Practitioner?s Guide to COMPAS Core

 

 

 

MARCH 19, 2015

POINTE

Contents
Table of Contents

i

1 Introduction
1.1
Overview for Practitioners . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
2

2 Case
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9

Interpretation
Introduction . . . . . . . . . . . . . . . . . . . .
COMPAS Scores . . . . . . . . . . . . . . . . .
Levels of Interpretation . . . . . . . . . . . . . .
Criminological Theories . . . . . . . . . . . . . .
AIPIE . . . . . . . . . . . . . . . . . . . . . . .
Basic Descriptive Information for the Scales . .
Conversion of Raw Scale Scores to Decile Scores
Interpreting Decile Scores . . . . . . . . . . . .
Norm Groups . . . . . . . . . . . . . . . . . . .

3 COMPAS Validity and Reliability
3.1
Predictive Validity of the COMPAS Risk Scales
3.2
Validity of COMPAS Core Needs Scales . . . . .
3.2.1 Criterion Validity . . . . . . . . . . . . . .
3.2.2 Construct Validity . . . . . . . . . . . . .
3.2.3 Content Validity . . . . . . . . . . . . . .
3.3
Internal Consistency Reliability . . . . . . . . .
3.4
Test-Retest Reliability . . . . . . . . . . . . . .
4 Treatment Implications for Scales
4.1
Risk Scales . . . . . . . . . . . . . . . .
4.1.1 Pretrial Release Risk . . . . . . .
4.1.2 General Recidivism . . . . . . . .
4.1.3 Violent Recidivism . . . . . . . .
4.1.4 Recidivism Risk Screen . . . . . .
4.1.5 On Counter-Intuitive Predictions
4.2
Criminogenic Need Scales . . . . . . . .
4.2.1 Cognitive Behavioral . . . . . . .
4.2.2 Criminal Associates/Peers . . . .
4.2.3 Criminal Involvement . . . . . . .
i

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

3
3
4
4
5
7
7
8
11
11

.
.
.
.
.
.
.

12
12
16
16
19
22
23
25

.
.
.
.
.
.
.
.
.
.

26
26
27
27
28
28
28
32
32
32
33

 ii
4.2.4
4.2.5
4.2.6
4.2.7
4.2.8
4.2.9
4.2.10
4.2.11
4.2.12
4.2.13
4.2.14
4.2.15
4.2.16
4.2.17
4.2.18
4.2.19
4.2.20

Criminal Opportunity . . . . . . . . .
Criminal Personality . . . . . . . . . .
Criminal Thinking Self-Report . . . . .
Current Violence . . . . . . . . . . . .
Family Criminality . . . . . . . . . . .
Financial Problems . . . . . . . . . . .
History of Non-Compliance . . . . . .
History of Violence . . . . . . . . . . .
Leisure/Boredom . . . . . . . . . . . .
Residential Instability . . . . . . . . .
Social Adjustment . . . . . . . . . . .
Social Environment . . . . . . . . . . .
Social Isolation . . . . . . . . . . . . .
Socialization Failure . . . . . . . . . .
Substance Abuse . . . . . . . . . . . .
Vocation/Education . . . . . . . . . .
The Lie Scale and Random Responding

5 Typology
5.1
Interpretation . . . . . .
5.2
Male Typology . . . . .
5.2.1 Type Descriptions .
5.3
Female Typology . . . .
5.3.1 Type Descriptions .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
Test

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

Bibliography

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

34
35
35
36
36
37
38
39
39
40
41
42
42
43
43
44
45

.
.
.
.
.

47
48
50
50
55
55
58

➞2015 Northpointe Inc., All Rights Reserved.

 Chapter 1
Introduction
The Practitioner’s Guide provides an overview of the COMPAS Core Module in the Northpointe Suite. The Northpointe Suite is an integrated web-based assessment and case management system for criminal justice practitioners. The Northpointe Suite has modules designed
for pretrial, jail, probation, prison, parole and community corrections applications. COMPAS
Core is designed for both male and female offenders recently removed from the community
or currently in the community. The Practitioner’s Guide to COMPAS Core covers case interpretation, validity and reliability, and treatment implications. Most of the information
provided is specific to COMPAS Core. Throughout this text we use the term COMPAS
Core to distinguish an element (scale, typology, decile type) specific to COMPAS Core from
general elements in the Northpointe Suite, such as scales found in both COMPAS Core and
COMPAS Reentry.
COMPAS is a fourth generation risk and needs assessment instrument. Criminal justice
agencies across the nation use COMPAS to inform decisions regarding the placement, supervision and case management of offenders. COMPAS was developed empirically with a focus
on predictors known to affect recidivism. It includes dynamic risk factors, and it provides information on a variety of well validated risk and needs factors designed to aid in correctional
intervention to decrease the likelihood that offenders will reoffend.
COMPAS was first developed in 1998 and has been revised over the years as the knowledge base of criminology has grown and correctional practice has evolved. In many ways
changes in the field have followed new developments in risk assessment. We continue to
make improvements to COMPAS based on results from norm studies and recidivism studies
conducted in jails, probation agencies, and prisons. COMPAS is periodically updated to
keep pace with with emerging best practices and technological advances.
COMPAS has two primary risk models: General Recidivism Risk and Violent Recidivism
Risk. COMPAS has scales that measure both dynamic risk (criminogenic factors) and static
risk (historical factors). Additional risk models include the Recidivism Risk Screen and the
Pretrial Release Risk Scale.
Statistically based risk/needs assessments have become accepted as established and valid
methods for organizing much of the critical information relevant for managing offenders
in correctional settings (Quinsey, Harris, Rice, & Cormier, 1998). Many research studies
1

 2

have concluded that objective statistical assessments are, in fact, superior to human judgment (Grove, Zald, Lebow, Snitz, & Nelson, 2000; Swets, Dawes, & Monahan, 2000). COMPAS is a statistically based risk assessment developed to assess many of the key risk and
needs factors in adult correctional populations and to provide information to guide placement
decisions. It aims to achieve these goals by providing valid measurement and concise organization of important risk/need dimensions. Northpointe recognizes the importance of case
management and supports the use of professional judgment along with actuarial risk/needs
assessment. Following assessment, a further goal is to help practitioners with case plan
development/implementation and overall case management support.
In overloaded and crowded criminal justice systems, brevity, efficiency, ease of administration
and clear organization of key risk/needs data are critical. COMPAS was designed to optimize
these practical factors. We acknowledge the trade-off between comprehensive coverage of
key risk and criminogenic factors on the one hand, and brevity and practicality on the
other. COMPAS deals with this trade-off in several was; it provides a comprehensive set of
key risk factors that have emerged from the recent criminological literature, and it allows
for customization inside the software. Therefore, ease of use, efficient and effective time
management, and case management considerations that are critical to best practice in the
criminal justice field can be achieved through COMPAS.

1.1

Overview for Practitioners

COMPAS Core is comprised of a total of forty-three scales, including four higher order
scales that use items from several domains and seventeen scales from the women’s risk
and needs assessment (WRNA) developed by Van Voorhis, Wright, Salisbury, and Bauman
(2010). This document provides an overview of COMPAS Core. Supplemental materials are
available that provide details about the scales not covered in the Practitioner’s Guide (see,
e.g., “Measurement and Treatment Implications of the COMPAS Core Scales”).
The COMPAS Core assessment is designed to be configurable by the user at decision points
within the local criminal justice system and with different populations. For example, Pre-trial
Services may choose to use only the Pretrial Release Risk Scale to make recommendations to
the court regarding pre-trial release. Probation may then use the Violent Recidivism Risk and
General Recidivism Risk Scales to “triage” their caseloads by recidivism risk, and choose to
complete the full assessment only on the higher risk individuals. The full assessment provides
a holistic view of the person to address supervision and treatment needs for rehabilitation.

➞2015 Northpointe Inc., All Rights Reserved.

 Chapter 2
Case Interpretation
2.1

Introduction

This chapter gives an introduction to the interpretation of a COMPAS assessment. After
completing an assessment in the COMPAS software, the practitioner will generally interpret
the bar chart that displays scale scores. The bar chart indicates in what areas the person
scores higher or lower – that is, which risks or criminogenic needs may exist. The practitioner
will also interpret the type assigned by the typology if enabled by the site. The implications
for treatment and intervention are discussed in chapter 4.
Collecting assessment information is important, yet the information is only helpful when we
can make sense of it and understand how it can inform our case planning and interaction
with the offender. Interpretation skills and activities include accessing and using:
1. The assessment results
2. The criminological theories used in COMPAS
3. The Typologies
A model that everyone can relate to is the medical model for interpretation of information
gathered on a person. Think about the different steps taken in the medical field to find a
solution to an illness or a problem. When you don’t feel well and you go to the doctor, what
is the first thing that the doctor does – Asks about symptoms, when did they start, how
severe are they? She asks about your medical history, are you taking any medications, have
you had this or a similar problem before? And, she runs tests, takes your temperature, takes
your blood pressure, blood tests, MRIs, etc. What does she do with all of this information?
She makes a diagnosis and prescribes an effective treatment.
Case interpretation involves connecting the dots to understand the relationship between a
person’s criminal behavior and her history, beliefs, and skills.

3

 4

2.2

COMPAS Scores

The COMPAS assessment system consists of predictive risk scales for risk prediction and
separate need scales for identifying program needs in the domains of employment, housing,
substance abuse, and others. Agencies commonly adhere to the risk principle to target
individuals for treatment programs who have high recidivism risk scores and high need for
treatment (e.g., high substance abuse scores).

2.3

Levels of Interpretation

Skills and issues to consider when interpreting assessment information:
1. Interpretation is a skill that needs to be honed over time.
2. People are complex and multi-faceted. Interpretation is hard, yet is necessary for
understanding behavior and for determining strategies for intervention.
3. From research in the field we have several criminological theories to help us understand
the paths to criminal behavior.
There are different levels of interpretation.
1. Level 1: “Big bars, bad – little bars, good.” Crime-producing issues are viewed largely
in isolation, thus disregarding the influence high-scoring needs have on one another.
This is a simplistic interpretation that fails to consider a chain of possible precursors
and antecedents. It is, however, a good place to start, by identifying the areas of need
for further consideration.
2. Level 2: Helps strengthen the interpretation process beyond Level 1 by identifying
criminogenic factors that are interrelated. In particular, level 2 begins the process of
looking at areas of need that influence one another. Palmer (1994) identified three
areas of commonality: environmental issues, skill deficiencies, and cognitive/mental
health/psychological areas. This level of interpretation allows practitioners to begin
developing interventions that might address clusters of needs, rather than individual
needs in isolation of others.
3. Level 3: This is a fully integrated interpretation, using criminological theories to explain patterns of criminal behavior and help practitioners begin understanding possible
underlying causes or contributors to the person’s behavior. This approach enables the
practitioner to consider a mix of explanatory theories that help “connect the dots” of
need and other influencing factors to paint a picture of the individual’s pathway to
crime.
The needs measured by the scales are often interwoven and co-occurring. Accurately interpreting a COMPAS bar chart requires the practitioner to take into account all the high
scoring needs. Criminological theories provide a framework to help understand the interrelationship between the different needs.

➞2015 Northpointe Inc., All Rights Reserved.

 5

2.4

Criminological Theories

People are complex creatures. To obtain a holistic picture of an individual, salient life
events and influences must be considered. Criminological theories explain how people become
involved in criminal behavior and may provide guidance for effective interventions. Several
important criminological theories are outlined below.
Social Learning Theory
1. This theory matches the traditional way we think about learning through modeling of
behavior.
2. The basic principle of the theory is that behavior is modeled, imitated, and if reinforced,
then likely to occur again.
Sub-Culture Theory
1. The theory was developed from the Chicago School on Gangs.
2. The theory was developed to explain delinquency and gang behavior.
3. The theory suggests that norms are transmitted through social interactions.
4. Norms in subcultures are different than those in the main culture.
5. Certain behaviors (crime, substance abuse) become the cultural norm within the subculture.
6. All individuals in society are driven toward economic success. Some subcultures aim
to achieve that success through illegitimate means.
7. Fischer (1995) defines subculture as “a large set of people who share a defining trait,
associate with one another, are members of institutions associated with their defining
trait, adhere to a distinct set of values, share a set of cultural tools and take part in a
common way of life” (p. 544).
Control/Restraint Theory
1. This theory suggests there are different types of control. These include internal control
(bonding to values, beliefs, etc.), external control (bonds to family, friends, social
networks, co-workers), and psychological control (emotional attachments, cognitions,
etc.).
2. The lower an individual’s level of social bonding (or less pro-social) and self control,
the more crime-prone they will become (less to lose).
3. Or, they may be bonded to antisocial social norms values and associations, and their
level of status depends on adherence to the restraints of that norm group.

➞2015 Northpointe Inc., All Rights Reserved.

 6

Sociopathic/Socialization Breakdown Theory
1. Within this theory lies the concept of the sociopathic offender, which has more layers
than the commonly stated “criminal personality.”
2. Sociopathic is a specific personality disorder. Personality disorders can be described
as a person’s world view. A person with a personality disorder does not usually see
themselves as needing help to remedy their behavior and typically blames consequences
on other people and events.
3. A sociopath is characterized by selfishness, ruthlessness, and the inability to feel guilt
or empathy.
4. This cluster of deviant personality traits and behaviors may not include criminal behavior.
Criminal Opportunity Theory (including Routine Activity)
1. This theory draws on the economic theory of markets to describe and predict criminal
behavior.
2. The theory suggests that if you alter the quality of opportunity for crime you will
reduce criminal behavior.
3. Both individual and environmental factors across time affect criminal acts.
4. The convergence in time and place of a motivated offender, suitable target, and absence
of guardianship are strong predictors of criminal behavior.
5. Crime is most likely to occur in the presence of a suitable target (victim) and a motivated offender, and in the absence of inhibiting factors (law enforcement, neighbor,
witnesses).
Social Strain Theory
1. This sometimes is referred to as the “means–end” theory of deviance.
2. Crime breeds in the gap between culturally induced aspirations and structurally distributed possibilities for success.
3. It is the combination of cultural emphasis and social structure which produces intense
pressure for deviation-criminal behavior.
4. This is an economic explanation for crime. Crime occurs largely in poverty-stricken
areas where opportunities to attain the “American Dream” by legitimate means is
blocked, producing frustration and a desire to pursue monetary success by any means
necessary.

➞2015 Northpointe Inc., All Rights Reserved.

 7

2.5

AIPIE

Interpretation and the related events around case management can be a complex set of
activities for professionals. One model that helps to explain the procedures of evidencebased practice is known as AIPIE. The AIPIE model is sequenced so that information triggers
decisions which trigger actions.
A = Assessment (COMPAS or other tool)
I = Interpretation of the results
P = Plan, create an action plan based on the information gathered
I = Implement the plan
E = Evaluate the results of the actions and outcomes
The AIPIE model is linear and cyclic, that is, the steps are sequential and inform ongoing
practice.
Risk and need scales have been discussed at length in this document. The other element
to consider for supervision is responsivity. An offender’s responsivity, or any person who
is considering making some kind of change, can be understood as their level of readiness
and their skill set to make the changes. Responsivity to intervention includes the person’s
motivation for change and the type of intervention offered. If the intervention does not fit
the need, then responsivity factors are lost. If there is good fit, then there is better chance
for success.

2.6

Basic Descriptive Information for the Scales

The scales are divided into two categories:
1. The Need Scales provide measures of relatively simple constructs (e.g., financial problems). These scales are not meant to be predictive but aim simply and accurately to
describe the offender.
2. The Risk Scales were developed using methods and strategies for predictive modeling. The purpose of the risk scales is prediction - the ability to discriminate between
offenders who will and will not recidivate.

➞2015 Northpointe Inc., All Rights Reserved.

 8

2.7

Conversion of Raw Scale Scores to Decile Scores

The COMPAS scale scores are transformed into decile scores. Deciles are obtained by ranking
the scale scores of a normative group in ascending order and then dividing these scores into
ten equal sized groups. Deciles range from 1 (lowest) to 10 (highest). These scores thus
proceed in roughly 10% steps from lowest to highest (1 through 10). A decile rank of 1
indicates that the scale score is in the lowest 10% of all scores in the normative group. A
decile rank of 2 places the scale score above 10% and below 20% of the scores, and so on, up
to a decile of 10, which places the scale score in the top 10% of all scores in the normative
group.
In general the decile rank has the following interpretation:
❼ 1 – 4: scale score is low relative to other offenders in norm group.
❼ 5 – 7: scale score is medium relative to other offenders in norm group.
❼ 8 – 10: scale score is high relative to other offenders in norm group.

Note however that the location of the decile cut-points vary depending on the type of COMPAS scale. Table 2.1 shows the cutting points for each type of COMPAS scale. Table 2.2
lists each COMPAS scale and its type.
Table 2.1: Cutting Points for COMPAS Scale Types.

Type
Type
Type
Type

1
2
3
4

Low (1-4)
Unlikely (1-2)
Unlikely (1-5)
Unlikely (1-4)

Medium (5-7)
Probable (3-4)
Probable (6-7)
Probable (5-7)

High (8-10)
Highly Probable (5-10)
Highly Probable (8-10)
Highly Probable (8-10)

The decile cutting points for the scale scores in the COMPAS Core composite norm group
(n=7381) are shown in Table 2.3. The column labeled D1 contains the cut-off for the first
decile, D2 the cut-off for the second decile, and so on. Thus, for the Criminal Personality
Scale (CrimPers), roughly one-tenth of the offenders scored 23 and lower, another one-tenth
scored 24 through 25, and so forth. If a score covers more than one decile, we use the
convention of assigning it to the lower decile category. For instance, 30% of the composite
sample have a score of 0 on the History of Noncompliance Scale (HistNonC), covering D1
through D3 in the table, but this score is assigned to the lower decile (D1). This characteristic
is associated with the granularity of certain COMPAS Core scales, which is discussed in the
next section.

➞2015 Northpointe Inc., All Rights Reserved.

 9

Table 2.2: COMPAS Core Scales and Types.
Scale

Scale Type

Violent Recidivism Risk
General Recidivism Risk
Pretrial Release Risk
Criminal Involvement
History of Noncompliance
History of Violence
Current Violence
Criminal Associates/Peers
Substance Abuse
Financial Problems/Poverty
Vocational/Education Problems
Criminal Thinking
Family Criminality
Social Environment Problems
Leisure and Recreation
Residential Instability
Social Adjustment Problems
Socialization Failure
Criminal Opportunity
Criminal Personality
Social Isolation

➞2015 Northpointe Inc., All Rights Reserved.

1
1
1
1
1
1
1
4
2
3
3
3
3
3
3
3
3
3
3
3
3

 10

Table 2.3: Decile Cut-Points for COMPAS Core Scales in the Composite Norm Group.
D1
CrimInv
1.0
HistNonC
0.0
HistViol
1.0
CurrViol
8.0
CassPeer
7.0
SubAbuse
10.0
Financ
5.0
VocEd
14.0
FamCrim
6.0
SocEnv
6.0
Leisure
5.0
ResInst
9.0
SocAdj
16.0
EJuvSoc
10.0
CrimOpp
15.0
Soc.Isol
11.0
CrimAttC
13.0
CrimPers
23.0
PretrialRisk
2.89
ViolRecidRisk −2.90
GenRecidRisk −1.30

D2

D3

D4

3.0
6.0
7.0
0.0
0.0
1.0
1.0
1.0
1.0
8.0
8.0
8.0
8.0
8.0
9.0
10.0
11.0
12.0
6.0
7.0
8.0
15.0
16.0
17.0
6.0
6.0
7.0
6.0
6.0
6.0
5.0
6.0
7.0
11.0
11.0
12.0
17.0
18.0
19.0
11.0
12.0
13.0
17.0
18.0
20.0
13.0
15.0
16.0
15.0
18.0
20.0
25.0
27.0
29.0
3.08
3.24
3.39
−2.50 −2.20 −2.00
−0.90 −0.70 −0.40

D5

D6

D7

D8

D9

D10

9.0
11.0
12.0
13.0
15.0 19.0
2.0
2.0
3.0
5.0
7.0 21.0
1.0
2.0
3.0
4.0
6.0 20.0
8.0
8.0
8.0
9.0
11.0 14.0
10.0
10.0
11.0
12.0
15.0 22.0
13.0
14.0
14.0
15.0
16.0 20.0
9.0
9.0
10.0
11.0
12.0 15.0
18.0
19.0
21.0
22.0
24.0 30.0
7.0
8.0
8.0
9.0
10.0 12.0
6.0
7.0
7.0
8.0
10.0 12.0
8.0
9.0
10.0
12.0
14.0 17.0
14.0
15.0
16.0
18.0
20.0 30.0
20.0
21.0
22.0
24.0
25.0 35.0
13.0
14.0
15.0
17.0
19.0 30.0
21.0
22.0
24.0
25.0
28.0 40.0
18.0
19.0
21.0
23.0
26.0 40.0
21.0
22.0
23.0
25.0
28.0 50.0
30.0
32.0
34.0
36.0
40.0 59.0
3.54
3.69
3.86
4.08
4.38 8.01
−1.70 −1.50 −1.20 −1.00 −0.60 1.90
−0.20
0.00
0.20
0.40
0.60 1.90

➞2015 Northpointe Inc., All Rights Reserved.

 11

2.8

Interpreting Decile Scores

It is important to note that decile scores can only be interpreted in a relative sense, and
are always linked to the norm group. If, for example, the norm group that is referenced for
decile scoring of the Violent Recidivism Risk Scale happens to consist of offenders at high
risk of violent recidivism, then low decile scores would not necessarily indicate low risk of
violent recidivism. Similarly, if the norm group happens to consist mainly of offenders with
low risk of violent recidivism, the decile scores for Violent Recidivism Risk would be biased
in the other direction – high scores could be associated with individuals who are actually
not high risk for violent recidivism.
It is also important to note that for some scales, it is not always possible to break the sample
into ten groups of exactly equal size. Hence, for some scales it was necessary to skip over
some decile scores.
When it was not possible to divide the sample into ten groups, an algorithm was used to
identify cutting points that divided the offenders into as many roughly equal-sized groups as
possible and that used the full range of decile values (i.e., 1-10).
The issue of clumping affects a limited number of scales. In addition to Violence History,
other COMPAS Core scales that exhibit clumping of decile ranks include Current Violence,
Family Crime, and Social Environment. Overall, the use of decile ranks has clear advantages
over the use of raw scale scores in terms of interpretability. Low scores (e.g., 1 thru 4)
directly reflect the lowest ends of the distribution, and high scores (e.g., 8 thru 10) reflect
the highest ends of the distribution.

2.9

Norm Groups

The COMPAS Core normative data were sampled from over 30,000 COMPAS Core assessments conducted between January 2004 and November 2005 at prison, parole, jail and
probation sites across the United States. The Core Norm Group was compiled to obtain
proportions of prison, parole, jail, and probation assessment data that reflect proportions
of adult correctional populations in the criminal justice system. Based on recent criminal
justice statistics, 21.6% of persons under adult correctional supervision during 2011 were
in prison, 12.2% were on parole, 10.5% were in jail, and 56.9% were on probation (Bureau
of Justice Statistics, 2012). The Composite Norm Group consists of assessments from state
prisons and parole agencies (33.8%); jails (13.6%); and probation agencies (52.6%). The Core
Norm includes 7,381 offenders. Men represent 76.9% of the Core Norm Group (n=5,681),
and women represent 23.1% of the Core Norm Group (n=1,700). The median age at assessment is 31.0 (M = 32.6) in the Core Norm Group. The racial composition of the Core Norm
Group is 61.6% Caucasian, 24.9% Black, 10.3% Latino and 3.2% other racial groups.
In the current version of COMPAS Core, scale scores can be referenced to the scale distributions of eight normative subgroups: (1) male prison/parole, (2) male jail, (3) male probation,
(4) male composite, (5) female prison/parole, (6) female jail, (7) female probation and (8)
female composite.

➞2015 Northpointe Inc., All Rights Reserved.

 Chapter 3
COMPAS Validity and Reliability
In this section we summarize research findings from multiple studies that demonstrate COMPAS Core is reliable (test-retest and internal consistency), that its scales measuring needs
have construct validity and behave consistently and that its risk scales have predictive validity.1 An overall conclusion is that COMPAS Core was found to be reliable and has good
predictive and construct validity.
Northpointe has an established history of working in partnership with our clients to advance
knowledge and practice. From our early work in jail classification to our recent partnerships
with the California Department of Corrections and Rehabilitation (CDCR) and the University of Cincinnati, Northpointe leverages the opportunity of public and private partnership
to test and advance knowledge. Our research and evaluation findings are publicly shared
through conference papers, technical reports, peer-reviewed articles and book chapters to
advance the availability of current information for use in practice.

3.1

Predictive Validity of the COMPAS Risk Scales

COMPAS distinguishes between risk scales (designed to predict recidivism) and needs scales
(designed to measure needs, inform case plans and identify intervention targets). This approach of separating risk and needs aligns with current best practices in risk assessment
(C. Baird, 2009; S. D. Gottfredson & Moriarty, 2006). COMPAS has two main risk models:
General Recidivism Risk and Violent Recidivism Risk. Some researchers believe risk scales
should be dynamic (composed of dynamic, criminogenic needs) so that one can measure
change in risk of recidivism over time. Others have argued that risk models should be composed of static criminal history factors available in criminal justice information management
systems, arguing that these models are more objective, reliable, and efficient (Barnoski &
Drake, 2007). Our risk scales make limited use of dynamic variables. Our methods for
developing and validating the General Recidivism Risk Scale were strongly influenced by
1

The General Recidivism Risk and Violent Recidivism Risk scales are used in both COMPAS Core and
COMPAS Reentry. Identical linear equations are used to calculate the risk scales in the two applications.

12

 13

the research of John Copas and colleagues who have developed an outcomes-based recidivism scale for England and Wales (Copas & Marshall, 1998). The methods used to develop
both risk scales are described in various books on regression modeling and machine learning
(see,e.g., Harrell, 2001; Hastie, Tibshirani, & Friedman, 2008; Kuhn & Johnson, 2013).
Northpointe is committed to testing, evaluating, and improving our risk models. The General
Recidivism Risk and Violent Recidivism Risk scales have been validated with prospective
outcomes in new samples in several different studies since they were first developed.
When possible we include an outcomes component in the pilot test of COMPAS in new
jurisdictions. This component is designed to evaluate the predictive validity of the risk
scales. In 2006 we conducted pilot tests in the New York Office of Probation and Correctional
Alternatives (OPCA), the New York State Division of Parole (NYSDP), and the Michigan
Department of Corrections (MDOC). These three pilot studies all had outcomes studies
built into them. In 2008 we conducted outcomes studies at all three sites using their pilot
data. We also conducted separate studies in the California Department of Corrections and
Rehabilitation (CDCR) and the OPCA. This latter study was published in the Journal of
Criminal Justice and Behavior (Brennan, Dieterich, & Ehret, 2009).
Table 3.1 below shows the results of subsequent tests of the predictive validity of the COMPAS risk scales. These outcomes studies were conducted on large samples in the Michigan
Department of Corrections (Brennan & Dieterich, 2008; Dieterich, Oliver, & Brennan, 2011;
Dieterich, Brennan, & Oliver, 2011); the New York State Office of Probation and Correctional Alternatives (Brennan & Dieterich, 2009; Brennan et al., 2009; Lansing, 2012); and
the California Department of Corrections and Rehabilitation (Farabee, Zhang, Roberts, &
Yang, 2010).
The table shows the area under the receiver operating characteristic curve (AUC) for the
General Recidivism Risk and Violent Recidivism Risk scales. The AUC is the most widely
used measure of predictive accuracy in criminal justice, psychology, medicine, and related
fields. AUCs of 0.65 to 0.69 indicate modest to moderate predictive accuracy. AUC’s of 0.70
to 0.75 indicate moderate to good predictive accuracy. Note that for arrest, felony arrest,
noncompliance and return to prison outcomes, the General Recidivism Risk Scale is tested.
For person offense arrests the Violent Recidivism Risk Scale is tested.
The results of these studies indicate that the COMPAS risk scales generally fall into the
moderate to good range of predictive accuracy. They also indicate that COMPAS generally
meets or exceeds the AUC values produced by competitive instruments such as the LSI-R.

➞2015 Northpointe Inc., All Rights Reserved.

 14
Table 3.1: Summary of AUC results for the General Recidivism Risk and Violent Recidivism
Risk Scales in several outcomes studies.

Study
NY Probationa
NY Probationb
MDOC Reentryc
MDOC Probationd
CDCR Reentrye

N
(n=2,328)
(n=13,993)
(n=25,347)
(n=21,101)
(n=25,009)

Any
Year Arrest Felony Person NonComp. Return
2009 0.680
2012 0.710
2011
2011
2010 0.700

0.700

0.710

0.710
0.670

0.700
0.740
0.650

0.690
0.710

0.720

a

(Brennan et al., 2009).
(Lansing, 2012).
c
(Dieterich, Brennan, & Oliver, 2011).
d
(Dieterich, Oliver, & Brennan, 2011).
e
(Farabee et al., 2010).
b

Differential Validity
A few independent outcomes studies have examined the predictive validity of the COMPAS
risk scales for gender and racial groups. Brennan et al. (2009) found that the COMPAS
recidivism models preformed equally well for African American and White men at predicting
the arrest outcomes in a probation sample. A prior study examined the predictive accuracy
of the COMPAS for different ethnic groups, and that study reported much weaker results
for African American men (Fass, Heilbrun, DeMatteo, & Fretz, 2008). In predicting rearrest
within 1 year of release, Fass et al. (2008) reported AUCs for the COMPAS Recidivism Risk
Scale of .81 for Whites, 0.67 for Hispanics, 0.48 for African Americans, and 0.53 for the total
sample assessed with COMPAS (N = 276). However, their study has at least one critical
weakness that renders its findings unreliable. Their small overall sample size and base rates
resulted in extremely small effective sample sizes for the ethnic groups (African American =
36, Hispanic = 4, White = 1). These effective sample sizes are too small for ROC analysis
and unreliable results were obtained.
Farabee et al. (2010) report separate findings for men and women in a California Department
of Corrections and Rehabilitation sample of released prisoners onto parole with 2 year followup. They present a matrix with bivariate correlation coefficients for the General Recidivism
Risk Scale and any arrest separately for men and women. The Pearson product moment
correlation between the General Recidivism Risk Score and any arrest is 0.32 for men and
0.32 for women, thus providing evidence that the risk scale has similar predictive validity for
men and women.
Table 3.2 displays AUCS for the any arrest outcome for the data set used by Farabee et al.
(2010). The AUCs in the table give an indication of how well the General Recidivism Risk
Scale discriminates the offenders who are rearrested from those who were not rearrested.
The results are for the entire sample (All) and for Men, Women, White, Black, and Hispanic
groups. The values for the AUCs are very nearly the same.

➞2015 Northpointe Inc., All Rights Reserved.

 15
Table 3.2: AUCS for the General Recidivism Risk Scale for a California prison sample. The
AUCs are calculated separately for the different subgroups defined by gender and ethnicity/race. The lower (Low) and upper bounds (High) of the 95 percent confidence interval
are displayed along with the number of failures (Nfail) and the number of offenders in the
sample (N).
Men
Women
White
Black
Hispanic
All

AUC Low High Nfail
N
0.71 0.70 0.71 14819 21015
0.69 0.67 0.71 1595 2638
0.70 0.69 0.71 4683 7268
0.69 0.67 0.70 4813 6447
0.71 0.70 0.72 5980 8514
0.70 0.70 0.71 16414 23653

Table 3.3 displays AUCS for a large reentry sample from the Michigan Department of Corrections. The outcome was any arrest within 3 years following release from prison into the
community. Offenders who did not have opportunity to fail in a three year period were
excluded from the sample. As in the previous analysis, the results are for the entire sample
(All) and for the Men, Women, White, Black, and Hispanic groups.
The AUCs in Table 3.3 vary from 0.71 (Black) to 0.78 (Hispanic). The effective sample size
for the Hispanic group is relatively small, which results in a broad 95% confidence interval.
The AUCs for Men (0.73) and Women (0.74) are nearly the same. The AUCs for White
(0.75) and Black (0.71) do noticeably differ but both values are reasonably high.
These results taken together are encouraging. They suggest that the predictive validity of
the General Recidivism Risk Scale is good overall and nearly equivalent for the men and
women, and for the White, Black and Hispanic offenders.
Table 3.3: AUCS for the General Recidivism Risk Scale and the any arrest outcome for a
Michigan reentry sample. The AUCs are calculated separately for the different subgroups
defined by gender and ethnicity/race. The lower (Low) and upper bounds (High) of the 95
percent confidence interval are displayed along with the number of failures (Nfail) and the
number of offenders in the sample (N).
Men
Women
White
Black
Hispanic
All

AUC Low High Nfail
N
0.73 0.72 0.74 5427 13439
0.74 0.71 0.77
341
961
0.75 0.74 0.76 2807 7177
0.71 0.69 0.72 2720 6571
0.78 0.73 0.84
89
289
0.73 0.72 0.74 5768 14400

➞2015 Northpointe Inc., All Rights Reserved.

 16

Examples of Validity Results for Different Tools and Outcomes
Here we provide examples of the AUCs obtained with other risk tools to help contextualize
the findings of our studies. Perhaps the best known instruments are the Violence Risk
Appraisal Guide [VRAG] (Quinsey et al., 1998); the Level of Services Inventory-Revised
[LSI-R] (Andrews, Bonta, & Wormith, 2006); and the Psychopathy Checklist-Revised [PCLR] (Hare, 1991). The AUC values for these instruments in recent studies are quite varied
depending on the populations, outcome periods, and dependent variables used in specific
studies.
VRAG: Quinsey et al. (1998) found an AUC of 0.76 in a large scale, multiyear recidivism
study. Barbaree, Seto, Langton, and Peacock (2001) reported AUCs of 0.69 in predicting
serious reoffending and 0.77 when predicting any re-offense for sex offenders. Kroner, Stadtland, Eidt, and Nedopil (2007) obtained an AUC of 0.70 in a study of re-offending among
mentally ill offenders.
LSI-R: A review by Andrews et al. (2006) did not provide AUCs. However, Barnoski
and Aos (2003) found AUCs of 0.64 - 0.66 for the LSI-R in predicting felony and violent
recidivism among Washington State prisoners. Flores, Lowenkamp, Smith, and Latessa
(2006) reported an AUC of 0.689 using the LSI-R to predict re-incarceration among federal
probationers. Dahle (2006) reported an AUC of 0.65 using the LSI-R to predict violent
recidivism. Barnoski and Drake (2007) reported an AUC of 0.65 using the LSI-R to predict
felony sex recidivism.
PCL-R: Predictive accuracy varied across studies. For example, a Swedish study of mentally
ill violent offenders (Grann, Belfrage, & Tengstrom, 2000) found AUC levels of 0.64 - 0.75
based on various follow-up time frames. Barbaree et al. (2001) reported AUCs of 0.61, 0.65,
and 0.71 for the PCL-R in predicting various recidivism outcomes among sex offenders.

3.2
3.2.1

Validity of COMPAS Core Needs Scales
Criterion Validity

In contrast to the COMPAS risk scales, the COMPAS Needs Scales have a separate purpose
and were developed using different methods. The risk scales were developed using methods
and strategies for predictive modeling. The purpose of the risk scales is prediction - the
ability to discriminate between offenders who will and will not recidivate.
The need scales are not meant to be predictive but aim simply and accurately to describe the
offender along dimensions relevant for correctional practice. Research findings indicate that
individuals involved in the criminal justice system often have problems and deficits in the
domains of education, housing, employment, substance abuse, relationships, and cognition.
The need scales should be valid and reliable measures of constructs in these domains and
other aspects of the person-in-environment that represent potential targets for interventions.
The need scales guide individualized decisions for case planning, including identifying targets
and choosing interventions. Within some theoretical frameworks, needs are expected to be

➞2015 Northpointe Inc., All Rights Reserved.

 17

criminogenic, suggesting that they cause recidivism and that recidivism can be reduced if
the criminogenic need is effectively addressed. But research results indicate many constructs
in these domains are only modestly correlated with recidivism, and evidence of a causal link
between needs, treatment, and recidivism is lacking (e.g., Monahan & Skeem, 2014). Here we
focus only on correlations to demonstrate that the COMPAS Core need scales are relevant
and useful measures for correctional practice.
The following tables show measures of association between the COMPAS Core scales and
recidivism in large samples from two COMPAS outcomes studies. The results obtained in
the respective COMPAS outcomes studies provide evidence of the criterion validity of the
COMPAS Core scales. The results demonstrate that in general the COMPAS Core need
scales measure factors associated with recidivism, and hence, they are useful measures of
potential intervention targets. The results can be compared with the results from published
studies. For example Barnoski and Aos (2003) conducted an outcomes study in a sample of
22,533 offenders and provide a table with similar measures of association between the LSI
subscales and recidivism.
Table 3.4 shows measures of association between the COMPAS Core scales and any arrest
within 2 years in the study sample used by Farabee et al. (2010). The sample consists
of 23,635 soon-to-be-released inmates assessed with COMPAS Core who were followed for
two years after release from prison. The first column shows the correlation between each
COMPAS Core scale and recidivism. For correlations between a continuous variable (e.g.
Voced, Subabuse, etc.) and a dichotomous variable (recidivism), we estimate the point
biserial correlation (rpb ).2 The point biserial correlation is mathematically equivalent to the
Pearson product moment correlation (r). J. Cohen and Cohen (1983) provide the following
conventions for interpreting r: 10 = small; 30 = medium; 50 = large. The next column
shows the area under the receiver operating characteristic curve (AUC). The AUC is a rank
measure indicating how well the respective scales discriminate recidivists from nonrecidivists.
An AUC equal to 1 indicates that the scale discriminates perfectly. An AUC equal to 0.50
indicates that the scale does not discriminate any better than chance. By convention an AUC
of 0.70 is regarded as good in criminal justice settings. The AUC is 0.60 for the Criminal
Associates Scale - a modest result if this were a standalone risk scale, but for a needs scale,
the result indicates good criterion validity. The last column shows the odds ratio. The
odds ratio indicates how much the odds of recidivating change for every one-unit increase in
the respective COMPAS Core scale. The odds ratio for Criminal Associates is 1.09, which
indicates that for every one-unit increase in the Criminal Associates raw score the odds of
recidivism increases by 9%. There is solid evidence of criterion validity in this study sample
for most of the COMPAS Core scales.

2

In a previous version of the Practitioner’s Guide, the biserial correlation was reported. The biserial
coefficients are inferred estimates of what the Pearson correlation would be if both variables were continuous
and normally distributed. We now use the Pearson product moment correlation (r), which is usually called
the point biserial correlation (rpb ) when one of the variables is dichotomous.

➞2015 Northpointe Inc., All Rights Reserved.

 18

Table 3.4: Measures of Association Between COMPAS Core Scales and Any Arrest Within
Two Years in Farabee et al. Study Data.

COMPAS Scale
General Recidivism Risk
Criminal Involvement
Noncompliance History
Violence History
Current Violence
Criminal Associates
Substance Abuse
Financial Problems
Voced Problems
Family Crime
Social Environment
Leisure
Residential Instability
Social Adjustment
Socialization Failure
Criminal Opportunity
Social Isolation
Criminal Thinking
Criminal Personality
Cognitive Behavioral

Point-Biserial
Correlation AUC
0.34
0.20
0.16
0.11
−0.05
0.14
0.02
0.09
0.17
0.07
0.10
0.11
0.08
0.15
0.18
0.19
0.04
0.11
0.13
0.23

0.70
0.61
0.61
0.58
0.52
0.60
0.51
0.55
0.61
0.54
0.56
0.57
0.55
0.60
0.62
0.62
0.52
0.57
0.58
0.63

Odds
Ratio
3.31
1.10
1.11
1.06
0.92
1.09
1.02
1.08
1.11
1.11
1.12
1.07
1.04
1.10
1.13
1.10
1.02
1.04
1.05
1.05

With n=23,635, a correlation of .013 is significant at p < .05
(2-tailed).

➞2015 Northpointe Inc., All Rights Reserved.

 19
Table 3.5: Measures of Association Between the COMPAS Core Scales and Any Arrest
Within 1 Year in the 2010 New York Probation Study Data.

COMPAS Scale
General Recidivism Risk
Criminal Involvement
Noncompliance History
Violence History
Current Violence
Criminal Associates
Substance Abuse
Financial Problems
Voced Problems
Family Crime
Social Environment
Leisure
Residential Instability
Social Adjustment
Socialization Failure
Criminal Opportunity
Social Isolation
Criminal Thinking
Criminal Personality

Point-Biserial
Correlation AUC
0.27
0.08
0.13
0.08
0.04
0.13
−0.07
0.05
0.17
0.12
0.09
0.14
0.08
0.17
0.18
0.22
0.06
0.09
0.15

Odds
Ratio

0.71
0.56
0.59
0.55
0.53
0.60
0.55
0.53
0.63
0.57
0.57
0.59
0.56
0.63
0.64
0.66
0.55
0.57
0.62

2.94
1.05
1.15
1.09
1.12
1.15
0.93
1.06
1.12
1.22
1.18
1.11
1.06
1.12
1.15
1.14
1.03
1.04
1.06

With n=2,328, a correlation of .041 is significant at p < .05
(2-tailed).

Table 3.5 shows the point biserial correlations between the COMPAS Core scales and any
arrest within 1 year in the study sample from Brennan et al. (2009). The sample consists
of 2,328 probation intakes assessed with COMPAS Core. The results in Table 3.5 can be
compared to the results in Table 3 in Brennan et al. (2009). The sample and event of interest
(any arrest) are identical, but here we fit a logistic regression model with a binary outcome
(any arrest within one year), and in Brennan et al. (2009) we fit a Cox proportional hazards
model in which the outcome is defined as failure over the entire follow-up which ranged out
to 1,722 days.

3.2.2

Construct Validity

Construct validity refers to the extent to which a scale measures what it is supposed to
measure. Construct validity is tested by observing correlations between measures of the
same or divergent constructs. Construct validity is relevant only for the COMPAS need scales
and refers in part to unidimensionality of the scale and to its factor structure. Construct
validity additionally is based on establishing evidence that a scale correlates in an expected
manner with similar scales, and to other relevant variables in theoretically expected ways.

➞2015 Northpointe Inc., All Rights Reserved.

 20

To demonstrate the construct validity of a measure requires the testing of different types of
validity including convergent and divergent validity. Here we only address the convergent
validity of the COMPAS Core need scales. A direct approach to convergent validity is to
measure the correlation between matched scales of the LSI and COMPAS Core. The LSI-R
is considered a gold standard because it is the current industry leader. This would be a good
indication for how well the COMPAS Core scales are measuring the same concept. Results
from a study conducted in the California Department of Corrections and Rehabilitation
(Farabee et al., 2010) show a direct and high level of correlation between matching LSI-R
and COMPAS Core scales. The findings shown in Table 3.6 offer strong evidence of the
convergent validity of the COMPAS Core scales. Farabee et al. (2010) found high Pearson
product moment correlations between the LSI-R and COMPAS Core measures of Criminal
Involvement (0.64); Vocation/Education (0.51); Criminal Associates (0.48); Substance Abuse
(0.53); Financial (0.49); and Residential Stability (0.57).
Table 3.6: Correlations between COMPAS Core and LSI-R scales in Farabee et al., 2010
COMPAS
Criminal Involvement
Criminal Associates/Peers
Substance Abuse
Financial
Vocation/Education
Family Criminality
Leisure
Residential Instability
Criminal Attitudes

LSI-R
Criminal History
Companions
Alcohol/Drug Problem
Financial
Education/Employment
Family/Marital
Leisure/Recreation
Accommodation
Attitudes/Orientation

Correlation
0.64 (p < .0001)
0.48 (p < .0001)
0.53 (p < .0001)
0.49 (p < .0001)
0.51 (p < .0001)
0.16 (p > .10)
0.05 (p > .10)
0.57 (p < .0001)
0.20 (p = .08)

Shifting to more general issues of convergent validity, we consider additional evidence to support the convergent validity of the COMPAS Core need scales. For example, the COMPAS
Core substance abuse measure correlates positively (r = 0.44) with the Substance Abuse
Subtle Screening Inventory (SASSI) in the Michigan Department of Corrections pilot data.
Convergent validity is also demonstrated if a measure correlates in the predicted manner with
other variables with which it theoretically should correlate. For example, research in developmental delinquency (longitudinal research in which anti-social behaviors and attitudes are
studied over the life course) consistently finds that youth with early onset of delinquent behavior tend to have more serious delinquency trajectories and more negative emotionality,
lower achievement, and problems in social adjustment (Moffitt, 1993). Thus, when we consistently find, over multiple studies, that our Criminal Personality, Criminal Attitudes, Social
Adjustment and Vocational Educational scales correlate with age-at-first-arrest, just as developmental delinquency research predicts, we take this as evidence of convergent validity.
Note that these correlations with age-at-first-arrest hold up when current age is statistically
controlled.
Furthermore, age-at-first-arrest is a good external variable to demonstrate convergent validity
of the COMPAS Core need scales. Although age-at-first-arrest is collected inside COMPAS,

➞2015 Northpointe Inc., All Rights Reserved.

 21
Table 3.7: Concurrent correlations between COMPAS Core Scales and criminal history indicators in the Wisconsin Division of Adult Community Corrections sample.

COMPAS Scale
Criminal Associates
Substance Abuse
Financial Problems
Voced Problems
Family Crime
Social Environment
Leisure
Residential Instability
Social Adjustment
Social Isolation
Criminal Thinking
Criminal Personality

Age-atFirst

Prior
Arrests

Parole
Returns

Prior
Prisons

Assault
Tickets

−0.28
−0.07
−0.07
−0.24
−0.22
−0.16
−0.13
−0.09
−0.27
−0.02
−0.13
−0.24

0.23
0.22
0.08
0.16
0.10
0.10
0.04
0.10
0.22
0.08
0.06
0.15

0.17
0.14
0.06
0.13
0.04
0.11
0.02
0.11
0.13
0.07
0.03
0.11

0.17
0.15
0.05
0.11
0.02
0.13
0.00
0.09
0.11
0.06
0.02
0.09

0.19
0.06
0.04
0.15
0.07
0.13
0.05
0.11
0.15
0.07
0.10
0.16

With n=25,773, a correlation of .013 is significant at p < .05 (2-tailed).

it comes from official records, while the need scales are scored using a different method
(interview and self-report).
We have evidence of convergent validity of this type from psychometric studies in the Michigan Department of Corrections (MDOC), New York Office of Probation and Correctional
Alternatives (OPCA), New York State Division of Parole, Virginia Department of Corrections, South Carolina Department of Probation, Parole, and Pardon Services, and many
other sites. To illustrate our approach to demonstrating convergent validity, we present results in Table 3.7 from a sample in the Wisconsin Department of Corrections Division of
Adult Community Corrections (DCC). The DCC sample consists of 25,773 Core COMPAS
assessments conducted between July 1, 2012 and August 31, 2013. Men comprise 76.7% of
the sample.
There are many notable correlation patterns in Table 3.7 that provide evidence of convergent
validity for the COMPAS Core scales. For example, we see that age-at-first-arrest correlates
negatively with the higher-order personality scales Criminal Attitudes (r = −.13) and Criminal Personality (r = −.24). This comports with findings in developmental research that
indicate offenders with early onset are more likely to have high scores on similar types of
personality measures and more serious and persistent criminal involvement (Moffitt, 1993).
Similarly, we see that offenders with earlier age-at-first-arrest are more likely to have higher
scores on scales measuring factors that have been identified as criminogenic in longitudinal
developmental studies. These scales include Criminal Associates (r = −.28), Family Crime
(r = -.22), Vocational/Educational Problems (r = −.24), and Social Environment (r = −.16)
(Farrington, Jolliffe, Loeber, Stouthamer-Loeber, & Kalb, 2001).
Another pattern in Table 3.7 is defined by the correlations between previous arrests and the
scales Social Adjustment (r = 0.22), Criminal Personality (r = 0.15), Criminal Associates

➞2015 Northpointe Inc., All Rights Reserved.

 22

(r = 0.23) and Substance Abuse (r = 0.22) (Stouthamer-Loeber, Loeber, Wei, Farrington,
& Wikstrom, 2002).
There are modest, significant correlations between the assaultive misconduct item from
COMPAS Core and the scales Criminal Associates and Peers (r = .19), Vocational Educational Problems (r = 0.15), Social Environment (r = 0.13), Social Adjustment (r = 0.15),
and Criminal Personality (r = 0.16). In their meta-analysis, Gendreau, Goggin, and Law
(1997) found that antisocial attitudes and criminal peers were important individual level
predictors of prison misconduct.
There are small, significant correlations between the number of returns to custody for a
parole violation and the scales Criminal Associates and Peers (r = 0.17), Substance Abuse
(r = 0.14), Vocational Educational Problems (r = 0.13), Residential Instability (r = 0.11),
and Social Adjustment (r = 0.13). Substance abuse, residential stability, and employment
and education have been identified in past research as important factors associated with
reentry success (Nelson, Deess, & Allen, 1999; Petersilia, 2003; Solomon, Visher, La Vigne,
& Osborne, 2006; Travis, 2005). At least one study using self-report and qualitative methods
found that housing and employment problems did not distinguish between parole violators
and successes (Bucklen & Zajac, 2009).
Overall, the observed relationships between the COMPAS Core scales and criminal history
indicators in the Wisconsin DCC sample provide evidence of the convergent validity of the
scales. These correlations comport with relationships between risk factors and serious and
violent trajectories observed in developmental criminological research (Herrenkohl et al.,
2000; Tolan & Gorman-Smith, 1998).The significant correlations we have pointed out are
somewhat attenuated by variability in the base rates of the paired variables. These modest
associations are typical of correlations between need scales and criminal involvement variables
observed in many criminal justice research contexts.

3.2.3

Content Validity

Content validity refers to the coverage of key factors that are relevant in the criminogenic
domain. COMPAS Core was designed to have greater coverage of relevant scales than the
LSI-R. Content validity has a major role in any assessment field. It refers to the extent
to which an assessment comprehensively includes and assesses the key factors in a domain
of interest. The LSI-R includes 10 important criminogenic factors that assess constructs
supported in the literature.
A study conducted by Farabee et al. (2010) found that 9 out of these 10 LSI-R scales are
clearly matched to a similar scale in COMPAS Core. Thus, in terms criminogenic scale
coverage (content validity), COMPAS Core matches virtually all scales contained in the LSIR. However, the COMPAS Core system additionally includes another 14 scales that can be
utilized or turned on/off by an agency depending on its information needs. These additional
scales are supported empirically and cover constructs such as anger/hostility, history of noncompliance, low social supports, and socialization failure.

➞2015 Northpointe Inc., All Rights Reserved.

 23

3.3

Internal Consistency Reliability

For a scale to be useful it must be reliable. For example, if one were to carry out repeated
testing of a given respondent with different questions or tests, approximately the same scale
value should be obtained on each re-test. Generally, if the items entering a scale are highly
correlated (internally consistent), then the summated scale will be reliable. Internal consistency reliability - typically assessed by Cronbach’s Alpha Coefficient - is a widely used and
popular reliability approach. It is often used as a counterpart to test-retest reliability. By
convention alphas of 0.70 and above indicate acceptable internal consistency for most applications in the behavioral sciences. Low alphas indicate the scale has too few items or the
items don’t have much in common and possibly measure more than one construct (Nunnally
& Bernstein, 1994).
Table 3.8 shows the summary statistics and alpha coefficients for the COMPAS Core scales in
a sample of prison intake assessments in the Michigan Department of Corrections. We have
consistently found similar results in prison and probation study samples across numerous
jurisdictions.
The low alphas on Violence History (0.53) and Current Violence (0.52) reflect the fact that
these are indexes composed of different types of offenses that do not necessarily correlate
with each other. A low alpha does not indicate a problem because the items are not expected
to be highly correlated as they are in a scale. Family Crime (0.62) is a similar type of index
of problems experienced by family members.
Social Adjustment (0.54) and Criminal Opportunity (0.66) are higher order scales. They
are not unidimensional. Low internal consistency is less of a concern for these scales. They
are composed of two or three underlying constructs each. Cronbach’s alpha is less useful for
higher order scales, since the multidimensionality of the higher order scales makes it difficult
to ascertain what low alpha coefficients indicate. Conversely high alphas do not necessarily
indicate unidimensionality.

➞2015 Northpointe Inc., All Rights Reserved.

 24

Table 3.8: Summary statistics and alpha coeffients for the COMPAS Core scales in a prison
intake sample from the Michigan Department of Corrections .
Items
Criminal Involvement
Noncompliance History
Violence History
Current Violence
Criminal Associates
Substance Abuse
Financial Problems
VocEd Problems
Family Crime
Social Environment
Leisure
Residential Instability
Social Adjustment
Socialization Failure
Criminal Opportunity
Social Isolation
Criminal Thinking
Criminal Personality

4
5
9
7
7
10
5
11
6
6
5
10
15
13
14
8
10
13

N

Min

Max

Mean

SD

Alpha

15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315
15, 315

0.00
0.00
0.00
7.00
7.00
10.00
5.00
11.00
6.00
6.00
5.00
9.00
12.00
7.00
13.00
8.00
10.00
13.00

19.00
21.00
16.00
13.00
22.00
20.00
15.00
30.00
12.00
12.00
17.00
30.00
37.00
32.00
39.00
40.00
45.00
58.00

8.82
4.49
2.13
8.21
9.75
12.81
8.21
19.60
7.57
7.54
7.86
13.26
20.23
12.10
21.23
16.90
20.73
31.84

4.66
4.23
2.37
1.28
2.66
2.40
2.34
3.89
1.50
1.82
3.52
3.70
3.44
3.76
4.45
4.85
4.91
5.71

0.75
0.65
0.52
0.53
0.71
0.76
0.70
0.71
0.62
0.81
0.86
0.71
0.54
0.69
0.66
0.83
0.80
0.70

➞2015 Northpointe Inc., All Rights Reserved.

 25

3.4

Test-Retest Reliability

In a recent independent study by Farabee et al. (2010) the COMPAS Core scales showed very
high test-retest reliability, with correlations ranging from 0.70 to 1.00, and with an average
correlation above 0.80. Thus, the various COMPAS Core sub-scales demonstrated good to
excellent reliability over time. An important aspect of the Farabee study was a comparison
against the well-known LSI. Overall, the average test-retest correlation coefficient for the
COMPAS Core scales was 0.88; for LSI-R, the mean as measured in the same study was
0.64.

➞2015 Northpointe Inc., All Rights Reserved.

 Chapter 4
Treatment Implications for Scales
Each COMPAS scale has been constructed based on a variety of behavioral and psychological
constructs that are of very high relevance to recidivism and criminal careers. Included in
this section is a brief description of the area of research/literature that supports the scale
content and context. This material supplements the document “Measurement and Treatment
Implications of the COMPAS Core Scales.”
Interpretation of the scale scores and how they relate to case planning and intervention is
a key concept for COMPAS users. The information contained in this section is intended to
assist you in your interpretation of the COMPAS scores as you plan for meaningful interventions and plot the course of behavioral change with the individual. Some brief examples
of language for case planning are also offered with each needs scale description as a means
to generate thoughtful, individualized goals and tasks for a person under supervision. The
language (not considered a full treatment plan or goal/task statement) in the case planning
examples is action oriented in the goals and tasks. The concept of ”how” is defined through
behavioral statements. For example, how will the person find emergency housing, or how
will the person find new, healthy friends.

4.1

Risk Scales

In this section we describe the Risk Scales in COMPAS. We have developed risks scales
for general recidivism, violent recidivism, and pretrial misconduct. There are additional
risk scales under development. Northpointe’s Research Department also conducts outcomes
studies with clients and develops and validates customized risk assessment tools.

26

 27

4.1.1

Pretrial Release Risk

The Pretrial Release Risk Scale was developed through a pretrial release outcomes study
conducted in a large sample of felony defendants assessed with COMPAS in Kent County,
Michigan Pretrial Services (Dieterich, 2010). The Pretrial Release Risk Scale was constructed
to predict failure to appear (FTA) and new felony arrest among defendants on pretrial release.
Prior pretrial risk assessment research has consistently identified a set of factors that are
predictive of pretrial failure. The most common risk factors include current charges, pending
charges, prior arrest history, previous pretrial failure, residential stability, employment status,
community ties, and substance abuse (VanNostrand, 2003). We selected items from the
COMPAS assessment and included them as candidates for risk model development on the
basis of this prior research.
One purpose of pretrial release risk assessment is to sort a pretrial caseload into low-,
moderate-, and high-risk groups based on the likelihood of failure to appear in court or
commit a new crime pending trial. Use of the risk assessment tool by pretrial services agencies should result in consistent and equitable decisions regarding release and conditions of
release. The use of objective risk assessment tools is recommended by the National Association of Pretrial Services Agencies (2004). The risk assessment tool should be empirically
derived and have demonstrated predictive validity in the jurisdiction in which it is deployed.
The factors that enter into the risk assessment score should be consistent with applicable
state statutes.1 These and other guiding principles for pretrial risk assessment are outlined
in Pretrial Services Legal and Evidence-based Practices (VanNostrand, 2007).

4.1.2

General Recidivism

The recidivism risk scale was developed to predict new offenses subsequent to the COMPAS
assessment date. The outcome used for the original scale construction was a new misdemeanor or felony offense within two years of the COMPAS administration date.
The primary factors making up this scale involve prior criminal history, criminal associates,
drug involvement, and early indicators of juvenile delinquency problems. All of these risk
factors are well known predictors of recidivism.
Scores in the medium and high range garner more interest from supervision agencies than low
scores, as a low score would suggest there is little risk of general recidivism. It is important
to note that the risk scores are generally taken from static information and that current level
of needs, e.g. substance abuse or other issues can very much influence a person’s likelihood
of acting out or recidivating. In a later discussion the concept of Low risk/High needs will
be covered.
General recidivism refers to a broad range of potential acts, therefore, versatility is an element
for consideration. The COMPAS Typologies document delineates the typologies that have
been discovered through research at Northpointe. One trait that lends itself to recidivism is
versatility.
1

For example in New York a pretrial risk assessment instrument cannot be based on age, gender, or
marital status (Division of Probation and Correctional Alternatives, 2007).

➞2015 Northpointe Inc., All Rights Reserved.

 28

4.1.3

Violent Recidivism

This scale was originally developed in COMPAS Core assessment data on a large sample of
probation and presentence investigation (PSI) cases. The scale was subsequently added to
COMPAS Reentry. The scale inputs include history of violence, history of non-compliance,
vocational/educational problems, the person’s age-at-intake and the person’s age-at-firstarrest. The strong association of these factors with future violence has been established in
previous research and holds true for people who are considered “non-disordered” (Gendreau,
Goggin, & Little, 1996). Additionally, meta-analytic results from studies with disordered
persons show that a history of violent crime is one of the more potent predictors of violent
recidivism (Bonta, Law, & Hanson, 1998).
Similar to the General Recidivism Risk Score, attention to the medium and high scores on
this scale warrants careful planning for officer, institutional, and community safety. Some
offenders, based on their past history of violent acts may score in the high and medium range,
yet, show low or medium needs areas. Consideration for the current status of the offender
and the support network in place is, as always, recommended, yet in the case of a person
who scores high on this scale, special supervision conditions may be deemed necessary.

4.1.4

Recidivism Risk Screen

The Recidivism Risk Screen (RRS) is a brief recidivism risk scale developed to predict a new
misdemeanor or felony offense arrest within two years. The RRS consists of five salient risk
factors (age, age at first arrest, number of prior arrests, employment status, and prior parole
revocations). The RRS is particularly useful to agencies that apply a triage strategy as part
of their risk and needs assessment protocol to improve efficiency and reduce workload. The
RSS is suitable as a prescreen in correctional facilities to select high risk cases for further
assessment using a more comprehensive scale set from the Northpointe Suite. The RSS can
also be used in community corrections settings to screen candidates for administrative supervision or lower supervision levels. The RSS is not intended as a substitute for the standard
risk scales in the Northpointe Suite. The General Recidivism Risk and Violent Recidivism
Risk scales measure aspects of risk (both general and violent recidivism) not covered by the
RRS. Used in combination with the Current Violence Scale, the General Recidivism Risk
and Violent Recidivism Risk scales provide a complete recidivism risk profile.

4.1.5

On Counter-Intuitive Predictions

Sometimes the COMPAS risk score for a particular person does not match the practitioner’s
expectations or clinical judgment regarding the level of risk posed by that person. A case
in point is when an offender with no prior violence history scores medium or high on the
Violent Recidivism Risk Scale. Or, conversely, an offender with some violent history scores
low on the Violent Recidivism Risk Scale. This section explains how this occurs and why it
is not an indication that the risk scale has failed to work properly.

➞2015 Northpointe Inc., All Rights Reserved.

 29

The COMPAS risk scales are actuarial risk assessment instruments. Actuarial risk assessment
is an objective method of estimating the likelihood of reoffending. An individual’s level of
risk is estimated based on known recidivism rates of offenders with similar characteristics.
The Violent Recidivism Risk Scale is constructed from the following characteristics that we
found to be predictive of new person offenses (misdemeanor or felony):
❼ History of Noncompliance Scale
❼ Vocational Education Scale
❼ Current age
❼ Age-at-first-arrest
❼ History of Violence Scale

Each item is multiplied by a weight (w). The size of the weight is determined by the strength
of the item’s relationship to person offense recidivism that we observed in our study data.
The weighted items are then added together to calculate the risk score:

Violent Recidivism Risk Score = (age∗−w)+(age-at-first-arrest∗−w)+(history of violence∗
w) + (vocation education ∗ w) + (history of noncompliance ∗ w)

The strong association of each of these inputs with person offense recidivism that we observed in our studies has been established by many other researchers in criminal justice.
Meta-analytic results show that violent criminal history, education and vocational problems,
current age, and age-at-first-arrest are consistent predictors of violent recidivism. The Violent Recidivism Risk Scale has items in common with many risk assessment instruments
in use in corrections, including the Level of Service Inventory-Revised (LSI-R); the General
Statistical Information on Recidivism (GSIR); the Violence Risk Appraisal Guide (VRAG)
and the Sex Offender Risk Appraisal Guide (SORAG); and the Self-Appraisal Questionnaire
(SAQ).
Your auto insurance company uses a similar risk prediction approach to estimate your risk
of having an accident. Besides your age and accident history, the equation includes other
characteristics such as credit rating and gender. If you are under 25, male, and have poor
credit, you may be classified as high risk even though you have never had an accident.
In the context of Violent Recidivism Risk, if you are young, unemployed and have an early
age-at-first-arrest and a history of supervision failure, you could score medium or high on
the Violence Risk Scale even though you never had a violent offense arrest.
It is possible for a person’s score on the Violent Recidivism Risk Scale to deviate considerably from what one would expect given the person’s score on the History of Violence Scale.

➞2015 Northpointe Inc., All Rights Reserved.

 30

Consider a hypothetical person who scores high (D10) on History of Violence (2 prior misdemeanor assault offense arrests, 1 prior domestic violence offense arrest, 1 violent property
offense arrest, and 1 prior weapons offense arrest); medium (D6) on vocation / education
problems, and low on noncompliance history (D1). This person has a late age at onset (age
at first arrest = 33 yrs) and he is 51 years old. He has no history of noncompliance (D1)
and no vocation or education problems. All of these factors subtract substantially from his
Violent Recidivism Risk score, which falls into decile 3 (D3). Note that age is one of the best
predictors of violent recidivism, and it carries a lot of weight in the Violent Recidivism Risk
Scale calculation. If our hypothetical person were 25 years old and his age at first arrest
were 16 years old, his Violent Recidivism Risk score would jump to D8 (High).
Why Is the Current Offense Not Included in the Risk Score?
The Recidivism Risk Scale does not include current violent offense in its calculation. When
an offender with a current violent offense obtains a Low Score on the Violent Recidivism
Risk Scale, the Low Score may appear counterintuitive. The Violent Recidivism Risk Scale
was trained to predict general violent recidivism (misdemeanor or felony person offense).
During model development we generally find that violent current offense does not significantly
improve the prediction of general violent recidivism. However, an appreciation of the nature
and circumstances of the current offense remains essential for effective case management.
Current violent offenses are captured by the Current Violence Scale.
What About Domestic Assault or Sex Assault Offenses?
For both domestic assault and sex assault, details about the current offense are important
for understanding the risk of recidivism. If the current offense is domestic assault or sexual
assault, then it is recommended to use an index-offense-specific risk tool to assess risk of
recidivism. COMPAS includes secondary assessments for this purpose, including the Vermont
Assessment of Sex Offender Risk-2 (VASOR-2) (McGrath & Hoke, 2001; McGrath, Lasher,
Cumming, Langton, & Hoke, 2014) and the STATIC 99 (Hanson, 1997; Hanson & Thornton,
2000) for use with adult male sex assault offenders and the Revised Domestic Violence
Screening Instrument (DVSI-R) for use with adult domestic violence offenders (Williams &
Grant, 2006; Stansfield & Williams, 2014).
What Percent of the Assessments will have a Counterintuitive Pattern?
There are two counterintuitive patterns: (1) An offender with no prior violence history scores
high on the Violent Recidivism Risk Scale and (2) An offender with high violent history scores
low on the Violent Recidivism Risk Scale. The relative frequency of these patterns depends
on the relative frequency of violent history in the agency population. If a large percent
of the agency population has low violent history then pattern 1 is more likely. If a large
percent of the agency population has high violent history then pattern 2 is more likely. The
alignment between the agency data and the norm data will affect the proportion classified

➞2015 Northpointe Inc., All Rights Reserved.

 31

as high (or low) on the Violent Recidivism Risk Scale, which will also affect the likelihood
of counterintuitive scores.
Cases that have a counterintuitive pattern of History of Violence and Violent Recidivism Risk
should be examined closely and considered for an override. Persons who exhibit pattern 1 are
more likely to have early age at onset and younger age at assessment, and possibly a history
of noncompliance and vocational/educational problems. Persons who exhibit pattern 2 are
more likely to have late age at onset and older age at assessment, with minimal history of
noncompliance and few vocational/educational problems. In all cases a holistic framework to
case formulation should be applied that takes into account the varied aspects of the offender
as measured by the COMPAS risk and needs scales.
General Comments on Risk Prediction
Risk assessment is about predicting group behavior (identifying groups of higher risk offenders) - it is not about prediction at the individual level. Your risk score is estimated based on
known outcomes of groups of offenders who have similar characteristics.
The Violent Recidivism Risk Scale could be constructed in such a way that a high (low) score
can only be obtained for someone who has (doesn’t have) a history of violent offense arrests.
This could be accomplished for example by constructing the Violent Recidivism Risk Scale
entirely (or almost entirely) of violent history items. However, based on our own research
and that of many other researchers, a scale that depends too heavily on violent history items
will not have good predictive power.
Our risk scales are able to identify groups of high-risk offenders - not a particular high-risk
individual. We identify groups of offenders who score high, medium or low-risk. We expect
that the high-risk group will have higher recidivism rates for violent offenses relative to the
low-risk group - this, in fact, has been demonstrated in our outcomes studies.
It is also important to note that we would expect staff to disagree with an actuarial risk
assessment (e.g. COMPAS) in about 10% of the cases due to mitigating or aggravating
circumstances which the computer is not sensitive to. In those cases staff should be encouraged to use their professional judgment and override the computed risk as appropriate documenting it in COMPAS with the Override Reason - for monitoring by supervisory staff.

➞2015 Northpointe Inc., All Rights Reserved.

 32

4.2

Criminogenic Need Scales

Need scales measure a criminogenic need and help with case-planing. In the following section
we briefly describe each COMPAS Core need scale and give examples of the goals and tasks
that might be put into a case-plan.

4.2.1

Cognitive Behavioral

This is a higher order scale that incorporates the concepts and items included in the Criminal Associates, Criminal Opportunity, Criminal Thinking, Early Socialization, and Social
Adjustment scales.
This scale, as mentioned above, includes grouped items which represent areas of need that
can best be addressed in settings that include cognitive restructuring approaches. Concurrent drug/alcohol treatment or other interventions that address immediate needs are
recommended, a balanced approach is necessary to avoid overwhelming the person with interventions. For some people, implementing interventions before they are on community
supervision is the best approach as they will have the opportunity to focus on change their
thoughts, feelings and behavior in a controlled setting without the challenges of a community
setting. When a person scores in the medium and high ranges of this scale, considerations
for their world view must be made, beginning with the question, “does this person see a need
for change?”.
Table 4.1: Case Planning example for Cognitive Behavior
Goal
Task

Task

4.2.2

Build new and increase healthy coping skills
Immediate Needs: Identify sources/triggers of my anger, frustration, and feelings of being overwhelmed. Make separate lists for
each feeling, include what was going on in my immediate surroundings at that moment, who else was there, stressful incidents, and
any other information I think is significant.
Ongoing Needs: Use my healthy coping skills (from my skills
list/optional actions) to problem-solve in situations where I feel
stressed, angry, overwhelmed or when I recognize my triggers to
use old behavior to get through a situation.

Criminal Associates/Peers

An involvement with anti-social friends and associates is one of the “big five” risk factors
for criminality to emerge in meta-analytic research (Gendreau et al., 1996). Affiliating with
aggressive and criminal others is a significant risk factor for further violence and crime. This
is consistent with both social learning theory and sub-cultural theories of crime (Andrews et
al., 1990; Elliot, Huizinga, & Ageton, 1985).

➞2015 Northpointe Inc., All Rights Reserved.

 33

This scale assesses the degree to which a person associates with other persons who are
involved in drugs, criminal offenses or gangs, and determines whether they have a history of
arrests and incarceration. A high score would identify persons who are involved in a network
of highly delinquent friends and associates.
This domain is considered a strong area of influence for people in the criminal justice system.
Interventions in this area can be difficult for the person as their identity with a group as well
as a support system, albeit criminally involved, will be altered. Gang influence is particularly
difficult as a real level of threat could exist for the person who, by leaving/taking a break
from gang life, may be viewed as disrespecting those who have brought him/her to this point
in life. Compliance, rather than change is likely for some people, yet, it is a step forward
with respect to safety and recidivism.
Table 4.2: Case Planning example for Criminal Associates/Peers
Goal
Task
Goal
Task
Task

4.2.3

Increase my association with pro-social, healthy friends
Immediate Needs: Identify traits and behavior of positive, healthy
friends and family members
Reduce interactions with anti-social, potentially harmful friends
Identify friends and family who I tend to get into trouble with,
include any co-defendants or criminally involved associates
Create a plan to avoid interaction with criminally oriented
friends/family, include statements regarding what my actions will
be if I come into contact with the friends/family I have listed as
“trouble” for me.

Criminal Involvement

The degree of criminal involvement has consistently emerged as a major risk factor for
predicting ongoing criminal behavior. It is the most important of the major risk factors that
have emerged in various meta-analysis studies (Gendreau et al., 1996; Andrews & Bonta,
1994). Early juvenile delinquency involvement has also been linked to ongoing criminal
behavior (Moffitt, 1993).
This scale is defined by the extent of the person’s involvement in the criminal justice system.
A high score indicates a person who has had multiple arrests, multiple convictions, and prior
incarcerations. The items centrally defining this scale are the number of arrests and number
of convictions. A low score identifies the person who is either a first-time arrest or has
minimal criminal history. Thus, the central meaning of this scale is the extensiveness of the
criminal history.
Arrest history is useful here to see patterns (persons, places, things, time of year) and
other related elements that could be antecedents to recidivism and perhaps causal factors
(thoughts, feelings, beliefs, attitudes) that can be impacted by intervention. Cognitive behavioral approaches seem to work best in this life area to re-set a person’s response to triggers
and patterned responses.

➞2015 Northpointe Inc., All Rights Reserved.

 34

Case planning will be similar to criminal associates/peers, criminal personality and criminal
opportunity and some cognitive behavioral goals. See the goals listed in Table 4.2, 4.4 and
4.3.

4.2.4

Criminal Opportunity

We have developed a higher order scale to assess the concept of criminal opportunity. This
scale emerges from those criminological theories that stress the importance of routine daily
activities and the importance of occupying certain social roles (marriage, parenting, being an
employee). These roles tend to structure a person’s daily activities in a pro-social manner,
fostering social bonds and associated local social controls. The theoretical background to this
scale includes routine activities theory that emphasizes the importance of immediate local
daily activities that place a person in high risk or high opportunity situations (L. E. Cohen
& Felson, 1979). The second theoretical theme contributing to this scale is early social
control theory (Hirschi, 1969) which emphasizes the importance of social bonds as inhibitors
or constraints to crime. The third theoretical strand in which the concept of opportunity is
important is the “life cycle” theory of Sampson et.al. (1994). This asserts that age related
desistance from crime is linked to life cycle changes that increase both social bonds (wives,
children, jobs) and the immediate social controls of associated roles.
This higher order scale assesses criminal opportunity by using items that represent a combination of the following: time in high crime situations, affiliation with high risk persons
who often engage in illegal activities, an absence of pro-social or constructive activities (e.g.
working, spending time with family, etc.), an absence of social ties, high boredom, high restlessness and being in a high risk age group. The central items include: being unemployed,
living in a high crime area, having friends who engage in drug use, and having no constructive
activities.
A variety of life areas are represented within this scale. Interventions can be put in place
in concurrent waves—for example, seeking out new friends and activities that are pro-social
and have positive elements such as learning new skills, helping others, gaining awareness
and acting on the awareness at the same time. Structure is a key ingredient in reworking
previously idle or non-constructive time. Performance measures as a means of accountability
and tracking behavior are also useful tools in this area.
Table 4.3: Case Planning example for Criminal Opportunity
Goal
Task

Task

Increase positive activities
Immediate Needs: Set a date and time for any new activities to
help me follow through with the plans I make for new, positive
activities.
Ongoing Needs: Develop career aspirations, goals, and identify potential role models as a way to connect with others outside of my
family as a means to move forward. Create a plan with each item
listed, including dates, for behavioral actions on my part.

➞2015 Northpointe Inc., All Rights Reserved.

 35

4.2.5

Criminal Personality

Several personality dimensions have emerged from recent research as significantly related
to persistent criminality. These dimensions involve impulsivity, risk-taking, restlessness and
boredom, absence of guilt (callousness), selfishness and narcissism, interpersonal dominance,
anger and hostility, and a tendency to exploit others (Hare, 1991; Cooke, Forth, & Hare,
1998). Bonta (1996) reports that criminal personality was the second most important dynamic factor in predicting recidivism. Bandura (1996) also reports validating similar personality dimensions. Criminal personality was one of the ”big five” risk factors for criminality
in the meta-analysis of Gendreau (1996). The well known General Theory of Crime proposed by Gottfredson and Hirschi (1990) similarly invokes the personality concept of “low
self-control,” which includes similar dimensions of criminal personality. Prior research has
demonstrated a modest but significant relationship between psychopathy, low self-control
(variously defined) and both violence and general criminal behavior (Quinsey et al., 1998).
Quinsey el al. (1998) include the PCL (Hare, 1991) within their violence risk predictive
system – the VRAG.
The items in this scale cover the main dimensions identified as components of the criminal
personality (e.g. impulsivity, no guilt, selfishness/narcissism, a tendency to dominate others,
risk-taking, and a violent temper or aggression).
Personality is a complex concept and many social scientists believe personality is “set” in
childhood/adolescence. Given that many factors come together to create personality, the
idea of criminal personality is no less complicated. There are patterns seen in persons
who exhibit criminal personality traits. Intervention then, is based on cognitive behavioral
approaches that examine and offer alternatives to thoughts, feelings, beliefs and resultant
criminal behavior. A specific diagnosis of anti-social personality disorder is not necessary
when considering intervention, the area of focus is as listed above – what is the process the
person undergoes while deciding to engage in criminal behavior, what is his/her rationale,
and what is he/she willing to do about making changes?
Table 4.4: Case Planning example for Criminal Personality
Goal
Task

4.2.6

Build new and increase positive coping and communication skills.
Immediate Needs: Journal my behavior in the areas of thoughts,
feelings, attitudes and resultant behavior when I feel stressed, angry, or that something unfair has happened to me. Do my journal
entries daily for 5 days and bring to my next probation appointment.

Criminal Thinking Self-Report

Antisocial attitudes and beliefs are identified among the“big five”risk factors in meta-analysis
studies of factors that predict crime (Gendreau et al., 1996). However, there is no agreement
on the particular attitudinal dimensions or cognitions that are the most useful for predictive purposes. Various studies focus on aspects of thinking style, attitudes toward criminal

➞2015 Northpointe Inc., All Rights Reserved.

 36

justice, neutralization and excuses, tolerance for law violation, cognitive justifications, etc.
Clearly, this area could require a highly extensive inventory to map the full range of cognitive
dimensions relative to crime. In the absence of such consensus, we adapted the approach of
Bandura (1996). Bandura’s approach assesses several key cognitive dimensions that justify,
excuse, and minimize any damage caused by the person’s behavior/crime.
This scale brings together several cognitions that serve to justify, support, or provide rationalizations for the person’s criminal behavior. These dimensions include moral justification,
refusal to accept responsibility, blaming the victim, and rationalizations (excuses) that minimize the seriousness and consequences of their criminal activity. These include rationalizations such as: drug use is harmless because it doesn’t hurt anybody else, criminal behavior
can be justified by social pressures, theft is harmless if those stolen from don’t notice or
don’t need what was taken, etc.
The concepts discussed above as they relate to the Criminal Personality scale are also present
in this scale, and have been identified in further detail through the person’s own self-report.
A distinct pattern of rationalizations for criminal and/or harmful behavior is present for
those who score in the probable and highly probable categories. Interventions that focus primarily on cognitive behavioral approaches tend work best with those who evidence significant
criminal thinking.
Table 4.5: Case Planning example for Criminal Thinking Self-Report
Goal
Task

4.2.7

Modify criminal thinking, develop a positive attitude toward various life areas (see specific goals).
Immediate Needs: Create a list of what works for me (positive
thoughts and activities) and what doesn’t (negative thoughts and
activities) that keep me in the same cycle of getting into trouble.

Current Violence

This scale forms part of the general criminal history and measures the degree of violence in
the present offense. The central item that defines the scale is the presence of an assaultive
felony. Other key items involve whether or not a weapon was used, if there was injury to a
person, etc.
Research has shown that the level of violence in the instant offense is NOT a good predictor
of future crime. Keeping in mind the degree and type of violence in the instant offense as
compared to the person’s history of violence and current level of functioning/needs scores is
good practice. One area for clear consideration is that of family violence and how this will
affect any kind of living arrangement for community-based supervision.

4.2.8

Family Criminality

From a social learning theory perspective, participation in criminal behavior may be facilitated by significant others who model such behavior. Research has consistently demonstrated

➞2015 Northpointe Inc., All Rights Reserved.

 37
Table 4.6: Case Planning example for Current Violence
Goal
Task

Task

Build new and increase healthy coping skills.
Enroll in and successfully complete an Anger Management Program. Bring my homework and updates from the class to each
probation appointment.
Create a list of my healthy and unhealthy coping skills, then list
next to each one the usual outcome when I use that reaction or
response.

that delinquency and adult crime are both associated with parent criminality (West, 1973;
Lykken, 1995). Children may learn that violent and deviant behavior “work” in the context of their family. Aside from the social learning and role modeling perspective, other
intergenerational mechanisms may operate to transmit values and behaviors from parent
to child. Genetic influences, for example, may operate to transmit anti-social personality
disorder and criminality (Lykken, 1995). COMPAS therefore includes a measure of family criminality focusing on the criminality and drug use history of the mother, father, and
siblings.
This scale assesses the degree to which the person’s family members (mother, father, and
siblings) have been involved in criminal activity, drugs, or alcohol abuse. The items cover:
arrests of each family member, whether they have been in jail or prison, and whether the
parent or parental figure has a history of alcohol or drug problems.
Families can be significant positive resources for any person in the criminal justice system.
The presence of family criminality, however, can create a dichotomous situation in that, on
the one hand the family is a source of support, comfort, and hope, and on the other hand,
they may also be criminally involved and their support revolves around their criminal activity
and belief systems.
Table 4.7: Case Planning example for Family Criminality
Goal
Task
Task

4.2.9

Eliminate criminal involvement with family members.
Immediate Needs: List/identify family members (those who I have a
relationship with and spend time with) who are criminally involved.
Ongoing Needs: Create a time line of my involvement with these
family members and the consequences/benefits of spending time
with them, e.g. when did it happen and what happened while we
were together.

Financial Problems

This concept appears as one of the more modest risk factors in the Gendreau et al. (1996)
meta-analysis. It is linked to lower social class, poor housing, community disorganization,
and other factors. Homicides, for example, are disproportionately found in high poverty
areas. Numerous social dimensions related to poverty are linked to high crime, including

➞2015 Northpointe Inc., All Rights Reserved.

 38

residential mobility, family disruption, single parent families, crowded housing conditions,
and higher opportunity for violence (Sampson & Lauritsen, 1994). The measure of poverty
and financial problems in COMPAS focuses on the struggle to survive financially, problems
paying bills and other issues related to a shortage of money.
This scale assesses the degree to which a person experiences poverty and financial problems.
It assesses whether the person worries about financial survival, has trouble paying bills, and
has conflicts with friends or family over money.
Unpredictable economic times may play a role in this area, however, a person’s pattern
of earning (or not) and spending money is an important element. Education on money
management and fulfilling court ordered financial commitments is part of the necessary
approach when considering interventions. Assuming someone knows how to manage their
finances is an erroneous starting place, vocational training may also play a role in creating a
successful change plan.
Table 4.8: Case Planning example for Financial Problem Scale
Goal
Task
Task

4.2.10

Gain financial stability/independence.
Immediate Needs: Apply for financial assistance/emergency shelter
and/or food stamps (use other resources as referred by PO).
Immediate Needs: Inform my supervisor at work about my probation appointments and any terms and conditions that might impact
my ability to do my job.

History of Non-Compliance

This scale focuses on the number of times a person has failed when he or she has been
supervised in the community (probation or parole). The central defining items are the
number of times that probation or parole has been violated or revoked. Related items
include the number of times a new charge or arrest has occurred while the person was on
probation and the number of returns to custody for parole violations.
This scale focuses on the number of times the person has failed when he or she has been placed
on a community-based status. The central defining item is the number of times probation
or parole has been suspended or revoked. Related items include the number of times the
person has failed to appear for a court hearing, the number of times a new charge/arrest
or technical rules violation has occurred while on probation, parole and prior community
corrections program placement failures (i.e. electronic monitoring, community service work,
day reporting, etc.) Thus, the scale involves the risk of technical rules violation failure leading
to revocation of probation, pretrial release, or community corrections placement status.
Different states/agencies have different thresholds for supervision violation and suspension/revocation. While policy decisions do effect the person’s history ”on paper” it is
also important to understand the person’s willingness and ability to successfully complete
community-based supervision. Clearly articulated expectations with terms and conditions

➞2015 Northpointe Inc., All Rights Reserved.

 39

of supervision and case planning are key factors in laying the groundwork for success. Behaviorally stated goals and a high degree of structure with room for individual differences
and learning curves could enhance a person’s success rate.
Table 4.9: Case Planning example for History of Non-Compliance
Goal
Task

Note

4.2.11

Attend all probation meetings as scheduled.
Immediate Needs: Client and PO agree upon appointments for two
week intervals including attendance at Cog/Behavioral group 1x
week. Client to use pocket calendar for personal reminder of all
appointments, during this two week period (March 10-24, 2010)
client is to attend 2 scheduled appointments at this office (2/12
and 2/19 at 3pm) and the cog group on 2/15 and 2/22 at 6pm.
All case planning activities should include tangible sanctions should
the person fail to comply or engage in change behavior, and in the
cases when a very high degree of structure is put in place, those
sanctions may be stated on the case plan.

History of Violence

A history of violent behavior has been demonstrated to be one of the most powerful predictors
of future violence (Farrington, 1991; Parker & Asher, 1987). The likelihood of future violence
appears to steadily increase with each instance of a prior violent incident. Each prior arrest
for violent behavior increases the likelihood of further violence. Similarly, a history of juvenile
violence has been found to be a predictor of adult violence (Farrington, 1991).
The aim of this scale is to reflect the seriousness and extent of violence in an individual’s
criminal history. It focuses on the frequency with which violent felony offenses have occurred,
the use of weapons, and the frequency of injuries to victims. The frequency of several specific
violent offenses are also included in the scale (e.g., robbery, homicide, and assaultive offenses).
Multiple episodes of violence may suggest the need for further psychological evaluation. The
accumulation of multiples (events, victims, types of crimes against persons/animals) creates a
pattern of serious concern. Interventions may be targeted at cognitive behavioral constructs
to manage behavior, and highly structured supervision may be preferred by the supervising
agency.
While we are not going to change the past, we can teach people to intervene in old thought
processes and put in place, new, healthier thoughts that lead to pro-social responses rather
than reactions that always follow the same patterns.

4.2.12

Leisure/Boredom

Aimlessness in the use of leisure time is linked to several theories of crime. For example, it is
a component of Hirschi’s early Social Control theory representing an aspect of weak external

➞2015 Northpointe Inc., All Rights Reserved.

 40
Table 4.10: Case Planning example for History of Violence
Goal
Task
Task
Task

Increase my healthy responses to events that trigger an angry reaction for me.
Immediate Needs: List the way I have shown my thoughts and
feelings in the past.
Immediate Needs: Describe what happens when I lose self-control.
Immediate Needs: Describe what happens when I use positive, selfcontrol responses.

social bonding (Hirschi, 1969). Aimless use of leisure time is also included as a risk factor
in the LSI (Andrews & Bonta, 1994). The General Theory of Crime (M. R. Gottfredson &
Hirschi, 1990) includes aimlessness and the related concept of proneness to boredom within
the dimension of low self-control or criminal personality. It is also linked to routine activities
theory by the maxim of “Idle hands are the devil’s workshop” (Osgood, Wilson, O’Malley,
Bachman, & Johnston, 1996).
This scale assesses the degree to which the person experiences feelings of boredom, restlessness, or an inability to maintain interest in a single activity for any length of time. Thus,
this scale may be regarded as reflecting a psychological dimension rather than representing
the amount of constructive opportunities in the person’s community environment.
As noted above, the issue is not necessarily time management, but the person’s value of
experiences and relationships. Creating an understanding of these elements may be a first
step toward making changes for the individual. Some social or information processing issues may be identified through further assessment, and these issues can then be addressed
accordingly.
Table 4.11: Case Planning example for Leisure/Boredom
Goal

Task

4.2.13

Learn about the relationship between my level of participation
with other people/events/interests and my ability to be involved
in things outside of work or other required activities.
Immediate Needs: Create a plan for getting involved with my
friends who participate in the basketball league at the rec center.
List the night and time of the league and the person who I can talk
to get on a team. Ask my friend to go with me if I feel like I need
support in joining the team.

Residential Instability

An unstable lifestyle is one aspect of the second factor of Hare’s Psychopathy Checklist and
this is an obvious risk factor for crime and violence (Hare, 1991). Additionally, low social ties
and an unstable residential address are often used in pre-trial risk assessment instruments to
predict risk of flight. The absence of social ties, and the presence of social isolation are also

➞2015 Northpointe Inc., All Rights Reserved.

 41

seen in Social Control theory as the absence of restraints on deviant behavior that result
from weak social bonding. In addition, since change and stress are correlated, an unstable
lifestyle may be stressful. Finally, personal stress/distress appears as a risk factor with
modest predictive validity in meta-analysis studies (Gendreau et al., 1996).
The items in this scale measure the degree to which the individual has long term ties to
the community. A low score on this scale indicates a person who has a stable and verifiable
address, local telephone and long term local ties. A high score would indicate a person who
has no regular living situation, has lived at the present address for a short time, is isolated
from family, has no telephone, and frequently changes residences.
Community-based supervision requires a verifiable address. The reality is that some individuals end up in shelters right after release, or, they don’t have the financial means to
secure acceptable living quarters for months after sentencing/release. The historical nature
of the person’s residential stability is good information while the person is incarcerated in
that planning can be put into place to avoid the pitfalls aforementioned. Renewing and/or
creating family contacts and other potential support resources can be used as realistic goals
in establishing residential stability.
Table 4.12: Case Planning example for Residential Instability
Goal
Task

4.2.14

Seek and obtain sustainable living situation.
Ongoing Needs: Develop a workable budget that includes housing
costs that I did not list under my immediate needs such as pets,
additional furnishings, any agreements that I can lawfully enter into
to help reduce the cost of my rent.

Social Adjustment

Interpersonal problems may exist in each main social institution (family, school, work, etc.)
A pattern of interpersonal problems may indicate poor social skills. The present higher order scale was constructed to assess the recurrence of interpersonal problems across various
social contexts. Social skills training is often advocated as a treatment approach in preventing further violence and crime. Social adjustment problems are also implicated in several
theoretical perspectives of criminal behavior (e.g., weak social bonding in social control theory (Hirschi, 1969), stress (Gendreau et al., 1996) social cognitive models of crime (Dodge,
Pettit, McClaskey, & Brown, 1986; Dodge, 1998) and the erosion of social capital (Hagan,
1998)).
This scale is higher order in the sense that it uses items from other scales that crosscut
several domains. It aims to capture the degree to which a person is unsuccessful and conflicted in his/her social adjustment in several of the main social institutions (school, work,
family, marriage, relationships, financial.) A high score indicates a person who has been fired
from jobs, had conflict at school, failed at school or work, has conflict with family, exhibits
family violence, cannot pay bills, has conflicts over money, etc. Thus, the common theme is
problematic social relationships across several key social institutions.

➞2015 Northpointe Inc., All Rights Reserved.

 42

Areas for intervention will depend on the most pressing issue and need for support in that
area. Creating a sense of connectedness and responsibility for self and to others is a foundational element of many cognitive behavioral approaches. Structuring communication expectations and methodologies for the individual may be a starting place, many programs
provide sequenced awareness and practice options. The supervision professional may work
with the individual in identifying other community-based pro-social activities, as well.
Table 4.13: Case Planning example for Social Adjustment
Goal
Task

4.2.15

Increase positive social supports with family, friends, and community.
Immediate Needs: Create a plan for increasing my time spent with
positive, pro-social friends and family members.

Social Environment

Living in a high crime neighborhood is a well-established correlate of both delinquency and
adult crime (Thornberry, Huizinga, & Loeber, 1995; Sampson & Lauritsen, 1994). This risk
factor fits into several theoretical models of crime and delinquency (e.g., social disorganization, social learning, and sub-cultural theories). Disorganized and high crime communities
are characterized by perceived high crime rates, gangs, easy access to drugs, and inadequate
housing.
This scale focuses on the amount of crime, disorder, and victimization potential in the
neighborhood in which a person lives. High crime is indicated by the presence of gangs, ease
of obtaining drugs, the likelihood of being victimized, a belief that a weapon is needed for
protection, and so on.
Few scales reflect areas where the person has no direct control over the identified issues,
however, this scale is based on environmental factors that the individual has to cope with
on a daily basis. Problem-solving around the possibility of relocating or finding a safer
living arrangement may be paramount. Other risk factors come into play when considering
the person’s social environment (criminal opportunity, criminal peers, family criminality,
residential instability, etc.) and these factors may become more of a primary focus should
they be identified as active in the person’s life.
The Social Environment and Social Isolation scales will typically use case planning language
similarly. Increasing positive family and peer relationships, as we have seen in other scales
is a primary focus, as well as involvement in specific activities.

4.2.16

Social Isolation

Positive social supports appear to serve several functions that may reduce crime and violence.
Social support may act as a protective factor or mediator of stress, since stress and anxiety
may predispose a person towards anger and violence. Positive social support has been shown

➞2015 Northpointe Inc., All Rights Reserved.

 43

in research to act as a protective factor against risk of violence even in high risk environments
(Estroff & Zimmer, 1994). As described below, the COMPAS social isolation scale is bipolar
in that it serves to identify social isolation/loner behavior on one pole and strong social
supports at the other pole.
This scale assesses the degree to which the person has a supportive social network and is
both accepted and well integrated into this network. The scale is scored such that a high
score represents an absence of support, and the presence of feelings of social isolation and
loneliness. The defining items include: feeling close to friends, feeling left out of things, the
presence of companionship, having a close best friend, feeling lonely, etc.
As mentioned in other social support areas, intervention can be across many dimensions and
impact the person on both the awareness and practice levels. Strategies might include finding a mentor, joining known pro-social or support groups, learning new skills/hobbies, and
creating new social connections where the person’s new, healthy behavior will be expected
by those involved in the activities.

4.2.17

Socialization Failure

Socialization failure during childhood and adolescence has been consistently linked to crime
and delinquency. Problems in the family and inadequate parenting are the critical background issues (Lykken, 1995). We have constructed a higher order factor in COMPAS that
builds on the early onset of delinquency, problem behavior in school (dropout, suspensions,
fighting, etc.), inadequate parental socialization, and early drug use. These are all well
known risk factors for later criminality (Chaiken, Chaiken, & Rhodes, 1994; Lykken, 1995)
and all represent early socialization problems. Lykken (1995) in particular, explores the link
between socialization failure and criminal behavior in his concept of the sociopath.
This scale combines items reflecting family problems, early school problems, and early delinquency, all of which suggest socialization failure (how the person was socialized growing
up). The intent is to examine socialization breakdown through its early indicators in school,
delinquency, and family problems. A high score would represent a person whose parents
were jailed or convicted or had alcohol or drug problems. In addition, a high score is associated with early behavior problems in school (being expelled, failing grades, skipping classes,
fighting) and would also manifest serious delinquency problems.
This scale looks at the history or pathway that was involved in the person’s upbringing
that may have significantly affected his/her view of the world in terms of trust, respect
for reasonable authority, value of others, and the development of beliefs and attitudes that
are active and present today. High scoring individuals may need cognitive restructuring
programs to assist in an awareness of, and change plan for, some of the beliefs and attitudes
that lead to troublesome behavior for the person.

4.2.18

Substance Abuse

Numerous published research studies have established that substance abuse is a significant
risk factor for both general criminal behavior and violent behavior. Substance abuse emerged

➞2015 Northpointe Inc., All Rights Reserved.

 44
Table 4.14: Case Planning example for Early Socialization Failure
Goal
Task
Activity

Note

Build new and increase my positive coping skills and responses.
Attend and successfully complete cognitive behavioral program.
Complete first exercise in workbook by 3-20-10 and bring the completed exercise to the next probation appointment. Participate in
the cog group by engaging the exercise on My Thoughts, and participating in the role play discussion.
In the case of a structured, sequenced program, case planning will
often be stated as in the example above.

as one of the major risk factors in the meta-analysis studies of Gendreau et al. (1996).
The present scale is a general indicator of substance abuse problems. A high score suggests
a person has drug or alcohol problems and may need substance abuse intervention. The
items in this scale cover prior treatment for alcohol or drug problems, drunk driving arrests,
blaming drugs or alcohol for present problems, drug use as a juvenile, and so on.
The cut points on this scale are lower than the other needs scales due to the design of the
scale. A person who scores in the Probable range (3-4) is considered a person who is in need
of further evaluation (i.e. ASI, SASSI, etc.) and a person who scores in the Highly Probable
range (5-10) may have a serious alcohol or drug problem requiring a structured treatment
approach. Because of the high incidence of drug/alcohol abuse within the criminal justice
population, a primary intervention for many individuals to impact recidivism is assisting the
person to attain and maintain sobriety.
Substance abuse typically intersects every life area for a person. Therefore, cognitive behavioral restructuring and life skills planning may be needed following, or, in some cases during,
treatment. Case planning language varies in this area between the example shown under
the Socialization Failure (Table 4.14) scale regarding structured, sequenced steps, and, the
use of supervision focused goals and tasks as listed in Table 4.15.
Table 4.15: Case Planning example for Substance Abuse
Goal
Task
Task

4.2.19

Maintain Sobriety
Attend AA meetings 3 times per week and show my attendance
card to my PO at each meeting.
Call in for UA/BAC testing daily and report by 5pm on the day I
am to do my testing.

Vocation/Education

Another of the “big five” risk factors for crime and recidivism prediction in the Gendreau et
al. (1996) meta-analysis is labeled “social achievement.” This concept is an amalgam of educational attainment, vocational skills, job opportunities, a record of stable employment, good
income, and, more generally, the level of legitimate economic opportunity. Basically, persons

➞2015 Northpointe Inc., All Rights Reserved.

 45

with more social capital have higher “life chances” than other persons who may have very
restricted opportunities for success (Hagan, 1998; Coleman, 1990). The family is of critical
importance in building social capital. Parents either transmit positive and substantial social
capital to their child or fail in the socialization process. This scale is a higher order factor
in COMPAS, using items from both educational and vocational domains. Individuals differ
greatly in access to social capital or other resources. Social capital is somewhat dynamic.
It can be built or destroyed. For example, a record of serious criminal behavior or high
school dropout will clearly diminish life chances and social resources, whereas completing a
job skills training course or obtaining a GED may increase these chances.
This higher order scale assesses the degree of success or failure in the areas of work and
education. A high score represents a lack of resources. Those who score high will present
a combination of failure to complete high school, suspension or expulsion from school, poor
grades, no job skills, no current job, poor employment history, access only to minimum wage
jobs, etc. Thus, the scale represents a lack of educational and/or vocational resources.
A score in the Probable range is significant in that a person may be struggling to seek and
maintain employment that meets his/her skill set, ability, and interests. Vocational stability
plays a significant role in success on community supervision. Intervention can therefore be
initiated during incarceration or upon release. Education, or additional training may be the
reasonable answer to assisting the person to maintain employment. Therefore, it is important
to look at the whole picture in this domain when assessing paths and barriers to success.
Table 4.16: Case Planning example for Vocation/Eduation
Goal
Task

Task

Task

4.2.20

Develop vocational skills
Immediate Needs: Ask myself what it will take to meet the goals I
am setting, identify barriers that come from others/situations, and
those that I have put in place.
Immediate Needs: Identify methods to break down the barriers that
I have put in place, use my resources (supervisor, PO, instructor)
to move forward with my plan.
Enroll in vocational training program using the funding source I
found when I contacted the instructor at the school.

The Lie Scale and Random Responding Test

These validity tests provide alerts that the person being assessed by COMPAS is possibly
“faking good” or is responding randomly.
Items in the Lie Scale include questions about feeling unhappy or angry with the options
across a Likert scale that include “never.” Since it is highly unlikely that a person has never
felt unhappy or angry, the selection of “never” would suggest they are not telling the truth, or
perhaps they are being careless with their responses. If several of the items on the Lie scale
are given extreme answers, the criminal justice professional is then alerted to the possibility
that the person is not responding truthfully.

➞2015 Northpointe Inc., All Rights Reserved.

 46

The Random Responding scale is based on 37 highly correlated pairs of COMPAS scale
items. Some items appear more than once in the pairs as they relate to more than one
construct. Random responding has the effect of breaking these correlations. The cutting
score was internally set up to detect the 5% of the respondents at the extreme end of the
distribution who might be answering the questionnaire in a random fashion.

➞2015 Northpointe Inc., All Rights Reserved.

 Chapter 5
Typology
The fact that people respond differently to different treatments has been labeled as responsivity and repeats the conventional wisdom that “one man’s meat is another man’s poison.”
It indicates that the wrong treatment may make things worse and creates a need for careful matching of people to specific treatments. This is central to both “What Works” and
to the Risk-Needs-Responsivity (RNR) model. It also underlies Evidence-Based Practice
(EBP), since incorrect matching of a person to treatment may sabotage the effectiveness
of virtually any intervention. Thus, a challenge for treatment providers is to match intake
assessments to service plans in order to achieve good outcomes. Andrews et al. (2006) recently acknowledge that specific responsivity or differential matching is the least explored
of all the RNR principles. The traditional strategy for “’matching” has been to develop
treatment-relevant classifications to guide differential matching (Warren, 1971; Megargee &
Bohn, 1979; S. Baird, Heinz, & Bemus, 1979). Most of these classification efforts failed
because of a variety of technical problems (Harris & Jones, 1999).
However, we have developed risk and need typologies to facilitate the goals of specific responsivity and to guide the matching of interventions to client needs in the context of the
COMPAS system. We have developed treatment-relevant typologies for both males and females. These are now included as a standard component of the COMPAS software. These
typologies use advanced pattern recognition, cross-validation procedures and multiple methods to verify the stability of the typologies. Each person is now automatically classified on
the basis of “best fit” to one of several standard and replicated needs profiles. The class
profile of each person is automatically produced as part of the standard report to help treatment staff conceptualize the “kind” of client they are dealing with, and to develop a service
plan to meet the specific responsivity needs of that unique individual. It is important to
realize that no person is a perfect match to his/her class and will be unique in his/her overall
pattern of risks and needs. However, his/her assigned prototype membership will suggest a
beginning “framework” for a case plan that may then be customized according to the unique
risk and need patterns of each person. Thus, the default treatment plans for each prototype
will provide treatment staff a useful initial guide to the most likely kind of service plan for
each individual.
The scales required to determine a type in the COMPAS Core typology are: Criminal As47

 48

sociates, Substance Abuse, Financial Problems, Vocation/Education, Family Criminality,
Social Environment, Leisure/Boredom, Residential Instability, Social Isolation, Criminal Attitudes, Criminal Personality, and Age at Assessment.

5.1

Interpretation

Questions may arise as to how to interpret the COMPAS typology assignment and how
to integrate it into the case plan. Overall, we suggest that the typology results should be
interpreted in the context of the other three key classification elements that are provided in
the overall COMPAS Risk Assessment. These are as follows:
1. Risk Potential Scales (Predictive levels): These two (red) scales represent overall risk
potential scales. They include separate risk scales for Violent Recidivism and General
Recidivism.
2. Risk and Need Profiles (Prior history): Next, the profile chart provides the person’s
decile scores on all background scales (e.g., criminal history, drugs, peers, family,
work/education, etc.). These provide the basic data elements that drive risk predictions, needs assessment and treatment plans.
3. Explanatory Typology: This provides the closest fit of each person to one of eight
prototypical categories. The eight types represent different kinds of people. It is
important to remember that the profile chart of any individual person will never be an
exact match to his closest prototype. Many people are hybrids that may not fit well
into any typology.
These three elements may be used collectively to guide case formulation and to understand
what is “going on” with a case, and to select supervision levels and treatment interventions.
Other important elements that may influence case formulation are as follows:
Recommended Level of Supervision: The recommended level of supervision is found in
the Assessment Summary section. The Violence and Recidivism risk potential factors are
the main drivers of this recommendation.
Overrides of the supervision level: Overrides of the calculated recommended supervision
level are clearly appropriate when it is felt that the automated procedure is either over- or
under-estimating the risk level. This is especially true when the screener can identify the
presence of mitigating or aggravating factors. Examples of mitigating factors are such things
as your own street knowledge of the person, age and any extended periods of crime free
behavior, etc. Aggravating factors are such things as severity of offense, gang membership,
your knowledge of their street behavior, of non-apprehended crimes, or concerns on the Lie
Scale or Random Response Score (as applicable).
Common Prototypes versus Anomalous cases: There are several things to understand
about the typology label:

➞2015 Northpointe Inc., All Rights Reserved.

 49

1. The typologies represent “Common Types” of people: We have found that there are
eight common categories or prototypical offending and behavior patterns that often
re-appear in criminal justice populations. These eight prototypes are described in the
software and the software assigns each client to their nearest prototype. However,
please remember that no individual is ever an exact match to his/her typology. In
most cases there will be a good match to the closest fitting category, but will always
have some differences to the ideal prototype. However, some cases will NOT be a
good match to any prototype, or may straddle the boundaries between two prototypes.
These boundary or hybrid cases are not given a prototype assignment and must be
interpreted as unique cases.
2. What to do with the poor fitting/boundary cases: With boundary or hybrid cases, the
typology should be ignored, or used as a starting point for a more individualized interpretation. Such boundary types are often harder to interpret and are more complex.
If the screener’s judgment clearly disagrees with the computer-assigned prototype then
an override is appropriate. The anomaly should be reported and the counselor will
interpret the case using the individual’s case chart and other relevant information to
determine processing and treatment plans.
3. Typology Purposes are explanatory and for treatment planning: A main purpose of
the typology is to give an alert if a case belongs to one of the major case types (e.g.,
a young streetwise gang member; an older repeat drinking driver, etc). If a case is a
good fit this may help in understanding the case and it’s treatment needs since such
kinds of cases will have been seen before.
4. The Typology is not a risk classification! The typology emphasizes explanatory and
need profiles and treatment: The typology prototypes represent diverse profiles of need
factors, and are not designed as a predictive risk classification. Thus, the typology alone
should not be used to determine risk levels but it may often help in risk and placement
decisions if used in conjunction with the risk scales.

➞2015 Northpointe Inc., All Rights Reserved.

 50

5.2
5.2.1

Male Typology
Type Descriptions

Category 1 - Chronic drug abusers – most non-violent
The central theme of this prototype is long-term substance abuse and non-violent offences.
For example, serious substance abuse and use of alcohol/drugs at the current arrest. Problems often begin in adolescence, for example, with first arrests around age 16 or 17. Factors
underlying this type may include mixtures of family criminality, family disorganization, outof-home placements and some juvenile socialization problems. The profile appears in all
ethnic groups, but especially young Anglos. The social context does not suggest total social exclusion. For example, some members have relatively few social risk factors and some
strengths such as low poverty, educational-vocational resources, stable residence in good
neighborhoods and are not isolated, bored or socially rejected. Anti-social personality and
extreme criminal attitudes are mostly absent.
Official criminal histories support this profile. This type averages of 3 to 4 prior arrests mostly
for drug use or trafficking. This category is mostly non-violent with relatively low current
violence, low weapon offences and low victim injuries – although in some cases the current
charge includes assault. There is little evidence of domestic violence and sex offences.
Category 2 - Low risk situational – fighting/domestic violence caution
This type has several economic and educational “strengths” suggesting an apparently normal
citizen. They mostly avoid criminal associates and follow a low risk lifestyle. However, some
members of this group are involved in serious violence, thus caution is warranted. These
persons generally are not raised in high crime families, avoid drugs, have stable addresses
in safe areas and few financial problems. Personality and criminal attitudes appear average.
The profile offers no clear social or criminogenic explanation for offending or for violence.
This pattern may reflect the well known accidental or situational event that unexpectedly
occurs to create serious violence and an arrest situation.
The official criminal history reflects a low risk profile. The group, as a whole, has fewer
official arrests, convictions or prior violence than other types. The official data shows lower
violence history, lower weapons use, lower non-compliance, fewer probation episodes and
almost no burglaries, robberies, The current offense often is for DUI, substance abuse or
an assault (fight/no weapons). Many are incarcerated for the first time. However, as noted
above, some members of this group have been charged with a serious assault and/or domestic
violence. This category occurs in all ethnic or racial groups – a variant is found in Category
8.
Category 3 - Chronic alcohol problems – DUI, domestic violence
The dominant pattern of this category consists of older (40+), mostly relatively well-educated
men who function fairly well with stable jobs, finances and residences, but with recurrent

➞2015 Northpointe Inc., All Rights Reserved.

 51

alcohol problems and a history of DUI and/or domestic violence. They show the oldest age
at first arrest (27) and are thus late starters. A generally low risk lifestyle is reflected by few
criminal peers, educational-vocational and financial success, low crime families, stable and
safe addresses and pro-social structured leisure. They mostly avoid high-risk situations and
do not appear to hold anti-social attitudes or personalities. Thus, the explanation for their
offending would appear to relate to alcohol proneness perhaps in a context of family stress,
rather than social exclusion or environmental explanations.
The official data corroborates this pattern showing that this group has the highest score for
current DUI arrest and using alcohol (but not drugs) at the current arrest. Overall, they
have average criminal involvement and few violent offenses. However, domestic violence also
occurs for some of these people. DUI and alcohol abuse are the major problems since the category has lower clusters arrest rates than other clusters for current violence, weapon arrests,
assaults, juvenile felony arrests, fraud, property, burglary and robbery offenses. COMPAS
risk scales assign this category to low risk, although this is influenced by their older age
(since age lowers risk scores in the risk equations). Thus, they may be expected to have a
moderate recidivism risk mainly for drug/alcohol related offenses or domestic violence.
Category 4 - Socially marginalized – poor, uneducated, stressed, habitual offenders
The central problem in this type is socio-economic marginalization (e.g., educational-vocational
failure, poor job skills, poverty, unstable residence, poor social supports and social isolation).
This category is older (average age 37) and occurs in all ethnic groups. The social resources
for these men appear reasonable since they mostly do not have high crime families or antisocial peers, do not reside in high crime areas and do not hold extreme criminal attitudes – all
of which argue against a social learning explanation and do not suggest a high-risk lifestyle.
There is also little evidence of criminal personality.
Many of these cases are chronic repeaters with multiple arrests, probation terms and convictions. Their official criminal history coheres with the above profile in two main ways.
First, they are mostly late starters with a late age at first arrest (21), few juvenile felonies
and a relative absence of juvenile socialization problems. Second, their offense pattern of
fraud larceny (and some drug trafficking) and low robbery, suggests instrumental crime for
financial gain, or perhaps coping with poverty and unemployment. Finally, some of these
men exhibit prior domestic violence that coheres with prior weapons use and victim injury.
Substance abuse and criminal opportunity scores are about average.
Note: Mental health (MH) problems are often linked to social isolation and social adjustment problems. Thus, cases with MH and social withdrawal problems may enter this lonely
marginalized category. A mental health assessment is recommended to clarify MH issues.

➞2015 Northpointe Inc., All Rights Reserved.

 52

Category 5 - Criminally versatile – young marginalized persons often gang affiliated
This pattern exhibits multiple risk factors and several co-occurring causal processes linked to
criminality. First, is extreme social exclusion/marginalization (e.g., educational-vocational
failure, joblessness and poverty). Second is a lack of social control bonds, withdrawal from
education and work, boredom and little constructive use of leisure. Third, their high-risk
criminal opportunity lifestyle is reflected in weak pro-social bonds, boredom and higher than
average gang affiliation. Fourth, social learning is suggested by a pattern of anti-social attitudes, gang membership (for some), early school failure and out-of-home placements, all
implying affiliation with other rejected and weakly socialized peers. Finally, many of these
cases reflect an anti-social personality that has been empirically linked to family disintegration, family crime, juvenile felonies and early onset shown by many of these cases. These
themes reflect the sociopathic type of described by Lykken (1995) and Mealey (1995), and
others.
The criminal history of this category coheres with the above high risk profile. This young
group (22-23 average age) generally has an early age at first arrest (around 16), higher scores
than other types for juvenile felonies, weapons arrests, current violence, current property
and sex offense charges. However, there are two anomalies. First, they show relatively low
substance abuse. Second they score only average for prior arrests and convictions, perhaps
resulting from their youth (i.e., their early stage of a criminal career).
Category 6 - Socially isolated long term substance abuse – multiple minor and
mostly non-violent offenses
This group reflects four major criminogenic problems. First, many members exhibit serious
long-term substance abuse, suggesting addiction. Second, their extreme marginalization is
shown by social isolation, poverty, unstable residence, poor social adjustment, boredom and
a lack of pro-social leisure activities. Third, they appear embedded in a criminal drug culture
and exhibit high criminal opportunity. Finally, a disposition for criminality is shown by high
crime personality and antisocial attitudes. This type occurs in all ethnic groups.
The official criminal history matches this profile in several ways. Chronic criminality is
shown by multiple arrests, convictions and probations. Chronic substance abuse is confirmed by alcohol and drug offenses, using hard drugs (heroin, cocaine) as juveniles, being
high/intoxicated at current arrest and (in some cases) current drunk driving and/or drug
possession charges (but rarely trafficking). This category is difficult to treat as shown by
non-compliance, probation/parole revocations and FTA’s. They also exhibit above average scores for current fraud, prior domestic violence and burglary/larceny (but, rarely robbery). Criminal violence (except for domestic violence) is rare as shown by relatively low
arrests/convictions for weapons offenses and lower scores for assaultive felonies.

➞2015 Northpointe Inc., All Rights Reserved.

 53

Category 7 - Serious versatile high risk individuals
This type has the most serious and violent profile. It may warrant referral for a test such
as the Psychopathy Check List (PCL). This profile reflects a chronic, violent and versatile
criminal career as well as multiple criminogenic risk factors.
This profile reflects four major causal processes linked to high criminality. 1) A strong personal disposition to crime is shown by anti-social personality, antisocial attitudes/thinking,
early onset of crime, parental criminality and versatile criminal offences. 2) Social marginalization is shown by educational/vocational failure, unstable residence, poverty, boredom and
weak pro-social ties. 3) Social learning as reflected by anti-social peers, anti-social neighborhood, parent criminal behavior and anti-social thinking. 4) Poor socialization is suggested
by parental crime and family disorganization, early juvenile onset, early failure in school,
criminal attitudes.
The official criminal history matches this extreme criminogenic profile. It has the most
chronic and dangerous criminal career with the highest scores for criminal involvement,
juvenile onset, non-compliance and violent and versatile offending. These people have the
highest scores for arrests and convictions for robbery, burglary, weapon offenses, assaults,
injury to victims, violent felonies, fraud, drug possession and domestic violence arrests.
Category 8 - Low risk situational accidental category
Like Category 2, this category reflects lower criminogenic risks and more pro-social strengths
than most other categories. Thus, this profile offers no clear explanation for their engagement
in the criminal justice system. Like Category 2, these persons reflect perhaps “normal” folks
who became embroiled in a situational-accidental event that led to entry into the criminal
justice system. Many members of this category will have less poverty, more adequate jobs
and education, more stable residence in safer areas than most persons in this population.
They appear mostly to avoid anti-social peers and criminal opportunities and may have prosocial ties. Their attitudes and personalities are not clearly anti-social. They report low
drug use (compared to other groups), fewer criminal peers, lower family crime and positive
use of leisure.
The criminal history of this category confirms its low risk, non-violent status. Most have
few prior arrests and for many this may be their first incarceration. They generally have
fewer felonies or weapons offenses, and less history of probation or probation failure. Most
are assigned to the lowest risk category by the COMPAS risk models.
The current arrest pattern perhaps explains the situational nature of this category. Specifically, they have the lowest (mostly zero) scores for felony charges, assaultive felonies, weapons
offenses, victim injury, family violence, burglary/larceny, robbery and drug offenses. In many
cases their arrests are alcohol related, simple assault, drunk driving, non-felony fraud or minor property offense, or a sex offense. Thus, it is prudent to check the details (if available)
of the current offense of persons in this category.
An important caution is that a small percentage of this type may be “faking good” as indicated by the Lie Test score. Thus, while many are truly low risk (as confirmed by official

➞2015 Northpointe Inc., All Rights Reserved.

 54

history) a small percentage may be lying. Thus, it is still prudent to show caution with these
persons.

➞2015 Northpointe Inc., All Rights Reserved.

 55

5.3

Female Typology

5.3.1

Type Descriptions

Category 1 - Drug problems and anti-social sub-cultural influences – some with
relationship conflicts
This group (average age 35) appears locked into a high-risk sub-culture e.g. anti-social peers,
anti-social family and residence in a high-risk crime environment. Some reflect early onset
of teenage delinquency and cocaine use as a juvenile. Chronic drug problems are suggested
by above average scores for previous drug treatments and drug possession charges. Many of
these women hold anti-social attitudes. This profile suggests a social learning process where
these women are socialized within an anti-social drug sub-culture. However, some strengths
are still present for some of these women, e.g., stable housing, adequate use of leisure time
and apparently good social support. The group criminal history is about average and not
noticeably violent – although the group is above average for jail and probation terms, prior
convictions and non-compliance history. For some of these women their current domestic
violence charges suggest relationship conflict.
Category 2 - Family disorganization and inadequate parenting – residential instability and minor non-violent offences
This younger group has an average age of 25 years. Early family disorganization, abuse and
inadequate parents appear central. Their high scores for family criminality and juvenile outof-home placements suggests inadequate parenting. Their high juvenile socialization score
also suggests early onset of problems. Their adult life challenges include residential instability and social adjustment problems. However, several positive features emerge for some
of these women, i.e., lower than average scores for criminal peers, below average scores for
criminal attitudes and criminal personality. Many of these women appear to avoid drugs,
with relatively few reporting drug treatment or use of drugs as juveniles. The profile suggests some positive social supports and fairly constructive use of leisure time. The criminal
history is consistent with the above profile and is mostly non-violent and fairly low for noncompliance. The most common current charge is minor fraud. Mental health issues may be
explored given the possibility of early family abuse and/or neglect.
Category 3 - Chronic substance abusers – women with higher social resources
than other groups
This older (average age 38) category shows less poverty, more positive education and vocational skills and residence in an apparently safe low crime areas, than other categories. These
positive features are consistent with lower than average scores for criminal associates, lower
anti-social attitudes and a fairly positive use of leisure time. The group appears to have
relatively fewer social adjustment problems, better social supports and a lifestyle that avoids
high risks and criminal opportunity. They do not have high scores for criminal personality.

➞2015 Northpointe Inc., All Rights Reserved.

 56

The official data matches this profile with a relatively late onset, mostly minor offenses and
few juvenile problems.
DUI is the most frequent current offence among these mostly non-violent women – although
some also have domestic a violence record. However, the presence of prior convictions, prior
drug offences and frequency of prior treatments for drugs and/or alcohol underscores a clearly
chronic substance abuse problem.
Category 4 - Marginalized poor and isolated older women – economic survival
crimes
The average age of this group is 40 years. This group is characterized by poverty, social
isolation and a lower than average constructive leisure activities. This group has a late
onset with an average age at first arrest of 27 years. Their criminal history mainly involves
minor fraud. Aside from poverty they show few other criminogenic factors. For example,
they fall below average for criminal peers, antisocial attitudes, living in high crime areas or
following high opportunity lifestyles. Their family of origin appears relatively law abiding.
Their history exhibits few juvenile problems. It appears that their problems mostly emerge
in adulthood from poverty and poor social support. Their instrumental crimes such as minor
fraud and sex offences may be for economic survival. Their poor social adjustment and social
isolation suggest screening for mental health problems. The risk assessment assigns most of
these women to a low risk non-violent category.
Category 5 - Young antisocial poorly educated women with some violent offences
and early delinquency onset
This younger category (average age is 25) has a limited adult criminal history - with relatively
few adult arrests or convictions - but the highest score for a current violent offence, some involving felony and weapons charges. Their criminogenic factors include: early onset of delinquency, above average antisocial personality, antisocial attitudes, poor education/vocational
resources, bored/unproductive use of leisure hours and pessimism about finding a good job.
Early delinquency is reflected in higher than average juvenile marijuana and alcohol use (but
fewer hard drugs), high school dropout and the earliest first arrest. Surprisingly, the group
has relatively low affiliation with antisocial peers or gangs; no clear tendency to live in high
crime areas, abuse drugs, or to have extreme poverty or a high crime family background.
Their relatively low formal adult criminal histories, appear consistent with their average
scores on COMPAS risk assessment scales. However, the presence of early onset delinquency
and, in some cases, serious current violence suggests caution with this group.
Category 6 - Chronic long term criminal history A – multiple co-occurring social
and psychological risk factors
Drugs, extreme socio-economic marginalization, teen onset of problems and extreme problems in social relations characterize this high risk category. The recidivism risk computation

➞2015 Northpointe Inc., All Rights Reserved.

 57

identifies this group as high risk. Multiple criminogenic factors co-occur, including: antisocial peers, antisocial attitudes, antisocial personality, extreme substance abuse, high crime
family, poverty, extreme vocational and educational deficits, inability to use leisure time
constructively and a tendency to live a high risk life style. Problems started early and
these women report the highest levels for out-of-home placements as juveniles, the worst
school grades, the highest use of cocaine as a juvenile, the earliest first arrest and the highest number of juvenile felony arrests. This is a non-compliant group with multiple failures
and extreme drug problems. Social isolation and social adjustment problems are high. This
group commits a variety of offences, including: domestic violence, drug possession, and other
assault.
Category 7 - Chronic long term criminal history B – multiple co-occurring problems and high risk
This rare and infrequent group is a more serious version of type 6. While both categories have
multiple co-occurring risk and need factors group 7 is systematically higher than group 6.
This category has the highest scores for: violence risk, recidivism risk, FTA risk and technical
violation risk. They are highest for: overall criminal history, history of non-compliance, current violence and juvenile delinquency indicators. The multiple criminogenic factors include:
residential instability, family crime, vocational-educational failures, antisocial attitudes, antisocial personality, social adjustment problems, social isolation/withdrawal, extreme drug
use and so on. Compared to Category 6, this group has the highest scores for current violence, injuries to victims, current felony arrests and current robberies. They exhibit extreme
poverty, live in higher risk areas and report more gang affiliations.
Category 8 - Late starters with multiple strengths and fewer risk factors – minor
non-violent offence history
These women (like pattern 3) reflect higher resources than other groups for educational and
vocational scores, jobs, completing high school, living in safer areas, stable housing, better
social supports and fewer leisure problems. Their family background appears more prosocial and they report less poverty, antisocial attitudes or personality issues. This group
appears to adopt safer lifestyles by avoiding anti-social persons, fewer drug problems and
more pro-social leisure activities. While, we may be suspicious of this positive profile, their
official criminal history is consistent with this low risk profile showing the lowest criminal
involvement and incarcerations, the fewest arrests and convictions, the lowest arrest rate,
the lowest felony charges, the lowest pending charges, less non-compliance and the oldest
age at first arrest (average age is 27). Current charges reflect minor fraud and DUI. This
official data therefore coheres with this low need/risk profile. However, some women in this
category may be “faking good.” This was detected using the built-in COMPAS validity test
for defensive faking-good responses and notice should be taken of this warning.

➞2015 Northpointe Inc., All Rights Reserved.

 Bibliography
Andrews, D. A., & Bonta, J. (1994). The psychology of criminal conduct. Cincinnati:
Anderson Press.
Andrews, D. A., Bonta, J., & Wormith, J. S. (2006). The recent past and near future of risk
and/or needs assessment. Crime and Delinquency, 52(1), 7-27.
Andrews, D. A., Zinger, I., Hoge, R. D., Bonta, J., Gendreau, P., & Cullen, F. T. (1990).
Does correctional treatment work? A clinically relevant and psychologically informed
meta-analysis. Criminology, 28(3), 369-403.
Baird, C. (2009). A question of evidence: A critique of risk assessment models used in
the justice system. (Tech. Rep.). Madison, WI: National Council on Crime and Delinquency.
Baird, S., Heinz, R., & Bemus, B. (1979). Project report 14: Two-year follow-up (Tech.
Rep.). Wisconsin Department of Health and Social Services, Case Classification/Staff
Deployment Project. Bureau of Community Corrections.
Bandura, A., Barbaranelli, C., Caprara, G. V., & Pastorelli, C. (1996). Mechanisms of
moral disengagement in the exercise of moral agency. Journal of Personality and Social
Psychology, 71(2), 364-374.
Barbaree, H. E., Seto, M., Langton, C. M., & Peacock, E. J. (2001). Evaluating the predictive
accuracy of six risk assessment instruments for adult sex offenders. Criminal Justice
and Behavior, 28(4), 490-521.
Barnoski, R., & Aos, S. (2003). Washington’s offender accountability act: An analysis of the
Department of Corrections’ risk assessment (document no. 03-12-1202) (Tech. Rep.).
Olympia, WA: Washington State Institute for Public Policy.
Barnoski, R., & Drake, E. K. (2007). Washington’s Offender Accountability Act: Department
of Corrections’ static risk assessment (Tech. Rep.). Olympia, WA: Washington State
Institute for Public Policy.
Bonta, J. (1996). Risk-needs assessment and treatment. Thousand Oaks, CA: Sage.
Bonta, J., Law, M., & Hanson, K. (1998). The prediction of criminal and violent recidivism
among mentally disordered offenders: A meta-analysis. Psychological Bulletin, 123(2),
123–142.
Brennan, T., & Dieterich, W. (2008). Michigan Department of Corrections Core COMPAS
pilot study: One-year follow-up (Tech. Rep.). Traverse City, MI: Northpointe Institute
for Public Management.
Brennan, T., & Dieterich, W. (2009). Testing the predictive validity of the DPCA COMPAS
risk scales: Phase I (Tech. Rep.). Traverse City, MI: Northpointe Institute for Public
Management.
58

 59

Brennan, T., Dieterich, W., & Ehret, B. (2009). Evaluating the predictive validity of the
COMPAS risk and needs assessment system. Criminal Justice and Behavior, 36, 21-40.
Bucklen, K. B., & Zajac, G. (2009). But some of them don’t come back (to prison!):
Resource deprivation and thinking errors as determinants of parole success and failure.
The Prison Journal, 89, 239-264.
Bureau of Justice Statistics. (2012). Correctional populations in the United States, 2011.
Washington, D.C.: U.S. Department of Justice, Office of Justice Programs.
Chaiken, J., Chaiken, M., & Rhodes, W. (1994). Predicting violent behavior and classifying
violent offenders. In A. Reiss & J. Roth (Eds.), Understanding and preventing violence,
vol. 4: Consequences and control (pp. 217–295). Washington, D.C: National Academy
Press.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the
behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates.
Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine activity
approach. American Sociological Review, 44(4), 588-608.
Coleman, J. S. (1990). Foundations of social theory. Cambridge,Mass: Harvard University
Press.
Cooke, D. J., Forth, A. E., & Hare, R. D. (Eds.). (1998). Psychopathy: Theory, research
and implications for society. Dordecht, Netherlands: Kluwer Academic Publishers.
Copas, J., & Marshall, P. (1998). The offender group reconviction scale: A statistical
reconviction score for use by probation officers. Applied Statistics, 47, 159 - 171.
Dahle, K. P. (2006). Strengths and limitations of actuarial prediction of criminal re-offence
in a German prison sample: A comparative study of LSI-R, HCR-20 and PCL-R.
International Journal of Law and Psychiatry, 29(5), 431-442.
Dieterich, W. (2010). Kent County Pretrial Services outcomes study: Developing and testing
the COMPAS pretrial release risk scale. (Tech. Rep.). Traverse City, MI: Northpointe.
Dieterich, W., Brennan, T., & Oliver, W. (2011). Predictive validity of the COMPAS Core
risk scales: A probation outcomes study conducted for the Michigan Department of
Corrections (Tech. Rep.). Northpointe Inc.
Dieterich, W., Oliver, W., & Brennan, T. (2011). Predictive validity of the Reentry COMPAS
Risk scales: An outcomes study with extended follow-up conducted for the Michigan
Department of Corrections. (Tech. Rep.). Northpointe Inc.
Division of Probation and Correctional Alternatives. (2007). New York State pretrial release
services standards.
Dodge, K. A. (1998). The structure and function of reactive and proactive aggression.
In D. J. Pepler & K. H. Rubin (Eds.), The development and treatment of childhood
aggression (pp. 201–218). Hillsdale, N.J.: Lawrence Erlbaum.
Dodge, K. A., Pettit, G. S., McClaskey, C. L., & Brown, M. M. (1986). Social competence in
children. Monograph of the Society for Research in Child Development, 51(2), 1–85.
Elliot, D. S., Huizinga, D., & Ageton, S. (1985). Explaining delinquency and drug use.
Beverly Hills, CA: Sage Publications.
Estroff, S. E., & Zimmer, C. (1994). Social networks, social support, and violence among
persons with severe, persistent mental illness. In J. Monahan & H. J. Steadman (Eds.),
Violence and mental disorder: Developments in risk assessment (pp. 259–295). Chicago:
University of Chicago Press.

➞2015 Northpointe Inc., All Rights Reserved.

 60

Farabee, D., Zhang, S., Roberts, R. E., & Yang, J. (2010). COMPAS validation study: Final
report (Tech. Rep.). UCLA Integrated Substance Abuse Programs.
Farrington, D. P. (1991). Childhood aggression and adult violence: Early precursors and
later life outcomes. In D. J. Pepler & K. H. Rubin (Eds.), The development and
treatment of childhood aggression (pp. 5–29). Hillsdale, NJ: Lawrence Erlbaum.
Farrington, D. P., Jolliffe, D., Loeber, R., Stouthamer-Loeber, M., & Kalb, L. M. (2001).
The concentration of offenders in families, and family criminality in the prediction of
boys’ delinquency. Journal of Adolescence, 24, 579–596.
Fass, T., Heilbrun, K., DeMatteo, D., & Fretz, R. (2008). The LSI-R and the COMPAS.
Criminal Justice and Behavior, 35, 1095–1108.
Fischer, C. (1995). The subcultural theory of urbanism: A twentieth year assessment.
American Journal of Sociology, 101, 543 - 577.
Flores, A. W., Lowenkamp, C. T., Smith, P., & Latessa, E. J. (2006). Validating the Level
of Service Inventory-Revised on a sample of federal probationers. Federal Probation,
70(2), 44-78.
Gendreau, P., Goggin, C., & Little, T. (1996). A meta-analysis of the predictors of adult
offender recidivism: What works! Criminology, 34, 575-607.
Gendreau, P., Goggin, C. E., & Law, M. A. (1997). Predicting prison misconducts. Criminal
Justice and Behavior, 24(4), p414 - 431.
Gottfredson, M. R., & Hirschi, T. (1990). A general theory of crime. Stanford: CA: Stanford
University Press.
Gottfredson, S. D., & Moriarty, L. J. (2006). Statistical risk assessment: Old problems and
new applications. Crime & Delinquency, 52, 178-200.
Grann, M., Belfrage, H., & Tengstrom, A. (2000). Actuarial assessment of risk for violence:
Predictive validity of the VRAG and the historical part of the HCR-20. Criminal
Justice and Behavior, 27, 97-114.
Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C. (2000). Clinical versus
mechanical prediction: A meta-analysis. Psychological Assessment, 12(1), 19–30.
Hagan, J. (1998). Life course capitalization and adolescent development. In R. Jessor (Ed.),
New perspectives on adolescent risk behavior (pp. 499–517). New York: Cambridge
University Press.
Hanson, R. K. (1997). The development of a brief actuarial risk scale for sexual offense
recidivism. (User Report 97-04). Ottawa, Ontario: Department of the Solicitor General
of Canada.
Hanson, R. K., & Thornton, D. (2000). Improving risk assessments for sex offenders: A
comparison of three actuarial scales. Law and Human Behavior, 24, 119-136.
Hare, R. D. (1991). Manual for the Hare Psychopathy Checklist-Revised. Toronto, Canada:
Multi-Health Systems.
Harrell, F. E., Jr. (2001). Regression modeling strategies. New York: Springer.
Harris, P., & Jones, P. R. (1999). Differentiating delinquent youths for program planning
and evaluation. Criminal Justice & Behavior, 26(4), 403-434.
Hastie, T., Tibshirani, R., & Friedman, J. (2008). The elements of statistical learning: Data
mining, inference, and prediction (2nd ed.). New York: Springer.
Herrenkohl, T. I., Maguin, E., Hill, K. G., Hawkins, J. D., Abbott, R. D., & Catalano, R. F.
(2000). Developmental risk factors for youth violence. Journal of Adolescent Health,

➞2015 Northpointe Inc., All Rights Reserved.

 61

26, 176–186.
Hirschi, T. (1969). Causes of delinquency. Berkeley, CA: University of California Press.
Kroner, C., Stadtland, C., Eidt, M., & Nedopil, N. (2007). The validity of the violence risk
appraisal guide (VRAG) in predicting criminal recidivism. Criminal Behaviour and
Mental Health, 17(2), 89 - 100.
Kuhn, M., & Johnson, K. (2013). Applied predictive modeling. New York: Springer.
Lansing, S. (2012). New York State COMPAS-Probation Risk and Needs Assessment Study:
Evaluating predictive accuracy (Tech. Rep.). Albany, New York: New York State
Division of Criminal Justice Services, Office of Justice Research and Performance.
Lykken, D. (1995). The antisocial personalities. Hillsdale, N.J.: Lawrence Erlbaum.
McGrath, R. J., & Hoke, S. E. (2001). Vermont Assessment of Sex Offender Risk. Waterbury,
VT: Author
McGrath, R. J., Lasher, M. P., Cumming, G. F., Langton, C. M., & Hoke, S. E. (2014).
Development of the Vermont Asessment of Sex Offender Risk-2 (VASOR-2) Reoffense
Risk Scale. Sexual Abuse: A Journal of Research and Treatment, 26, 271-290.
Mealey, L. (1995). The sociobiology of sociopathy: An integrated evolutionary model.
Behavioral and Brain Sciences, 18(3), 523-599.
Megargee, E., & Bohn, M. (1979). Classifying criminal offenders: A new system based on
the MMPI. Beverly Hills, CA: Sage.
Moffitt, T. E. (1993). Adolescence-limited and life-course persistent antisocial behavior: A
developmental taxonomy. Psychological Review, 100(4), 674–701.
Monahan, J., & Skeem, J. (2014). Risk redux: The resurgence of risk assessment in criminal
sanctioning. Federal Sentencing Reporter, 26, 158-166.
National Association of Pretrial Services Agencies. (2004). National Association of Pretrial
Services Agencies standards on pretrial release (3rd edition) (Tech. Rep.). National
Association of Pretrial Services Agencies.
Nelson, M., Deess, P., & Allen, C. (1999). The first month out: Post-incarceration experiences in New York City [Unpublished monograph]. New York.
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3rd ed.). New York:
McGraw-Hill.
Osgood, D. W., Wilson, J. K., O’Malley, P. M., Bachman, J. G., & Johnston, L. D. (1996).
Routine activities and individual deviant behavior. American Sociological Review,
61(4), 635-655.
Palmer, T. (1994). A profile of correctional effectiveness and new directions for research.
Albany, NY: SUNY Press.
Parker, J. G., & Asher, S. R. (1987). Peer relations and later personal adjustment: Are
low-accepted children at risk. Psychological Bulletin, 102(3), 357–389.
Petersilia, J. (2003). When prisoners come home: Parole and prisoner reentry. New York:
Oxford University Press.
Quinsey, V. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (1998). Violent offenders:
Appraising and managing risk. Washington, DC: American Psychological Association.
Sampson, R., & Lauritsen, J. (1994). Deviant lifestyles proximity to crime and the offendervictim link in personal violence. Journal of Research in Crime and Delinquency, 27,
7–40.

➞2015 Northpointe Inc., All Rights Reserved.

 62

Solomon, A., Visher, C., La Vigne, N., & Osborne, J. (2006). Understanding the challenges
of prisoner reentry: Research findings from the Urban Institute’s prisoner reentry portfolio. Washington, DC: Urban Institute Press.
Stansfield, R., & Williams, K. R. (2014). Predicting family violence recidivism using the
DVSI-R. Criminal Justice and Behavior, 41, 163-180.
Stouthamer-Loeber, M., Loeber, R., Wei, E., Farrington, D. P., & Wikstrom, P. H. (2002).
Risk and promotive effects in the explanation of persistent serious delinquency in boys.
Journal of Consulting and Clinical Psychology, 70(1), 111–123.
Swets, J. A., Dawes, R. M., & Monahan, J. (2000). Psychological science can improve
diagnostic decisions. Psychological Science in the Public Interest, 1, 1–26.
Thornberry, T. P., Huizinga, D., & Loeber, R. (1995). The prevention of serious delinquency
and violence: Implications from the program of research on the causes and correlates
of delinquency. In J. C. Howell, B. Krisberg, J. D. Hawkins, & J. J. Wilson (Eds.),
Sourcebook on serious, violent and chronic juvenile offenders (pp. 213–237). Thousand
Oaks, CA: Sage.
Tolan, P. H., & Gorman-Smith, D. (1998). Development of serious and violent offending
careers. In R. Loeber & D. Farrington (Eds.), Serious and violent juvenile offenders:
Risk factors and successful interventions (pp. 68–85). Thousand Oaks, CA: Sage.
Travis, J. (2005). But they all come back: Facing the challenges of prisoner reentry. Washington, DC: Urban Institute Press.
VanNostrand, M. (2003). Assessing risk among pretrial defendants in Virginia: The Virginia
pretrial risk assessment instrument (Tech. Rep.). Virginia Department of Criminal
Justice Services.
VanNostrand, M. (2007). Legal and evidence-based practices: Application of legal principles,
laws, and research to the field of pretrial services (Tech. Rep.). National Institute of
Corrections and Crime and Justice Institute.
Van Voorhis, P., Wright, E., Salisbury, E., & Bauman, A. (2010). Women’s risk factors
and their contributions to existing risk/needs assessment: The current status of a
gender-responsive supplement. Criminal Justice And Behavior, 37, 261-288.
Warren, M. Q. (1971). Classification of offenders as an aid to efficient management and
effective treatment. Journal of Criminal Law, Criminology, and Police Science, 62,
239–258.
West, D. J. (1973). Who becomes delinquent? London: Heinemann.
Williams, K. R., & Grant, S. R. (2006). Empirically examining the risk of intimate partner
violence: The Revised Domestic Violence Screening Instrument (DVSI-R). Public
Health Reports, 121, 400-408.

➞2015 Northpointe Inc., All Rights Reserved.