
In California, for example, four counties developed their own RAIs; 18 counties use the Virginia
Pretrial Risk Assessment Instrument; 17 counties use the Ohio Risk Assessment System tool;
four counties use the COMPAS (Correctional Offender Management Profiling for Alternative
Sanctions) and two counties use the Public Safety Assessment.26 This patchwork shows how
variable justice can be even though RAIs purport to be “evidence-based.” Two similarly-situated
people, just a few miles apart, will be treated differently depending on what county they are in.

When the data used for that statistical analysis reflect
bias against Black, Latinx, Indigenous and poor people,
the resulting tool is inherently biased as well.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 6

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

Court Appearance and Arrest Rates Based on Risk Levels
Colorado presents its rates in terms of public safety (no arrest) and court appearance. New York looks
only at court appearance. Other tools consider both arrest rates and failure to appear, or a combination
of the two.
Name of Tool

Lowest Risk

Medium Risk

Highest Risk

91% – Public safety rate
95% – Court appearance
rate

69–80% – Public safety rate
77–85% – Court appearance rate
(representing the two
middle categories)

58% – Public safety rate
51% – Court appearance rate

For all charge types, release
on recognizance is recommended for those whose
scores suggest appearance
rates of 82.3–93%.

Depending on the charge
type (misdemeanor, nonviolent felony, or violent
felony), release on recognizance or consideration of all
options may be considered
for those whose scores
suggest appearance rates
of 71–76%.

For all charge types, release
on recognizance is not recommended for those whose
scores suggest appearance
rates of 41.7–63%.

0.3% – Arrest for violent
offenses
0.7% – Failure to appear

1.3% – Arrest for violent
offenses
2.5% – Failure to appear

2.9% – Arrest for violent
offenses
4.6% – Failure to appear

3.9% – New criminal arrest
7.5% – Failure to appear

10.9–15.1% – New criminal
arrest
13.9–19.8% – Failure to
appear
(representing the two
middle categories)

26.3% – New criminal arrest
32.1% – Failure to appear

6.1% – Any failure rate

14.9–21.4% – Any failure rate

37.1% – Any failure rate

RAIs based on likelihood of
success
Colorado Pretrial Assessment
Tool (CPAT)
CPAT has four risk categories
New York City Criminal Justice
Release Assessment
New York considers
appearance only
This assessment has four
recommendations based on
the score.
RAIs separating risk of arrest
and failure to appear
Federal Pretrial Risk Assessment (PTRA)
PTRA has five risk categories
Public Safety Assessment
(PSA)
As validated on data from
Kentucky
PSA has six risk categories
RAIs combining risk of
arrest and failure to appear
Virginia Pretrial Risk Assessment Instrument – Revised
(VPRAI-R)

(representing the two
middle categories)

VPRAI-R has six risk categories
Ohio Risk Assessment System
- Pretrial Assessment Tool
(ORAS-PAT)

5% – Any failure rate

18% – Any failure rate

29% – Any failure rate

ORAS-PAT has three risk
categories

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 7

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

V. RISK ASSESSMENTS CAN DRIVE UNNECESSARY AND UNPROVEN SUPERVISION
(AND DETENTION)
How “risk levels” are defined has serious implications for those being assessed. Some state
statutes or court rules require certain actions or decisions based on whether a person scores a
“Low,” “Medium,” or “High” on the RAI. This has a cascading effect; a higher risk designation
can result in more intensive levels of supervision, an increased likelihood of returning to jail on
a technical violation, and higher fees related to supervision conditions—all consequences that
have very real impacts on people’s lives. Most counties with pretrial services reported that they
charged people fees for some type of pretrial service, such as drug testing, surveillance technologies or supervision, which can lead to a cycle of debt.27 Very little data exists about how race
factors into the setting and enforcement of non-financial pretrial release conditions.
Subjective designations of high, medium or low risk are used to legitimize conditions
of pretrial release, such as supervision, drug testing, and electronic monitoring, and
contribute to our nation’s mass supervision and surveillance crisis. In most jurisdic-

tions, the results of an RAI are applied in a decision making framework or matrix to determine
the level of supervision. Typically, the score and the most serious charge are used together to
determine the supervision/surveillance to which a person will be subjected.
Notably, courts should not use the results of an RAI to determine detention; that decision
should be made in a separate hearing with full due process protections. However, according to
the PJI 2019 Scan of Pretrial Practices, many jurisdictions do use the RAI to determine detention. Nearly three out of four counties (73 percent) that had a pretrial assessment tool reported
using them to make the “release or detain” decision.28
There is little research demonstrating that supervision conditions actually improve court
appearance and public safety; what information is available on specific conditions, such as
drug testing and electronic monitoring, shows they are not effective but are rapidly expanding
in their usage. On a more nuanced level, there is no research showing that specific conditions
are appropriate for people with specific RAI scores. Instead, the type and dosage of supervision

If the pretrial field took the approach that no
specific condition could be assigned unless it was
proven effective, then supervision and surveillance
would be virtually obsolete.
pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 8

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

are determined by policy, and can vary widely across jurisdictions. If the pretrial field took the
approach that no specific condition could be assigned unless it was proven effective, then
supervision and surveillance would be virtually obsolete.29
Even the color-coding used by decisonmaking frameworks or matrices seem designed
to provoke alarm. Below is an example of a matrix used by Denver Pretrial Services Investi-

gation Unit as of 2018. Note the use of the color red for maximum supervision. Moreover, the
levels of supervision shown were created by policy makers and practitioners in Colorado, not
derived from any research showing these specific conditions are related to any “risk” purportedly
assessed.

Denver Pretrial Supervision Guidelines-Misdemeanor and Felony Offenses
Primary Charge and CPAT Category

Enhancers
(move up one level of supervision}

➔

•NonVRA
Misdemeanors

•NonVRA
Felonies

➔

Currently supervised on felony probation, parole, or pretrial

supervision for any criminal offense

•VRA
M isdemeanors

•M isdemeanor
DV

•VRA Felony
Crimes

•Indecent
Exposur e

•Felony DV

Two or more pending felony cases or misdemeanor assault cases
within one yeor of current offense date

•DFl

•

Offense Involves a knife or a firearm in current charge

•
•

High OOARA score (7+}
Felony Child Abuse or Felony Sex Assault on o Child

•Burglary of
a Dwelling

•

•Felony Sex
Offenses

Category 1
score: a to 11 (87" Success)
91" Public Sa
earonce

Category 2
Score: 18 to 37 (71"Success)
80" Public So
eoronce

Category 3
Score: 38 to SO (58" Success)

Category 4

Enhanced (Enh)

No Pretrial Services Supervision
Statuto

Conditions Onl

Notification of new arrest

Court Reminder Calls
Notification of new arrest

Check-in physically after court appearances

Court Rem inder calls
Telephone check ins after court

lnteNlve w/EM Monltorlnc
Int

a ea rances
Telephone check ins
1 to 4x per month (30 days)

Alcohol/ GPS/ Home Monitoring•

Check•in physically after court appearances

Curfew/Employment Leave Only•
Case Management meetings
1 to 2x per month (30 days)
Substance Testin if ordered

Case Management meetings
1 to 4x per month (30 days)
Tele hone Check Ins As Needed
Substance Testin if ordered

Denver Pretrial Decision Making Framework, as presented to
the Colorado Supreme Court Bail Blue Ribbon Commission, August 16, 2018.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 9

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

The Virginia Pretrial Risk Assessment Instrument-Revised (VPRAI-R) Praxis, as presented to
the Pretrial Services Stakeholder Group, June 11, 2018, contains six different types of outcomes,
raising questions of whether these modifications actually make a difference. The outcomes
include release with no supervision, release with supervision monitoring (court reminder and
criminal history check before every court date), three levels of supervision monitoring with
increasing frequency and compliance verification, and detention.

■
Levell

(0-2 points)

Level2
(3-4)

Level3
(5-6)

Level4
(7-8)

Levels
(9-10)

Level6
{11-14)

Non-Violent
Misdemeanor

Driving Under the
Influence

Non-Violent Felony

Violent
Misdemeanor

Violent Felony
or Firearm

Release

With no Supervision,
(Monitoring if
Supervision Ordered)

With no Supervision,
(Monitoring if

With no Supervision,
(Monitoring if
Supervision Ordered)

With Supervision
Level II

Supervision Ordered)

Release

With no Supervision,
(Monitoring if
Supervision Ordered)

With Supervision
Monitoring

With Supervision
Level Ill

Release

Detain

With Supervision
Monitoring

With Supervision
Level I

(Level Ill if Supervision
Ordered)

Release

Detain

With Supervision
Level I

With Supervision
Level II

(Level Ill if Supervision
Ordered)

Release

Detain

With Supervision
Level II

With Supervision
Level Ill

(Level Ill if Supervision
Ordered)

Detain

(Level Ill if Supervision
Ordered)

Praxis accompanying the Virginia Pretrial Risk Assessment Instrument - Revised (VPRAI-R)

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 10

CONCLUSION
It is time to transform pretrial reform. As an organization that was deeply involved in the proliferation of RAIs, PJI has spent years wrestling with the research on pretrial risk prediction. Our
central concerns now focus on the civil rights implications of the tools, the items on them, and the
ways in which courts are using these and other tools of reform that do not prioritize racial equity.
One major tenet in a racial equity transformation is reckoning with intention versus impact.
What was intended to support courts in better making decisions that honored the presumption
of innocence and held “detention as the carefully guarded exception” has had a devastating
impact on Black, Indigenous and Latinx communities. The consequences are reflected in the
data on mass pretrial detention and mass pretrial surveillance, often followed by forced pleas30
or dropped charges.31
If RAIs cannot predict the risk of fleeing prosecution and risk of violence against a specific
person or group; and they reflect systemic issues such as difficulty in keeping court appointments, case processing, and over-policing; and they are being used to legitimize an expansion
of surveillance, then we have an obligation to interrogate their use. Even relatively successful
attempts to produce “race equal” outcomes in RAIs cannot address the racialization of what is
criminalized and how we police. Subjective assessment of the very same items on an RAI—due
to requirements in statute or court rules, or simply based on professional experience—will only
reproduce these effects in a less transparent manner.
Pretrial justice requires a radical reconstruction, which prioritizes racial equity
in the presumption of innocence and pretrial liberty, and commits to long-term
racially equitable solutions.32 The Leadership Conference on Civil and Human Rights and
the Civil Rights Corps have proposed a vision that prioritizes upfront investments in community
programs and social services, including additional resources for education, housing, employment,
health care, social-emotional supports, while protecting the pretrial rights of people and requiring
collection of pretrial data relating to release, detention and race to empower communities to make
change. This broader view of what constitutes public safety is not only more equitable, but has the
potential to make our communities safer in the short and long-term by providing opportunities to
thrive and prosper.
It is time to put away RAIs and forge an approach that does not perpetuate racial inequality,
court involvement, debt or poverty, or create barriers to pretrial liberty and the presumption
of innocence. The focus should be on implementing a very narrow detention net and providing
robust detention hearings that honor the charge of the Supreme Court forty years ago.33 And we
must prioritize helping people succeed—from assistance with court appointments to connecting
people to support services—while addressing the needs of all people victimized by crime.

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 11

P.11
•

•

PRETRIAL
JUSTICE
INfflTUTE

ENDNOTES
1
Gouldin, L. (2018). Defining Flight Risk. University of
Chicago Law Review 85, 677, 716.
2
Id. at 722.
3
Id. at 710. Gouldin argues that examining ability to flee
requires consideration of ability and inclination to flee. Gouldin notes examples where courts found resources such as a
foreign passport, real property, bank accounts, access to large
sums of cash and foreign contacts to be persuasive examples
of ability to flee. “Successful fight from jurisdiction suggests
access to networks and resources that are not part of the
equation for the vast majority of nonappearing defendants.”
4
Id. at 689, citing Reaves, B.A. (2013). Felony Defendants
in Large Urban Counties, 2009—Statistical Tables. Bureau of
Justice Statistics.
5
Jones, A. and Sawyer, W. (2019). Arrest, Release, Repeat: How police and jails are misused to respond to social
problems. Prison Policy Initiative.
This measurement is known as the Area Under the
6
Curve (AUC) Receiver Operator Characteristics (ROC) estimate. DeMichele, M., Baumgartner, P., Wenger, M., Barrick,
K., Comfort, M. & Misra, S. (2018). The Public Safety Assessment: A Re-Validation and Assessment of Predictive Utility
and Differential Prediction by Race and Gender in Kentucky.
7
Desmarais, S. L., and Singh, J P. (2013). Risk assessment instruments validated and implemented in correctional
settings in the United States. Lexington, Kentucky: Council
of State Governments; Desmarais, S., Johnson, K. and Singh,
J. (2016). Performance of recidivism risk assessment instruments in U.S. correctional settings. Psychological Services.
8
Hosmer Jr., D.W., Lemeshow, S. and Sturdivant, R.X.
(2013). Applied Logistic Regression. 3rd Edition, John Wiley
& Sons, Hoboken, NJ.
See, State Court Caseload Digest. (2020). 2018 Data.
9
Court Statistics Project.
10 Results from First Six Months of the Public Safety Assessment - Court in Kentucky. (2014). Laura and John Arnold
Foundation.
11 Picard, S., Watkins, M., Rempel, M. & Kerodal, A.
(2019). Beyond the Algorithm: Pretrial Reform, Risk Assessment, and Racial Fairness. Center for Court Innovation.
12 Mayson, S.G. (2019). Bias In, Bias Out. 128 Yale Law
Journal, vol. 128, 2218.
13 Id.
14 Kochel, T.R., Wilson, D.B., & Mastrofski, S.D., (2011).
Effect of Suspect Race on Officers’ Arrest Decisions. Criminology 49(2), 473-512, 490 & 495-96.
15 Brame, R., Bushway, S.D., Paternoster, R., & Turner,
M.G. (2014). Demographic Patterns of Cumulative Arrest
Prevalence By Ages 18 and 23, Crime and Delinquency Vol.
60(3), 471-486.
16 Bishop, E.T., Hopkins, B., Obiofuma, C. & Owusu, F.
(2020). Racial Disparities in the Massachusetts Criminal Justice System,” Criminal Justice Policy Program, Harvard Law
School.

17 Berdejó, C. (2018). Criminalizing Race: Racial Disparities in Plea Bargaining. Boston College Law Review 59, 1187.
18 Rates of Drug Use and Sales, by Race; Rates of Drug
Related Criminal Justice Measures, by Race. (2016). The
Hamilton Project.
19 Desilver, D. (August 21, 2013). Black unemployment rate
is consistently twice that of whites. FactTank, Pew Research.
20 Bishop et al. (2020).
21 Owens, E., Kerrison, E.M., & Da Silveira, B.S. (2017). Examining Racial Disparities in Criminal Case Outcomes among
Indigent Defendants in San Francisco. Quattrone Center for
the Fair Administration of Justice.
22 Shannon, S.K.S., Uggen, C., Schnittker, J. et al. (2017).
The Growth, Scope, and Spatial Distribution of People With
Felony Records in the United States, 1948–2010. Demography 54, 1795–1818. Both of these percentages have grown
steadily since 1980, due to the federally-funded “war on
drugs” strategy leading to a disproportionate number of arrests of Black people. See also, Human Rights Watch. (2000).
UNITED STATES Punishment and Prejudice: Racial Disparities in the War on Drugs; Human Rights Watch. (2009). Decades of Disparity Drug Arrests and Race in the United States.
See also, for example, discussion of racially biased arrests
around drug charges. Ferrer, B., & Connolly, J. M. (2018). Racial Inequities in Drug Arrests: Treatment in Lieu of and After
Incarceration. American Journal of Public Health, 108(8),
968–969.
23 Gross, S.R., Possley, M. and Stephens, K. (2017). Race
and Wrongful Convictions in the United States. National
Registry of Exonerations.
24 Sutton, J.R. (2013). Structural bias in the sentencing of
felony defendants. Social Science Research 42, 1207-1221.
25 Sutton, J.R. (2013). Structural bias in the sentencing of
felony defendants. Social Science Research 42, 1207-1221.
26 Harris, H., Goss, J., & Gumbs, A. (2019). Pretrial Risk
Assessment in California. Public Policy Institute of California.
27 Scan of Pretrial Practices. (2019). Pretrial Justice Institute.
28 Id.
29 The single pretrial practice that research has consistently shown to improve court appearance is court reminders.
See, Use of Court Date Reminder Notices to Improve Court
Appearance Rates. (2017). National Center for State Courts’
Pretrial Justice Center for Courts.
30 Petersen, N. (2020). Do Detainees Plead Guilty Faster?
A Survival Analysis of Pretrial Detention and the Timing of
Guilty Pleas. Criminal Justice Policy Review, 31(7):10151035.
31 Starger, C. (2020). The Argument that Cries Wolfish.
MIT Computational Law Report.
32 King County, Washington county council and executive
rejected the use of RAIs in its Superior Courts in 2012, partly
out of concerns that it would perpetuate racial disproportionality in the jails.
33 U.S. v. Salerno, 481 U.S. 739 (1987).

pretrial.org | The Case Against Pretrial Risk Assessment Instruments | November, 2020 | 12