Shannan.
launwxrlwudm.
rfi.:n...... X
Al no

.. é: .
‘aafiivig . .. ..
e

.r
3:.-

.ihfln .:
$11.3 1

at...
as...
A, I.

.3 £3.27:
:tiixaii E...
trtivi!

S}...

. .. Eu“!
ifluufiraifti
nuts: .5

 

 

 

 

IF. 7n:

Zoo 3

LIBRARY
Michigan State
University

This is to certify that the
dissertation entitled

Risk Assessment: Toward Valid and Accurate Delinquency
Predictions with African-Americans and Girls

presented by

Eyitayo Onifade

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in , Psychology

 

 

52 @«M

Major Professor’ 5 Signature
(jg/£14429?

Date

MSU is an affinnative-action, equal—opportunity employer

A- - .I-o-o-----u--o-I-n-o--o-n-I-I-o----u---o-----n-o--nut-n-s—

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K,IProi/Acc8Pres/ClRC/DateDue indd

 

RISK ASSESSMENT: TOWARD VALID AND ACCURATE DELINQUENCY
PREDICTIONS WITH AFRICAN-AMERICANS AND GIRLS

BY

Eyitayo Onifade

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Psychology

2008

ABSTRACT

RISK ASSESSMENT: TOWARD VALID AND ACCURATE DELINQUENCY
PREDICTIONS WITH AFRICAN-AMERICANS AND GIRLS

BY
Eyitayo Onifade

Recidivists typically comprise less than ten percent
of the offender population, yet can commit the majority of
offenses in a given year. Re-offenders share
characteristics in common (criminogenic risk factors),
which can be used to identify young offenders that are most
likely to recidivate. These criminogenic risk factors are
modeled and combined into multi-domain measures that
juvenile justice systems use in a practice known as risk
assessment. While the YLS/CMI is generally considered a
good risk measure, little work has been done validating its
predictive validity with African Americans and girls.

This study explored the risk-recidivism relationship
between YLS/CMI score and re-offense rates over a 24 month
period for African Americans, Whites, boys, and girls. The
YLS/CMI demonstrated its ability to differentiate non-
recidivists and re-offenders moderately well for each of
the subgroups. Recidivists on average scored four points
higher on the YLS/CMI than non-recidivists. In turn, the

YLS/CMI had an AUC of .63, however, subgroup differences in

distribution of recidivism by risk level were found.
African American low and moderate risk boys had
substantially higher re-offense rates than their
counterparts, while White low and moderate risk girls had
substantially lower re-offense rates. This led to over and
under predictions of re-offense for the two respective
subgroups.

In analyzing the risk-recidivism relationship in this
manner, the study was narrow in scope in that it did not
account for risk factors excluded from the YLS/CMI on the
community level like police surveillance levels that may
moderate these relationships. Moreover, this study did not
seek to ascertain the causal pathways between delinquency
and the various risk factors. Further, socioeconomic status
was not controlled for in parceling out possible moderating
effects on the risk—recidivism relationship. Instead given
the current stage in research on the dimensional identity
of risk assessments like the YLS/CMI, the first priority
was determining whether offense rates differed within risk
levels by gender and racial group. Commentary could then be
made about whether the instrument over—predicted or under-
predicted future delinquency for the various groups. With
that information future research can now focus on improving

the dimensional identity of the YLS/CMI.

Copyright by
Eyitayo Onifade

2008

To Eniola for just being

ACKNOWLEDGEMENTS

First and foremost, I would like to acknowledge and
thank my beautiful wife, Tiffiani. We have been together on
this journey so long, it's difficult, sometimes, to
separate who I am from what she's encouraged me to become.
We play the what if game, and in all honesty, her beautiful
spirit inspires me to be better than I am. So without a
doubt this dissertation, let alone my Ph.D would not have
been possible without her to support and encourage me along
the way. I thank Joy for letting me be a better scholar
than husband at times. I thank Joy for listening to the
same speech over and over again about the just us system
and what we need to do about it. I thank Joy for being a
great wife and even better Joy. You have my love, heart and
soul plus.

Secondly, I would like to acknowledge and thank my
beautiful family, Eniola, Sade, Kike, Kemi, Ciru, and
Mommy. Believe it or not, our experiences together are the
backbone of the dream, I dream for our people.

Thirdly, I would like to acknowledge and thank my
brilliant father, who is my hero and the only human being I
have ever strived to be like. It’s my nature to toss
directions to the side and dive into putting the pieces

together based off a picture on the box. Fortunate for me

vi

the blueprint has never been too far from hand and customer
support is always available 24—7. Now that I am older and
slightly wiser, I find myself reaching for the directions
more and more and the product is finally manifest. Daddy,
thanks for being my teacher.

Finally, I would like to acknowledge and thank my
advisor, Dr. William Davidson. Four years ago, I walked
into his office and I immediately realized this was the
place for me, in spite of the weather and distance from
home. Bill gave me room to grow and showed me real power
and influence. He pushed me when I needed to be pushed and
kept me grounded, when I was dreaming. I learned so much
about research and research that makes a difference. But
more importantly he acknowledged the enormity of life and
allowed me to do the same. This made all the difference for
me as a father, husband, brother and son; affording me the
rare privilege of having a life beyond grad school. Thanks,

Bill for everything.

vfi

TABLE OF CONTENTS

 

 

 

 

 

 

 

 

 

 

 

LIST OF TABLES iv
INTRODUCTION 1
CHAPTER 1

The Criminal as Different: Crime Theory and the Impact of
Poor Application 10
CHAPTER 2

Study Rationale 27
CHAPTER 3

Risk Assessment 44
CHAPTER 4

Validity, Accuracy and Dimensional Identity of Risk
Assessments 57
CHAPTER 5

Rationale for Validating the YLS/CMI 79
CHAPTER 6

Methodology 83
CHAPTER 7

Results 91
CHAPTER 8

Discussion 105
CHAPTER 9

Conclusions. 115

LITERATURE CITED 119

 

viii

LIST OF TABLES

Table 1. Description of validation studies with risk
assessment instruments 6O

 

Table 2. Results of validation studies 61

 

Table 3. Dimensional identity of risk assessment
instruments 63

 

Table 4. Corrected Item-Total Correlations for the Intake

 

 

and Probation Samples 86
Table 5. Item Frequencies (% Yes) 91
Table 6. Subscale scores and ROC by race and gender ............. 96
Table 7. Descriptive Statistics 97

 

Table 8. Hierarchical Logistic Regression for YLS/CMI score
and New Offenses 100

 

ix

I NTRODUCT I ON

There exists a rarely spoken of nexus between
recidivism risk assessment and the reduction of minority
and marginal group contact with the juvenile justice
system. In 2007, the Office of Juvenile Justice and
Delinquency Prevention [OJJDP] began to include
standardized risk assessments in its database of best
practices for systems plagued by disparate incarceration
and arrest rates for girls and minorities (OJJDP, 2007).
That said, the extent to which different racial and gender
groups officially offend at disparate rates and severity is
reflected in annually reported justice statistics and is
hardly a matter of debate. For instance, African Americans
comprise 16% of the youth population in the United States,
yet comprise 29% of the delinquency caseload (Snyder &
Sickmund, 2006). Moreover, despite accounting fOr no more
than a third of the juvenile population, African Americans
and Hispanic youth comprise nearly two-thirds of the
juvenile detention population (Snyder & Sickmund, 2006).
With respect to gender, girls are entering the justice
system at an alarming rate for offenses traditionally

associated with their male counterparts (Chesney-Lind &

Sheldon, 2004). For instance between 1970 and 2004, there
was a 322% increase in violent crime for girls compared to
a 107% increase in violent crime for boys in that same
period (Chesney-Lind & Sheldon, 2004). While girls have
historically comprised the majority of status offenders
(e.g. truancy, running away, consumption of alcohol,
incorrigibility); between 1994 and 2004 girls experienced
an increase in arrest rates for Index Crimes like homicide,
robbery, aggravated assault, arson, weapon possession and
drug offending (Chesney-Lind & Sheldon, 2004). So while
girls commit fewer officially reported offenses than boys,
there is some concern with the extent to which the rate of
increase is outpacing that of their male counterparts.

The causes of these disparities in officially recorded
offense rates have been the subject of many empirical
studies and qualified conjecture by professionals in fields
as dissimilar as criminology and public health. However,
one perspective in particular, the conflict theory of law,
has managed to sustain itself over the past decades
(Jackson, 1989). This theory suggests power is unequally
distributed and crime control is often used to mitigate
threats to the power group's interests (Jackson, 1989).
Moreover, disparate arrest rates are a product of certain

racial and gender groups receiving inordinate attention

from the criminal justice system that occurs on multiple
levels of the justice process from differential street
level policing to differential sentencing (Liska, 1992).
This perspective may be reflected in the conflicting
delinquency rates provided by national self report surveys
versus official arrest report data. For example, two
national self report studies [Monitoring the Future Survey
and National Youth Survey] showed no significant increases
in juvenile violence between 1980 and 2000, despite the
aforementioned substantial fluctuations in arrest rates
across race and gender groups (Howell, 2003).

In a pivotal meta-analysis, Pope and Feyerham (1992)
found that most of the disparity in marginal group contact
with law enforcement was found at the arrest and intake
level of the justice process. Other studies attribute this
differential treatment to unfavorable attitudes and
perceptions that too many practitioners and professionals
have of members of certain social groups (Guevera, Herz, &
Spohn, 2006; Leiber & Fox, 2005; Schrantz & McElroy, 2000).
In essence, the system is subject to a Pygmalion effect, a
self-fulfilling prophecy where certain groups are expected
to commit crimes and therefore are targeted by law
enforcement while other groups are not subject to such

treatment. This extra attention results in a larger

proportion of undesirable social behavior of certain groups
being discovered and receiving a formal system response.
While the validity of this perspective is widely and
hotly debated, the presumed role human bias plays in the
disparity in minority and marginal group contact with the
justice system has resulted in far-reaching policy
implications. The Juvenile Justice and Delinquency
Prevention Act of 2002 now requires that states receiving
formula grants:
“address juvenile delinquency prevention efforts and
system improvement efforts designed to reduce, without
establishing or requiring numerical standards or
quotas, the disproportionate number“ of juvenile
members of minority groups who come into contact with
the juvenile justice system” (see section
223(a)(22)).”
In the midst of this civil debate and system change,
criminal justice experienced a concurrent but separate
evolution of the practice known as recidivism risk
assessment. A plethora of research has demonstrated that as
few as 6% of offenders can commit the majority of offenses
(Cottle, Lee, & Heilbrun, 2001; Wolfgang, Figlio, & Sellin,
1972; see also Schumacher & Kurz-Gwen, 2000). Consequently,

a cornerstone strategy of crime reduction is the

identification of individuals comprising that small group
of offenders (Bonta, 1996). In the past, this process of
risk classification was managed by a practitioner or
clinician who predicted whether a child was likely to re-
offend based on the practitioner’s professional opinion and
experience. Since then psychometricians and researchers
have demonstrated that practitioners and clinicians often
ignore base-rates in addition to making spurious
correlations and incorrect causal attributions (Grove and
Meehl, 1996; Bonta, 1996). Formal actuarial assessments are
consequently considered more accurate and less susceptible
to bias than informal assessments (Grove and Meehl, 1996;
Bonta, 1996). As such, a myriad of risk prediction
instruments have become popular in the prediction of
delinquency and recidivism.

Formal risk assessment instruments are typically
comprised of risk factors (e.g. age of first offense) that
have been demonstrated to be associated with delinquency
and recidivism (Bonta, 1996). The cumulative effect of
multiple risk factors present in a youth’s life is then
used to predict the likelihood of that child experiencing a
particular undesirable outcome like further delinquency or
recidivism. Moreover, modern instruments like the Youth

Level of Service/Case Management Inventory utilize both

static and dynamic risk factors. That is to say these
instruments assess risk factors that are either malleable
or unresponsive to change efforts. For example, age of
first offense is a static risk factor while education is
dynamic because it can change. This practice of risk
assessment is evolving in the direction of using
combinations of risk factors to predict recidivism with
recent research demonstrating risk patterns are just as
important as total risk (Onifade et al., 2006). The early
verdict of most researchers has largely been in favor of
risk assessments, showing that the most recent generation
of risk assessments are both valid and reliable predictors
of delinquency and recidivism (Schwalbe, 2007).

Therein lays the nexus between risk assessment and the
reduction of disproportionate minority and marginal group
contact with the juvenile justice system. Theoretically, in
so much as risk assessments are valid predictors of
recidivism with diverse populations, risk assessments by
their very nature could reduce the system's contribution to
disparities in adjudication rates through squeezing out the
biases found at the early stages of the justice process.
However, little study has been conducted on the validity of
risk assessments with girls, and African Americans.

Therefore this posited curative relationship between risk

assessment and the reduction of disparities is presumptive
at best. Furthermore, these instruments are administered by
practitioners, begging the question how the presumed bias
of the practitioners will not be reflected in something as
basic as the scoring of the risk assessment. Moreover, it
is likely that the risk-recidivism statistical relationship
is naturally different for members of marginal groups. If
conflict theory is correct, members of these groups would
be at greater risk of having their criminal activity
exposed due to disparate levels of policing, surveillance,
and attention from the justice system in their communities.
As even multi-domain risk assessments do not include
policing levels, community surveillance, and justice system
attention as predictive risk factors, the instruments
should under-predict offending for these groups. This is
especially disconcerting when one considers there is
evidence that the very factors used in some risk
assessments have been shown to have little to no—
relationship with delinquency for girls and other
minorities. One should also keep in mind “arrest rates”
only reflect the rate of crime officially recorded by
police. Moreover, the differential validity of the
instrument may be reflective of the inherent bias in the

recorded and reported arrest rates, if juveniles with the

same risks have substantially and significantly different
arrests rates.

The danger in presuming a curative relationship
between risk assessment and justice system bias is
demonstrated in the case of the North Carolina Assessment
of Risk Instrument (NCAR). In a recent validation study of
the measure, researchers compared the validity coefficients
of African American (n=233) and White young offenders
(n=231) over a 12 month period (Schwalbe, Fraser, Day, &
Arnold, 2004). A Cox linear regression revealed that the
NCAR showed significant differences between the high,
moderate, and low risk offenders in their time to new
offense. Furthermore, high risk offenders were nearly twice
as likely to re—offend as moderate risk offenders. However,
when the hazard analysis was extended partitioning out
African Americans, Whites, boys and girls, the researchers
found significant differences in the risk/offending
patterns of the Subpopulations. Males reached the base—
rate of re-offense with lower scores on the NCAR than
girls. In turn, African Americans also reached the base-
rate of offense with lower scores on the NCAR than Whites.
There were also differences in the pattern of offense with
regards to the time it took for an offender of a particular

group with an NCAR score to re-offend.

The NCAR study offered an ideal model of the
validation process for risk assessment measures used in
juvenile justice systems. The focus, then, of this study
was to similarly examine the predictive validity of a risk
assessment used in the local courts, the Youth Level of
Service Case Management Inventory (YLS/CMI), across gender
groups and with Whites and African-Americans in a young
offender population. Clearly there is much work to be done
in the field of risk assessment, validating this measure
with a cross section of juvenile offenders. Thus, this
study answers this question of predictive accuracy with
regards to recidivism based on cumulative risk across
racial and gender groups.

In addition to providing a systematic synthesis of
juvenile risk assessment literature, the following
literature review offered a review of the historical
context of the criminal justice response to undesirable
social behavior, provided a review of evidence and theories
explaining disparate offending and incarceration rates, and
reviewed ecological theory’s role in improving the accuracy
of risk assessments and improving upon their dimensional

identity.

CHAPTER 1

The Criminal as Different: Crime Theory and the Impact of
Poor Application

If one thinks of crime as undesirable social behavior
so egregious that informal sanction, mores, and traditions
are incapable of providing a sufficient response, then the
criminal justice system is a formalized system of social
control by society (Cullen & Agnew, 2006). Society has
mechanisms in place that compel individuals to conform to
its expectations and norms through punishment, deterrence,
or rehabilitation and criminalization of those respective
behaviors (Kraska, 2006). Actors in systems of justice
response are faced with a tough question whose simple
wording belies our collective difficulty in responding to
it. Why do some people commit crimes, while others do not?

There are a number of theories offered in response to
this basic question. Biochemistry, Strain, Labeling,
Rational Choice, Social Control, Social Disorganization,
and Social Learning are all popular theoretical
perspectives explaining undesirable social behavior so
egregious that informal sanctions and traditions are
incapable of providing a sufficient response (Cullen &

Agnew, 2006).

10

Biogenetic theories offer perhaps the oldest
explanation of criminal behavior as exemplified by Cesare
Lombroso, who in as early as 1876 contended criminality was
an inheritable trait with identifiable physical
characteristics (Burke, 2001). Contemporary biogenetic
theories suggest heredity, temperament, traits and physical
maladies cause a predisposition to crime (Walsh, 2000).

Strain theorists following in the line of Durkheim,
Merton, Cohen, Agnew and Messner essentially state social
structures within society encourage or discourage citizens
to commit crime (Akers, 2000). Depending on the adequacy of
social structures in facilitating the goal achievement of
individuals, those individual’s perception of their means
and opportunities affect their willingness to adhere to
society rules in achieving those goals (Akers, 2000).

Labeling theorists hypothesize individuals are highly
susceptible to the stigmas society assigns to them;
consequently the individual has little choice but to
conform to the behavioral expectations of them whether
those behaviors are deviant or not (Braithwaite, 1989).

Rational Choice theorists offer that individuals are
rational and engage in cost/benefit analysis when
determining a best course of action; therefore, individuals

freely choose to act in their best interest according to

11

their calculations (Liska, 1987). Crime then is a result of
law's failure to act swiftly, with certainty, and severity
to deviant behavior, consequently criminal behavior is
individually calculated to be in the best interest of the
individual offender (Liska, 1987).

Social Control theorists believe that people
accumulate relationships, values, norms, and beliefs
through a socialization process which encourage or
discourage crime (Cullen & Agnew, 2006). The socialization
process occurs through direct control (punishments and
rewards), indirect control (identification with non-
criminals) and internal control (an active conscience).
Crime is therefore a product of degradation of social
controls (Cullen & Agnew, 2006).

Social Disorganization theorists attribute crime to
the absence or breakdown of community institutions and
communal relationships that typically facilitate
cooperation for mutual benefit among members of a community
(Jensen, 2003).

Social Learning theorists explain crime as a learned
behavior which occurs through the modeling of others’
behavior, the results of which are reinforced by certain

outcomes (Cullen & Agnew, 2006).

12

The one common denominator in most of these
perspectives is that those that commit crimes are different
in their person or context; in turn, the justice response
has always hinged on the identification of individuals with
those differences and addressing the criminogenic (crime
causing) factors that comprise those differences between
offenders and law-abiding citizens. In identifying
criminogenic differences, society has tried everything from
phrenology to profiling to standardized risk assessments.
Each of these presupposes criminals have identifiable
differences that can be relied upon to separate out
individuals that society can target with criminal justice
resources. The differences between these methods depend on
the theory from which they originate. With the wide cross-
section of theories explaining crime, the resulting system
is an amalgamation of crime reduction models with often
counterintuitive and conflicting means that either widen or
narrow the adjudication net depending on any number of
factors including but not limited to age, race, class,
gender, community, and policing philosophy (Pope &
Feyerham, 1992). Specifically, these disparate methods of
crime control occur in three ways:

> through emphasizing law enforcement in particular

communities

13

> the criminalization of undesirable behavior
associated with certain groups

> the decision making of criminal justice
practitioners at various junctions of the
adjudication process (Schrantz & McElroy, 2000).

Disparate.Methods of crime Control

First, with regards to over-emphasis of law
enforcement in certain communities, communities vary in the
degree to which they use the criminal justice system to
respond to undesirable and maladaptive behavior (Stucky,
2005). It is also widely observed that socioeconomic
conditions like adequate education and employment
opportunities, access to health care, and treatment
programs can reduce problem solving to the overuse of the
criminal justice system in certain communities (Loftin &
McDowall, 1982; Maguire, 2001; Stucky, 2005).

Indeed, studies of police strength and policing levels
have demonstrated that economic class and race are
significant factors in policing levels (Stucky, 2005). Both
economic and racial conflict theories presume resources are
finite and, consequently, power groups act in their best
interests, which is often detrimental to other groups. As
this relates to economic class or race, policing levels are

less related to crime levels than they are the perceived

l4

threat from the marginalized group (Stucky, 2005). Thus,
local government systems that are more susceptible to
political pressure from the power group are more likely to
increase policing resources in marginal communities
(Stucky, 2005; Wilson, 1968).

In support of this contention, several studies have
demonstrated a positive association between inequality and
policing levels (Sever, 2003; Jackson & Carroll, 1981;
Loftin & McDowell, 1982; Maguire, 2001). In a three-year
longitudinal study of 943 cities, Stucky (2005)
demonstrated that there was a significant positive
association between African-American presence and police
employment. Moreover, cities that had partisan elections
were more likely than their counterparts to have more
police employees per 1000 residents. Kent and Jacobs (2005)
found similar results, when examining policing levels in
large US cities between 1980 and 2000. Cities with
appointed managers as opposed to elected officials were
less likely to hire additional police support in spite of
minority presence.

The difference in communities’ willingness to use the
criminal justice system is also dependent upon the
availability of alternative systems that can appropriately

address the respective undesirable and maladaptive

15

behaviors of individuals in that community. For instance,
it is commonly observed that nearly 7% of inmates admitted
to jails annually have a mental illness (Freudenberg,
2002). Moreover, 75% of this number has a co-occurring
substance abuse disorder (Freudenberg, 2002). Consequently,
mental health professionals and other advocates often
recommend diversion to community-based services that treat
illnesses; however, in 1994 as few as 34% of jails
responding to a survey stated that they offered diversion
services for the mentally ill (Freudenberg, 2002).

Due to economic restraints, huge gaps in allocations
of mental health resources remain intractable for low
income communities, making criminal justice a primary
source for treatment of the homeless and indigent. This may
explain why some communities treat maladaptive behavior as
a treatable illness; while other communities are more
likely to use punitive sanctions available through law
I enforcement due primarily to disparate funding levels
between the criminal justice and mental health systems
(Weisman, Lamberti, & Price, 2004).

This is further demonstrated by the extent to which
juvenile offenders often have dual status as victims in the
child welfare system. Some communities use contact with

juvenile justice as a trigger mechanism for diversion into

16

the child welfare system that is better suited to deal with
family issues (Phillips & Bloom, 1998). However, this
practice varies by community and thus contributes to
disparities in officially reported offense rates for
various groups (Phillips & Bloom, 1998). Some communities
allow their schools to adopt zero— tolerance practices
which ultimately decrease the ratio of reported and
unreported delinquency (Mauer, 1999). Consequently, these
communities will have a higher rate of “crime” than their
counterparts whether the prevalence of the underlying
undesirable behavior is similar or not.Differences in the
degree to which communities rely on the juvenile justice
system to respond to delinquency affect the reported crime
rate for that community.

Second, communities can selectively enforce laws based
on group membership (Taxman, Byrne, & Pattavinia, 2005).
For instance, African Americans are more likely than Whites
to be arrested for drug offenses, yet Whites have higher
drug usage rates according to public health sources
(Freudenberg, 2002). Over the past two decades, sentencing
guidelines for drug involvement have dramatically changed.
The majority of changes are in regards to the use of
prisons to incarcerate and thus incapacitate offenders for

long periods of time. Possession of a controlled substance

17

typically resulted in a sentence of 18 months in 1990, yet
had increased to 30 months over the next ten years (BJS,
2004) .

Given the differences in arrest rate for drug offenses
in African American communities, the further.
criminalization of drug involvement disproportionately
affected this group (Taxman, Byrne, & Pattavinia, 2005). A
specific example commonly offered is that of laws targeting
groups associated with the trafficking of crack cocaine.
Crack cocaine use and sale is largely associated with
African Americans, while powder cocaine is associated with
other groups; however sentencing mandates require
significantly longer sentences for possessing smaller
amounts of crack than powder cocaine, in spite of their
comparable deleterious effects on the user (Mauer, 1999).
While the justification for this difference in sentencing
is that the trafficking of crack has far greater
deleterious effects on communities, the selective
criminalization in itself contributes to disparities in
system contact. This effect is further seen in the
treatment of female status offenders. In a review of self-
report studies conducted at the peak of arrest disparities
between girls and boys for status offenses; Steffensmeier

and Steffensmeier (1980) found that boys reported running

18

away just as much as girls, yet girls had a substantially
higher likelihood of being arrested for this behavior.

One also notes selective criminalization of marginal
groups can occur through targeting certain neighborhoods
and areas with justice resources and policies (Eck &
Weisburd, 2004). Curfews are often passed that make it a
violation for youth of a certain age to be out after a
certain time at night, yet those ordinances are differently
enforced at a neighborhood level.

In an effort to fight the proliferation of gangs in
certain communities, ordinances have been passed that
prohibit individuals that “look like” gang-members from
loitering or aggregating in a particular area for extended
periods of time. The Supreme Court struck down a city
ordinance that used just such language in prohibiting said
behavior, but recommended the law be re—drafted to specify
such loitering cannot be done with the purpose of
facilitating “criminal conduct.” With greater surveillance
of these communities, the contributions of criminals to the
larger cultural trends of the community they belong to can
also become a source of selective criminalization.

For instance, the City of Atlanta has passed an
ordinance prohibiting “sagging”, a fashion trend that

originated in prisons but matriculated into the larger

19

youth culture over time and is now commonly associated with
African American males. Opponents of the code contend this
law unfairly targets African American youth while allowing
other popular fashion trends such as low ride out jeans and
tank tops that are popular amongst other groups.

Finally, disparities in contact can occur through the
basic decision-making of justice officials. Prosecutors,
for instance, often have wide latitude as to plea
negotiations. These decisions can be swayed by prevailing
public opinions at the time of those decisions. Despite
sentencing guidelines and mandates, prosecutorial
discretion allows for significant departures from these
guidelines based on the decisions of the prosecution which
are not reviewable under the law (Kerstetter, 1990).
Prosecutors thus have discretion in three ways: the filing
of charges, the level of charges and the dismissal of
charges (Albonetti, 1987).

The US Sentencing Commission revealed that in 2002,
the average departure rate was 17%; however, districts
varied as much as 35% across the United States (Hartley,
Maddan, & Spohn, 2007). Moreover, this rate of departure
from sentencing guidelines varies by offense type with drug
offenses and property offenses differing in departure rates

by as much as 15% (Hartley, Maddan, & Spohn, 2007).

20

Numerous studies have indicated that while legal factors
such as strength of evidence (Albonetti, 1987) and
seriousness of the crime (Albonetti, 1987; Schmidt &
Steury, 1989) affect a prosecutor’s decision to pursue a
case, characteristics that are legally irrelevant also have
impact, such as racial composition of the suspect and
victim (Keil & Vito, 1989; Spohn & Spears, 1996), the
defendant’s gender (Nagel & Hagan, 1983; Spohn & Spears,
1996) or employment status (Schmidt & Steury, 1989).
Moreover, studies have shown that aforementioned crack-
powder cocaine sentencing disparity is exacerbated by
prosecutorial discretion where prosecutors are less likely
to depart from sentencing guidelines for African Americans
than Whites (Hartley, Maddan, & Spohn, 2007).

Furthermore, police have some degree of choice in
whether to warn and let go, or arrest and detain, with that
decision affected by many factors including their appraisal
of the offender. Moreover, a defendant's access to
resources also plays a part in whether alternative means of
treatment or problem-solving are used to deal with their
behavior.

Disparate System Contact Rates

One then quickly arrives at the sensitive subject

spoken to earlier, where individuals that share

21

characteristics in common, criminogenic or not, have
disparate amounts of contact with the justice system as a
result of the aforementioned crime control methods. For
instance, in the past decade, the number of girls referred
to the court for property and person offenses increased at
a greater rate than for boys. Indeed, young girls with
maltreatment and neglect histories were twice as likely as
their peers to be arrested in adulthood (Maxfield & Widom,
1996) with much of that disparity attributed to a failure
of service providers to adequately identify and respond to
the needs of this vulnerable population when the youth were
most responsive to care. While 9% of female status
offenders were placed in detention facilities in the mid
803, only 1.5% of their male counterparts received such
treatment (Schwartz, Steketee, & Schneider, 1989).
Moreover, young black offenders were 30 times more
likely than white offenders to be placed in detention
facilities for drug offenses despite sharing similar drug
usage rates (Pope and Feyerherm, 1992). In a widely cited
criminal justice study, Alfred Blumstein attributed 24% of
racial disparities in imprisonment to criminal histories
and racial bias. While African Americans constituted 15% of
drug users in 1998, African Americans comprised 37% of

those arrested for drug offenses. The numbers of

22

disparities are innumerable and as such warrant special
attention to any crime reduction methods that depend on
identifying differences.

Impact of Disparate Contact

The impact of disparate contact with the justice
system is far-reaching; affecting individuals, the
community and law—enforcement at large. For our justice
system to be viable, the public must have confidence in the
justice process. While it is convenient and even effective
to identify criminogenic differences amongst individuals,
this element of law enforcement should not devolve into
discrimination. Quite simply, disparate contact naturally
concentrates iatrogenic effects of the justice process in
communities that experience the disparity. Moreover, we see
iatrogenic effects across a broad spectrum of social
living, from educational achievement to community health to
political empowerment.

For instance, a growing body of research has
demonstrated the role that disparate contact plays in
destabilizing at-risk communities via the breakdown of
families (Courtwright, 1996; Freudenberg, 2002). Latina and
African American women are 3 and 7 times, respectively, as
likely as White women to be incarcerated in their lifetime.

Women that are incarcerated are often separated from their

23

children, and their children often end up in non-familial
foster care where the children are at higher risk for
psychological, educational and social problems (Phillips &
Bloom, 1998). Moreover, nearly 48% of women behind bars
report being sexually assaulted while incarcerated (Harlow,
1998). Thus, these women are disproportionately at risk for
sexually transmitted infections in addition to having
mental health issues due to violent sexual trauma
(Freudenberg, 2001). These personal and social problems are
compounded by the fact that nearly 40% of incarcerated
women find themselves homeless upon release (Richie, 2001).

One observes further iatrogenic effects in the cases
of education and family ecology. Congress passed
legislation in 1994 prohibiting prison inmates from
receiving Pell grants in pursuing higher education, thus
limiting prisoners’ access to professional skill sets that
ease re—entry to society (Fellner & Mauer, 1998). Thus, for
groups such as that of the African American community,
where males have a 29% chance of spending time in prison at
one point in their lives, disparate contact with the
justice system affects this group’s very ability to gain
upward social mobility through education.

The African American community also has a considerable

imbalance in the adult male-female ratio (Courtwright,

24

1996). While boys outnumber girls at birth, by the age of
40 this cohort has 86 males for every 100 females. With the
rate at which the remaining men are incarcerated or under
criminal justice supervision, the number of eligible males
for marriage is chronically low. Furthermore, felons are
largely disenfranchised; consequently a group like African
Americans has nearly 13% of its adult male population
ineligible to vote (Fellner & Mauer, 1998), weakening the
very ability of these communities to influence the policies
that set the parameters of their lives.

Communities experiencing disparate contact with the
criminal legal system have fewer informal mechanisms of
crime control. Joblessness, poverty, and a highly
transitory population are believed to be associated with
poor neighborhood cohesion. As a result, the ability of
authority figures to transmit values in tune with the
larger society expectations to children is highly
threatened. These communities develop a somewhat
antagonistic relationship with law enforcement, where
police are perceived as equally as threatening as
criminals. This leads to lower rates of participation in
crime reduction programs or even basic reporting of crimes

to the police.

25

The very basis of democracy is the presumption of
equality; therefore it is generally agreed that it is wrong
in principle to treat equals differently unless there is
some reasonable circumstance warranting such treatment (May
& Sharratt, 1994). In considering most crime theory, we see
that crime reduction methods based on identifying
differences saddles the line between violating our
principles and addressing a social problem in the most
practical way, despite iatrogenic effects. So in resorting
to reducing crime through the identification of
differences, we must carefully consider whether there is
some reasonable circumstance warranting such treatment.
Juvenile Justice as Differential Treatment

The very basis of the argument for the necessity of
juvenile justice is the presumption that youth require
differential treatment from their adult counterparts due to
differences in their etiological pathways to crime
(Shepard, 1998). Adolescence is second only to infancy with
regards to rate of change and growth, where identity
formation and social integration primarily occur (Erikson,
1963). Both of these developmental stages are highly
susceptible to the influence of peer, educational, familial

and social factors.

26

Adolescents typically are moving toward autonomy from
parental figures, whilst undergoing rapid neurobiological
changes that affect emotional and abstract reasoning.
Developmental neurobiologists have gone so far as to use
magnetic resonance imaging of the brain to show how youth
and adults differ in the way they think, with adolescents
using their limbic system whereas adults use their frontal
cortex in decision-making of an emotional nature (Siegel,
1999). The limbic system is responsible for instinctive
reactions whereas the frontal cortex is usually relied upon
for complex reasoning and thought.

Developmental psychologists observe that youth
involved with the justice system typically come from homes
and communities that are marked by harmful environmental
factors and thus they lack the consistent and reliable
care-giving necessary for their moral development (Bandura
& McDonald, 1963). Youth presumably are most given to
undesirable social behavior that violates codified laws
because of some lack of supervision or failure on the part
of the adult and communal actors in the child’s life;
therefore warranting rehabilitation of the youth as opposed
to retributive justice which is traditionally considered

the priority of adult systems.

27

One notes that without this perspective and in
treating juveniles like adults, society used methods that
would be considered senseless and draconian today. Prior to
the adoption of this perspective, at least ten children
under 14 years of age were executed for their crimes
(Streib, 1987). Juvenile courts were thus born out of the
“parens partriae” doctrine, which implied the courts have
an obligation to take charge of and rehabilitate delinquent
juveniles (Anderson, 1988; Scott & Grisso, 1997).

The Source of Disparity

Between system contributions to disparities and
intrinsic differences between groups like that of juveniles
and adults, there is considerable room for debate as to why
these differences exist. Moreover, in acknowledging
identifiable differences and acting to reduce crime through
special treatment of individuals possessing those
characteristics, society has impetus to consider the
advantages and disadvantages of just such a practice.

Higher education, for instance has experienced a
chronic under-representation of minorities and has long
since incorporated standardized tests into their
admissions, in an effort to “objectively” identify students
most likely to succeed in college. The median correlation

of the SAT I with GPA for Blacks reported in a 2001

28

comprehensive review was .33 for African Americans and was
.44 for Whites (Young, 2001). Moreover studies consistently
demonstrated that the SAT I under—predicts college GPA
performance for women as compared to men. As standardized
tests are so important in American society because of their
role in college admissions; essentially deciding who and
who does not get access to our educational resources the
bedrock of social mobility in this country, the persistence
of differential predictive validity in standardized testing
is problematic to say the least. As a result, the education
field is constantly in a state of debate and legal
challenge as to the merits of its classification and
prediction system (Zwick, 2004). The parallel example of
the SAT I that is taken by 40% of graduating US high school
students (Young, 2001) is quite salient in that it reflects
how measures can become so entrenched in practice, the
field is almost wedded to them. This is despite empirical
evidence pointing to fundamental flaws in their predictive
validity with girls and some minority groups (Young, 2001).
Attribution Theory (Heider, 1958) could explain the
affinity for most of the paradigmatic traditions cited in
these debates. Self-serving causal attributions or
fundamental attribution errors are at play casting the

individual or the environment wholly responsible for the

29

disparate outcome. Criminal justice theorists have long
since developed any number of theories to explain away
specific differences in differential offending/system
contact rates that focus either on the individual or the
environment. For instance, Strain Theory suggests
disadvantaged groups feel some conflict between aspirations
of success and their expectations of achievement given
societal and structural systems (Agnew, 1992). When one
turns her or his attention to biological deficit theories,
one finds they typically imply criminal behavior is by its
nature pathological. As a result, disparities in
incarceration rates reflect endemic maladaptive disorders
in those groups (Loeber & LeBlanc, 1990). Simply put, some
groups merely have a disproportionate number of individuals
that are genetically or otherwise biologically predisposed
to crime. On the external attribution side, social dis-
organization theories attribute delinquency to spatial and
ecological factors that encourage crime in some contexts
and reduce it in others (Sampson, Raudenbush & Earls,
1997).

While these debates are common and are certainly
conflict-laden, clearly some degree of support and evidence
exists for each one of these perspectives. Thus, the

greater axiomatic conclusion that we can draw from multiple

30

competing theories concurrently explaining a single social
condition is that the causal pathways are in fact quite
complex and interactive. The framework then that is best
suited to account for differential offending and, in turn,
our differential responses to undesirable social behavior
of juveniles must account for behavior in context and thus
must be comprehensive.

If we review the logic statement to this point we are
essentially saying the following: Crime warrants a formal
response from society. Criminals are different than law-
abiding citizens; ergo the justice system is heavily
dependent upon identifiable differences to prosecute
criminals for their deviant behavior. Moreover, the
accuracy of that system depends on the comprehensiveness of
the theoretical framework the system response is derived
from. Hence, as much of the theoretical basis for crime
reduction is based either on external or internal causal
attributions of crime, unjust disparities that disadvantage
certain social groups arise. If the juvenile justice system
incorporates risk assessment as a mechanism for objectively
and correctly identifying youth most likely to offend that
model’s success not only hinges on the presumption of
differences amongst individuals. The reliability or fallacy

of that presumption also rests on how comprehensive the

31

theoretical framework one works from in parceling out
relationships between crime’s cause and effect.
A comprehensive Theoretical Framework

An alternative and more comprehensive model to use in
validating and improving upon risk measures is
Bronfenbrenner’s Bioecological Systems theory (1979).
Bronfenbrenner (1979) offered that the development of a
child occurs within a larger relationship context between
its biology and the various dimensions of its environment.
Thus, as the child matures the interaction between the
child's own biological make-up, his immediate
family/communal environment and societal background
collectively impact the child’s development.

When the child experiences a negative outcome or
engages in undesirable social behavior like crime, the
causal source lays not only in the realm of their
environment or the individual (cognitive, genetic and
physical attributes), but rather the cumulative interaction
between the two, which are both multi-leveled.

Within the immediate environment of the child
(separate of their cognitive, genetic and physical
characteristics) rests the microsystem (Bronfenbrenner,
1979). This system includes the family, school,

neighborhood and childcare structures. As these factors are

32

closest to the child, they have the greatest impact on the
child’s well-being. It should be noted that the effect is
bidirectional; so for example, just as the child is
affected by their school institution, the school
institution is affected by the child. This creates multiple
feedback loops that reinforce and initiate behaviors
throughout the maturation process. Between each of these
structures comprising the microsystem lays a connecting
tissue referred to as the mesosystem (Bronfenbrenner,
1979). Just as the family is connected to child, the family
is also connected to the childcare andschool and
neighborhood.

Adjacent to the microsystem and mesosystem is the
exosystem which is comprised of the larger social systems
in which the child is not directly involved
(Bronfenbrenner, 1979). For instance, a child may not have
any interaction with a school board but their teacher does,
and whatever effect the school board has on their teacher
is augmented and filtered down to the child through the
child and teacher’s interaction in the classroom.

The outermost layer is referred to as the macrosystem
(Bronfenbrenner, 1979). The macrosystem is comprised of
values, customs and laws. These principles give shape to

the interactions at the other levels. For instance, a

33

society that uses work hours and the demand of the skill to
determine the amount of compensation for labor will
inevitably require that workers with skills in less demand
work long hours to support their families. This will affect
children by limiting the amount of time the parent spends
with the child.

This ecological framework has implications for
society’s response to undesirable social behavior. At the
macrosystem level our values and principles affect the
basic determination of which undesirable behaviors are so
egregious they warrant formal sanctions. They affect our
attitudes, which in turn affect our implementation and
consistency of application of these laws in the
microsystems of youths’ lives. These effects ripple into
the exosystem of principle actors of the juvenile’s life in
many ways. For instance, attitudes in a community that
reflect zero-tolerance for drug use exert considerable
pressure on law enforcement officers to satisfy their
expectation of having a drug-free community. This in turn
affects the officer’s relationship with the youth directly.
It may even affect the youth indirectly through the parent
or school via the mesosystem in a positive or negative
manner depending on the internal environment of the child

and their respective affinity for drug involvement.

34

When one traces these causal pathways from
criminogenic factors to delinquent acts to justice
response, the inordinate number of possible mediators and
moderators becomes obvious. Moreover, one realizes that an
interactive system such as this is primed for any number of
differential outcomes and differential treatments of
individuals. The system is then faced with a question of
how to accommodate the complexity of the system yet manage
the group with the single relevant outcome of offending. In
criminal justice, when powerful individuals and groups have
sole discretion in deciding what those relationships are
and in turn what those relationships mean with regards to
differences between groups and individuals in behavior, the
system becomes highly reflective of the beliefs and
prejudices most prevalent in that population. Therein lie
the cost of change models that are informal, of limited
causal attribution scope and unchecked for inconsistencies
in implementation.

From this premise one arrives at the notion of
creating additive ecological models that account for each
of these factors at the various levels and then determines
critical points. When certain thresholds are met the
likelihood of offending becomes so great, specific action

from the justice system addressing those needs is

35

warranted. This is the theoretical framework behind risk
assessment. The problem then arises when we consider that
typical risk assessment goes no further than assessing the
micro and mesosystem. Rarely does this expand to include
the macrosystem. Nor does the common practice of simply
adding the number of risks that exist in a youth’s life
account for the interaction between those risks and how
particular combinations of criminogenic factors either
mitigate or exacerbate delinquency. So in considering risk
assessment as a solution to disparate contact, we must also
recognize that this solvency may be limited by the
comprehensiveness of risk measures and the lack of
validation with diverse populations.

In the following sections, the rationale for using
risk assessments is reviewed. In addition, previous
research validating contemporary risk assessment with
regards to predictive accuracy and dimensional identity is
reviewed. The remainder of this review will demonstrate the
necessity of validating the YLS/CMI with a diverse
population, using both additive models and risk

combinations to predict the likelihood of offense.

36

Chapter 2

Study Rationale

Given that over 2 million young offenders are arrested
annually, costing $14.4 billion for law enforcement,
courts, detention, residential placement and block-grants
(National Center on Addiction and Substance Abuse, 2004),
there is considerable impetus to identify repeat offenders
early in the justice process. Recidivists are of particular
concern because as few as 6% of offenders can commit the
majority of offenses through recidivism (Snyder & Sickmund,
2006). Moreover, treating each juvenile as though they have
an equal likelihood of returning to the system without
intervention leads to wasted resources and inadequately
serviced youth in need (Howell, 2003). There is great need
for matching interventions with the need set of young
offenders, consequently, dependable tools for identifying
re-offenders and assessing areas of criminogenic risk are
highly desired by systems (Howell, 2003).

The process for sorting offenders into groups based on
criminogenic risk is vulnerable to enabling wide
disparities in system contact based on group membership
status for the previously mentioned reasons. Risk

assessments and decisions based on group membership status

37

can lead to over-predictions and under-predictions of
future criminal activity. For example, since the 19605 at
least two-thirds of studies have found some degree of
disproportionate racial minority contact with law
enforcement that was attributable to over-reliance on
subjective human judgments at the intake and arrest level,
according to a Pope and Feyerham’s meta-analysis (1992).
Furthermore, girls are increasingly becoming
vulnerable to a “net-widening” effect due to a difference
in their criminogenic developmental pathways (Chesney-Lind
& Sheldon, 2004). Girls are unique in that their first
contact with the system is often due to relatively minor
offenses, while their male counterparts exhibit a far
greater range of severity in offending (Chesney-Lind &
Sheldon, 2004). Programming often mixes girls regardless of
stage in criminogenic development. Net-widening for girls
is exacerbated by their status offenses being formally
prosecuted at a greater rate than their male counterparts
due to prevailing opinions about their risk for truancy,
incorrigibility and running—away. As such, this initial
contact invites greater attention from the legal system,
and this attention facilitates contact with higher risk
offenders in the system and essentially exacerbates their

criminogenic risk through peer group association.

38

There is also some evidence that commonly recognized
risk factors for boys do not have the same degree of
association with offending for girls, so girls that exhibit
similar risks as boys are not necessarily as likely to re-
offend. This can lead to over-programming of girls in the
justice system when they exhibit risks similar to that of
their male counterparts. Risk assessments can as result
over-predict delinquency for girls if not properly
validated with this group.

These factors coupled with the observation that minor
and low risk offenders are made worse by extended and
intense contact with the justice system have driven the
resurgence in popularity of actuarial risk assessment
measures (Andrews, Kiessling, Robinson, & Mickus, 1986).
Indeed, in the 2003 Comprehensive Framework for Juvenile
Justice, the US Office of Juvenile Justice and Delinquency
Prevention offered actuarial prediction instruments as
tools in creating a system of equitable and accurate
disposition decision-making (Howell, 2003) by standardizing
assessment practices and using prediction methods that are
similar across jurisdictions and populations (Grove &
Meehl, 1996).

As the use of risk assessments has become widespread,

the success of juvenile justice has become dependent upon

39

the validity, accuracy and dimensional identity of the
prediction instruments used in this cornerstone of the
justice process. Risk assessment as it is referred to here
is the prediction of offending and classification of
juveniles based on their likelihood to offend (Bonta, 1996;
see also Gottfredson & Moriarty, 2006).

The predictive validity of risk assessment measures
thus refers to the extent to which these instruments
predict delinquency, which is reflected in the significance
and strength of relationship between risk scores and
delinquency outcomes (Gottfredson & Moriarty, 2006). In
turn, accuracy is the rate at which risk assessments
correctly identify non-offenders and offenders. Dimensional
identity is said to exist when the same relationship
between delinquency and risk assessments is true for all
subpopulations upon which the measure has been administered
(Von Eye & Bergman, 2003).

In systems that use risk assessments in their
decision-making, the success of the system at reducing
delinquent offending is directly related to the performance
of their measures in the aforementioned regards to
validity, accuracy and dimensional identity. Wiebush and
researchers (1999) found that 39 states reported using risk

assessments and classification instruments with young

40

offenders. The use of actuarial risk assessments is thus
quite common across many jurisdictions. However, there has
been little attempt to systematically review the success of
these measures over the past decade in predicting
delinquency.

In so much as risk assessments are valid predictors of
recidivism with diverse populations, risk assessments by
their very nature could reduce the systems contribution to
disparities in adjudication rates through squeezing out the
biases found at the early stages of the justice process.
Theoretically these instruments should do a better job than
informal assessments at identifying youth likely to offend
repeatedly, regardless of race or gender, thus members of
marginal groups that are low risk would conceivably be
eligible for the same programming as their counterparts,
which in many cases includes being refracted from the
system. For these groups, this in turn results in
“narrowing of the net”, the reduction of disproportionate
marginal group contact with law enforcement.

However, little study has been conducted on the
validity of risk assessments with girls and African
Americans, so this posited curative relationship between
risk assessment and the reduction of disparities is

presumptive at best. Moreover, it is likely that the risk—

41

recidivism statistical relationship is naturally different
for members of marginal groups.

If conflict theory is correct, members of these groups
would be at greater risk of having their criminal activity
exposed due to disparate levels of policing, surveillance,
and attention from the justice system in their communities.
As even multi-domain risk assessments do not include
policing levels, community surveillance, and justice system
attention as predictive risk factors, the instruments
should under-predict offending for African Americans. This
is especially disconcerting when one considers there is
evidence that the very factors used in some risk
assessments have been shown to have little to no
relationship with delinquency for girls leading to over-
predictions of delinquency for this group and further
entrenching girls in the justice system when case-
management decisions are based on those assessments.

Consequently, there is need for validation of
instruments used by systems to identify youth most likely
to recidivate. This study was an attempt to validate the
YLS/CMI, which is used in the local jurisdiction with a
wide cross-section of young offenders including African-
Americans and girls. In choosing a predictive instrument, a

comparative review of the performances of various popular

42

risk assessment instruments was necessary. In the following
section, a synthesis of this literature is presented. The
purpose then of the following review is to summarize and
critique the extent to which delinquency risk measures have
been validated with attention to the usefulness of these
relationships with regards to their accuracy and
dimensional identity. Moreover, attention is given to the
specific performance of the YLS/CMI compared to other

instruments.

43

Chapter 3

Risk Assessment

Risk assessment rests on the premise that a small
number of offenders commit the majority of offenses through
repeat offending, which is well evidenced in the literature
(Wolfgang, Figlio, & Sellin, 1972; Moffit, 1993; see also
Snyder & Sickmund, 2006). For instance, Schumacher and
Kurz-Gwen (2000) demonstrated in a longitudinal study with
70,000 youth that as few as 8 percent of offenders commit
more than half of offenses in a given year. In order for
risk assessment to work, this small group of offenders must
share characteristics in common that differentiate them
from their fellow offenders.
Risk Factors

The literature is thus replete with research
indicating that characteristics of an individual’s life on
multiple levels mark those offenders most likely to commit
the bulk of offenses (Cottle, Lee, & Heilbrun, 2001).
Moffit (1993) articulated these differences as risk
factors, which set offenders on either of two delinquency
paths: adolescent limited and life course persistent
offending. In the case of the former, the majority of

offenders limited their delinquency to adolescence, while

44

for the latter; a minority of young offenders (6%)
continued to offend well into adulthood. Moffit noted that
life-course persistent offenders were characterized by
committing offenses at young ages and poor prenatal care.

Since then a number of risk factors like family
structure (Myner, et al., 1998), learning disabilities
(Duncan, et al., 1995), gender (Dembo, et al., 1998),
offense type (Archwamety & Katsiyannis, 1998) etc, have all
been associated with delinquency. In turn, a Cottles, Lee
and Heilbrun’s (2001) meta-analysis of 23 studies,
representing 15,265 youth revealed that age at first
offense, family problems, peer groups and family history of
criminality were the strongest single predictors of
recidivism.

It should be noted, however, that the use of single
risk factors to predict the outcomes of offenders is
replete with high false positive and false negative
classification rates thus making single dimensioned
assessments inadequate in the case management process
(Risler, Sutphen, & Shields, 2000). For example, we can
know in general that individuals of certain racial groups
or of certain gender or of a certain age raised in a
distressed community are more likely to commit offenses

than their peers without those risk factors. We also can

45

know individuals that do share those risks in common can
have very different outcomes, some desirable while others
are undesirable. Thus, in classifying youth solely on the
basis of these risks, the usefulness of our predictions
will be hampered by high false positive and false negative
rates.

Risk Factors for Girls

Moreover, using single risk factors is also
problematic because there is some indication that certain
risk factors are more salient with some groups than others.
Specifically with regards to gender, criminological theory
has largely focused on males and been generalized to
females (Belknap, 1996). Simound and Andrews’s (1994) meta-
analysis went as far to suggest that predictors for males
and females are largely the same, with criminal history,
antisocial behaviors, and delinquent peers each serving as
strong predictors of delinquency for both genders. Hubbard
and Pratt (2002), who also conducted a meta-analysis,
criticized Simound and Andrews’s conclusions for its lack
of focus upon the maladjustment of young female offenders.
Hubbard and Pratt (2002) asserted that there should have
been special emphasis on sexual abuse and maltreatment of
girls in childhood because of the disparate rate of such

abuse against females.

46

Other studies have suggested that mood disorders
(Mcreynolds, 2005), neighborhood disadvantage (Mcreynolds,
2005), and truancy (Hishinuma et. al., 2005) as risk
factors are moderated by gender. Hubbard and Pratt’s (2002)
meta—analysis included 97 effect size estimates
representing 5,981 cases, and revealed that while commonly
mentioned risk factors like peer deviancy, anti-social
behavior and prior history are strong predictors; childhood
maltreatment and school relationships are both strong
predictors of delinquency for females. Therefore in using
single factors that are highly correlated with delinquency
but not necessarily accurate in individually predicting
offense given race, gender and other moderating factors,
systems waste resources and perpetuate unjust biases
against entire classes of people.

Risk Factors for African Americans

While studies have indicated that African Americans
share many of the same risks for delinquencyas White youth
(i.e. family, substance abuse, education, personality,
offense history, etc.), researchers have noted the
relationship between those risks and delinquency are not
the same for the two groups (Myers, Taylor, Alvy,
Arrington, & Richardson, 1992). Certain criminogenic risks

are present to a greater degree in African American

47

communities than others and several factors moderate the
relationship between common risk factors and delinquency
for African Americans. For example, African Americans are
more likely to experience poverty, poor schooling, single-
parent households, and are more likely to witness violence
in their communities. Often the prevalence of these risk
factors can result in high levels of adaptability and
resiliency, confounding typical relationships between those
factors and offending (Kliewer & Kung, 1998). Studies of
coping strategies and delinquency have shown that emotion-
focused coping strategies are associated with negative
behavioral outcomes, whereas active—coping is associated
with positive behavioral outcomes (Compas et al., 1998).
These studies have also shown the difference in prevalence
of these strategies in the two groups is due to differences
in available resources.

Furthermore, some research has indicated that African
American youth are particularly vulnerable to certain risk
factors due to stress-induced factors related to their
status as African Americans (i.e., class oppression, low
self-efficacy, and barriers to goal achievement) (McGee,
2003). Moreover, some studies have indicated the single
factor of threat perception in response to everyday

situations has a substantial impact on delinquency risk

48

(Kliewer, Fearnow, & Walton, 1998). African American young
offenders in vulnerable communities often exhibit higher
levels of “threat awareness” resulting in a greater
likelihood of using coping strategies that may or may not
be socially desirable (Kliewer, Fearnow, & Walton, 1998).
Failure to account for factors related to resiliency can
lead to over-predicting offense rates in groups of African
Americans that have highly adaptive members. Moreover,
failure to account for structural factors and community
conditions that may exacerbate delinquency risk can lead to
under-predictions of risk.
Ecological Perspective and Risk Assessment

The ecological perspective may be more desirable as a
basis for developing risk assessment measures. As
articulated by Rappaport, the “ecological viewpoint should
be regarded as an orientation emphasizing relationships
among persons and their social and physical environment.
Conceptually the term implies that there are neither
inadequate persons nor inadequate environments, but rather
that the fit between persons and environments may be in
relative accord or discord (p. 2, Rappaport, 1977)”.

The failure of oversimplified risk models that do not
account for person-environment fit reflects the theoretical

work of Patterson (1989) and Elliot (1985) on the

49

developmental pathways of juvenile criminality, which
suggests that the undesirable behavior of youth is both
complicated and multifaceted. In some cases delinquency is
a product of psychological pathology, in other cases a
result of detrimental environmental factors or in still
other cases some combination of both, intrinsically
implying that juvenile offenders are a rather heterogeneous
population (Cullen & Agnew, 2006).Thus while a myriad of
risk factors have been individually correlated with
delinquency, it is the combination and cumulative effect of
risk factors that best predict delinquency (Schwalbe,
Fraser, Day & Cooley, 2006).

Psychometricians have consequently attempted to
develop instruments that account for cumulative risk across
several domains in additive models where higher scores
reflect higher likelihoods of delinquency, classifying
youth as high, moderate or low risk with corresponding
rates of re-offense associated with each risk category.

While the widespread use of risk assessments reflects
the resurgence in their popularity, the actual first use of
risk assessment dates back to as early as 1928 with the
Burgess instrument which was used by the Illinois parole
board in parole decisions (Burgess, 1928). The 21 item

Burgess Instrument outperformed prison psychiatrists in

50

predicting parole failures for 3,000 Illinois offenders
(Burgess, 1928). Its early use demonstrated the intended
purpose of structured risk assessments in juvenile justice;
that being to predict offending and in doing so risk
assessment aides in case management and policy decisions
through objectively and consistently assessing juveniles
given the complexity of juvenile criminality.
Re-Emergence of Dynamic Risk Assessment.Measures

In 1996, Bonta conducted a meta-analysis of
correctional classification literature and described the
evolution of risk assessment instruments, concluding that
there have been at least three generations of risk
measures. The first generation was comprised of clinicians
and practitioners offering their professional opinion about
the likelihood of particular delinquency outcomes (Bonta,
1996). While the clinical method is still in use today, the
accuracy and validity of predictions made in this manner
has been shown to be inadequate (Lowenkamp, Holsinger, &
Latessa, 2001; Van Voorhiss & Brown, 1997). This has been
demonstrated in the work of Grove and Meehl (1996) who
reviewed 136 studies since the 19203 across disciplines in
which clinical and actuarial practices were compared. The
researchers found that 64 of those studies favored

actuarial practices versus eight siding in favor of

51

clinical/human decision making (Grove & Meehl, 1996). In
assessing risk, human judgments result in high false
positive and false negative rates for several reasons.
Clinicians and practitioners often erroneously ignore base
rates of offense, make spurious correlations and are swayed
by incorrect causal attributions (Gottfredson & Moriarty,
2006) .

The second generation of risk assessment was comprised
of risk measures using static risk factors like race or age
of first crime to predict outcomes (Bonta, 1996). The risk
factors used in such measures were theoretically derived
and had statistically significant relationships with
delinquency. For instance, with regards to the latter
aforementioned risk factor, researchers like Moffit offered
the supposition that there are life course persistent
offenders and adolescence limited offenders, explaining
away differences in offense rates between juveniles as
being heavily influenced by age at which first offense is
committed (Moffit, 1993). This supposition was reflected in
offense rates which showed that age of first offense was a
strong indicator of future delinquency. In turn other risk
factors like race and gender were also strong indicators of
future delinquency and were thus included in many risk

assessment measures .

52

The risk measures offered an advantage over the
clinical method with regards to accuracy and validity
(Bonta, 1996). Thus risk assessment was initially steeped
in the individual differences tradition; however the
failure to consider the context of these factors across
multiple levels in individuals’ lives reduced the
effectiveness of these instruments in correctly predicting
crime outcomes and in guiding case management (Bonta, 1996;
Wiebush et a1. 1995). As static factors like race and age
of first offense are not malleable to change, second
generation risk assessment provided little useful
information that could be used in intervention efforts.

There was also considerable evidence that certain risk
factors included in second generation prediction
instruments were more salient with certain groups than
others. For instance, the occurrence of hyperactivity is
more common in boys than girls (Lahey et al., 1999) and in
turn neighborhood risks pertain to a larger portion of
Black youth than Whites (Loeber & Farrington, 1999)
therefore when using these factors to assess risk, some
groups automatically exhibit elevated risk levels on these
risk assessments that did not bear out in delinquency

out come 3 .

53

Differences in level of association between certain
risk factors and delinquency for different social groups
pose a threat to the dimensional identity of measures used
with diverse populations. Again, dimensional identity
refers to the generalizability of validity and accuracy of
prediction models across various subgroups of population
like race or gender. Specifically in the case of juvenile
justice it has become fundamentally clear that ignoring
poor dimensional identity and ignoring the person-
environment fit in risk assessment has tremendous costs in
terms of justice and distribution of resources (Bonta,
2002). This occurs through biases prejudicing the system
against members of certain groups or poorly distributing
resources to wrongly classified individuals.

The third generation of actuarial assessments
represented a departure from the individual differences
tradition and an attempt to adopt an ecological model with
dynamic risk factors. Bronfenbrenner (1979) offered that
the undesirable social behavior of youth was a product of
poor fit between various systems on multiple levels in
which an individual fits. The micro-level was characterized
as layer of interaction between the youth, family and
immediate environment, while the mesosystem was

characterized as connecting layer between the family,

54

social services and school. Finally, the macro-system
referred to cultural expectations, laws and views which
produce a cascade effect through the various layers of the
youth’s external and internal environment that ultimately
impact the child. As this pertains to risk assessment, the
failure or success of an instrument at predicting the
undesirable social behavior of juveniles is a direct
reflection on its ability to capture information on the
person in larger contexts of his or her environment. As
such, instruments of prior generations that overly focused
on individual and fixed characteristics of the individual
were inadequate and thus were doomed to failure. Third
generation actuarial assessments reflected these series of
interactions by using the cumulative risks of several risk
factors on multiple levels in predicting offense.
Consequently the power of risk assessments has become
dependent upon the risk factors or risk subscales that they
tap into. In a widely cited meta-analysis, Cottle, Lee and
Heilbrun's (2001) synthesis of risk factor literature
demonstrated that offense history, psychological factors,
school behavior, attitudes toward authority, gender factors
and peer factors are each strongly associated with re—

arrest, probation violation and commitment. In turn many

55

risk instruments tap into these domains in predicting
delinquency.

The third generation of actuarial assessments is
comprised of measures using dynamic and static risk factors
to determine the likelihood of the youth coming in further
contact with the system. Dynamic risk factors are risk
factors that are malleable to change efforts of clinicians
and practitioners, whereas static factors are fixed and
rigid characteristics of a youth or their environment that
cannot be changed. This cumulative multi-domain approach is
reflective of the axiom spoken to earlier: the more risks a
juvenile experiences in multiple areas of their lives the
more likely that juvenile will engage in some form of
delinquency in the future. For these instruments to be
effective, Bonta strongly advocated periodically validating
these measures on populations of interest to system
workers. Some researchers have recommended that measures be
validated every two years. With at least 39 states using
risk assessments, adherence to that recommendation would
mean that a number of studies have been conducted in the
past decade on the validity of risk measures.
Unfortunately, since 1996 there have been no systematic
reviews of studies validating juvenile risk assessments

qualifying as third generation instruments.

56

Chapter 4

Validity, Accuracy and Dimensional Identity of Risk
Assessments

Predictive validity in the case of actuarial risk
assessments refers to the extent to which the scale
predicts a delinquency outcome (Walters, 2006). In that way
an instrument can be validated in a number of ways and in
fact have been. For instance, a study could show that there
is a significant correlation between risk score and
recidivism. Another study could show that recidivists and
non-recidivists have significantly different scores on the
instrument. Yet still another study could show that there
are significant differences between juveniles that have
been classified as “low or high risk" in recidivism rates.
Researchers then have wide latitude in determining the
validity of their risk assessment measures, making
comparisons extremely difficult.

However, several researchers have made recommendations
as to how predictive validity studies be conducted and what
validity coefficients should be reported. Nunnally (1978)
noted that the accuracy of predictive measures are always
higher with samples with which they are created than with

independent samples; therefore validity studies should

57

utilize independent samples. DeComo, Wagner, and Baird
(1994) recommended that instruments differentiate between
high risk groups and low risk groups by at least 4 to 1 or
30% on predicted delinquency outcomes. This means that the
recidivism rate for juveniles classified as high risk
should be four times that of juveniles classified as low
risk using a prediction instrument.

Much of the risk prediction literature has focused on
the actual statistics and validity coefficients used in
describing the relationship between risk assessments and
delinquency outcomes. In a widely reported criticism of
validation practices solely using correlations or
association statistics to validate measures, Loeber and
Dishion (1983) argued that the validity of predictive
instruments depends on their ability to minimize false
positive and false negative error rates with regards to
predicting re-offense. The base-rate of re-offense
represents the odds of correctly guessing that any given
offender will re-offend. It is possible for a measure’s
risk scores to be significantly correlated with a
delinquency outcome yet incorrectly identify the vast
majority of its offenders as recidivists or non-
recidivists. Therefore, if any instrument does not

correctly identify re—offenders at a greater rate than

58

chance, such instruments offer little practical advantage
with regards to case management in spite of possible
significant correlations between risk and offense outcomes.

Since 1983 a number of validity coefficients which
account for base-rates have been offered as superior
alternatives to correlations such as Area Under the Curve
(AUCs), and Relative Improvement Over Chance (RIOC)
statistics. Given the myriad of validation protocols, this
review reported how common each of these methods/techniques
was in reported validity studies.
Results of Systematic Review

As reported in Table 1, between 1996 and 2006, 23
studies were published with the majority (61%) presented in
the past 5 years. Sample sizes ranged from 35 to 9,692
participants with a mean sample size of 1,498 (SD=2,748).
Of the studies reporting age (20 of 23), the average age of
participants was 14.5. Of the studies reporting gender (18
‘of 23), sample compositions for males ranged from 54% to
100%, with a mean of 70% (SD=11%). Of the studies reporting
racial demographics (20 of 23), 12% to 100% of samples were

White youth; in the typical study 45% (SD=25%) were Whites.

59

 

mmmH

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o x x x x x x x acumz m waa was 1H .oma .mqsn
a a «com
o x x x x x x x «ohms H z z z ama .aaHmz
_ a a moon
2 x x x x x x x ommom H waH was 0H mHH .umcuse
«com .
z x x x x x x amdmm m was was mH ans acmnmsEHn
aaaH
o x x x x x x Ham 2 wmm 2 SH aaa .suHsm
a moon
6 x x x x x asqomoo H won wsm aH amH .smxumnm
m a «com
2 x x x «moooz H wmm wam aH msw .mesns
a a woom
o x x x x x mauz a was wss «H mma .onHmsnom
a vooa
: x x x x x «402 H wom was aH «ma .onHm3nom
m w moon
0 x x x x x x x x Had a wooH 00H sH waH .monzoz
z moon
2 x x x x HamooH wow wam aH woe .muuncHuo
m aaaH
z x x x x x auH a wma was aH sHH .uusm
Hmoo: m aaaH

o x x x x x seem a wMH wss aH com .xcnm .
a oooa
o x x x x Hades a was was 4H HaH .uonHm
aaaH
x x x x x x x aumHm m was wmm mH aoa .xcmum
a Hooa
o x x x x x mamau H wan wwm mH ma .uaHso
o oooa
u x x x x . a mango m 2 was NH maH .momfiom
n Hooa
z x x x x asmu< H waH was mH «ma .oooz

Hum 93 D 0 oh 0

anon .n A m u 0 U u n a an» naH h “an: on m4 2 H2334

.mucmESHumnH ucmEmmmmmm xmwu nuw3 mwflcdum aoHumpwam> m0 Goflumfihommn

.H mHnms

6O

 

 

 

 

 

 

2 Sim .anwx 335qu 5333631 mem aaaH Janna
Hu>vq
o Ho.vm .Ha.uu ummnud smH>HoHomm mem Hooa .umHno
quunou mnoom
o Hove .H.ua .m.m socmsucHHon mem 82 .398:
Hw>mq
z Ho.vm .oannx ummnnd EmH>HoHomm mem Haas .9662
oaks
unuHoHuuooo oHfikus
nmauon >ufiuwfiu> uuouaomon uoanoufluo oHuom Hogans

 

 

 

 

 

 

 

.mmHUSDm coaumoflam> no muadmom .m mHQMB

maHummB wocmnowufio memo: u z HomHmun HmaoHumHmHHou u U n mmflnucm GEDHOU :mHmmow
“mom n H axhoumHm omcuwuo n n
“kuHHMGOmHmm
n m HucsuHuud n m Hmudmsmq n m Hmonmumndm u c «handEEou u o Hxaflemm n a HGoHumuzcm u m
mango: SH numnmq Q3u3oHHom u .D.mumamnmq nEdHou
uoz n Z "0502

w a N
0 aooa OOH ha mwm
09 OH on

u 0U

 

U x N x x x x N x H£U\W.Hw w «an. wmw ma had .uflfififiuw
Avoaawunoo .H canoes

61

 

mnoum

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

sm . "02 33.3 533303 smHm Son dengue:
Hwamuoos 9a 850m

icon .Ho.nam umuuud Emfl>flcHoom mem moon .xxuuumz
muoom

Hwioom 39.82 smH>HoHomm mem moon Susana
OHOUm

given among EmH>HUHowm mem aaaH 6:2.
mnoom

$.62 33.2 53383 033. Sea .853
ao.vm Hm>wq

.m.mmu ax ummuud 533303 033 mooa £958.

Hove mnoum woos

m.mm unx ummHH< Emfi>wUHuom xmflm .GOmuoEEHb
mnoom

ms.uuo< Hmong EmH>H8Humm mem aaaH .fiHEm
mHoum

$.69“ umouaa EmH>HoHomm 063 noon Suxumnm
Hmuoludo Hm>wq

ave «6“an pawn: 153333 mem moon snag

mnoom moon

Hove 60.751 93 $6.34 EmH>HcHoom mem 63958

mHoom soon

Ho.vm .m.mHon mmH ummnnfl EmH>HUHoom xmflm .onamznom
muoom

ss.»oo< 833280 5233631 mem 23 .8352

mdumum moon

Hove .H.aaum onoom mem mnoom 63 83:33 .muusfiuo
Ho>mq

Hove .mHuwx Havana 533303 32 aaaH dual
whoom

Halsm ”immune 5333031 mem aaaH 515E
wuoum

$.6on $32 533303 SE coon .uonHm

 

 

Hoosnaunou .u anueV

62

 

mo.vm mo.vQ mm.o vv.o H.o NH.o w.mH m.om omHH «mm soon .mouon
.Hm.n.H .mmiu o

¢\z mm.o oa.o ma.o mo.o (\z ¢\z <\z (\2 : soon .mmumz

Nv.nnm oo.Hn~m «\z «\z <\2 <\2 <\z <\z sz mva voom

u .GOmquEHo

ao.vd mo.vd aa a an o H.o mo.o a.aH H.nn ams awn moon .uocuss
{with .HN.n...H z

.m: <\z <\z <\z ¢\2 w.m a m m¢n wHH voom

z .onamzsum

mm.uUOHm wN.uUOHm mv.o mm.o ¢H.o H.o ¢\z ¢\z #Hm mmm 2 mama .mwflm

mm.nuom mm.nuom vm.o mm.o v.0 av.o w.va m.mH woa mm moon

0 .>oxumnm

.mc .HH. um ¢\z txz ¢\z «\z H.HH m.HH nsH om U mmmH .mnsn

mo.vm mu .va.nH hm.o ¢.o mH.o o s.mH H.mH hm ov moon

tmN . IN . U .uUfiEQUW

sH.I~m om.nnm <\z <\z ¢\z <\2 ¢\z <\z mom «HA 0 mama .xfldm

4:2 <\z 4\z <\Z <\Z <\2 ¢\z ¢\z Nfim mam U coon .momvom

<\z <\z ¢\Z 4\z <\z (\z m m vow or" 2 doom .QUUZ

m>om manwo whom uHHHU whom mauao whom mauuw mkon mHHwU
ucoHuHuuoou quoHHm> nmwm 30H ououm mmwz uNHm uHmEmm vogue: Honusd

.mUGmESHUMCH UGOEmmmmmm MQHH NO

 

 

oumm Emfl>HUwoom

hufluchH HMGmeCoEHQ .m sagas

 

 

 

 

 

 

 

 

H0>04
z Ho.vd .w.asu~x umwnum EmH>H6Howm mem aooa .aunoHa
whoom moon
6 ms.non< umonu< EmH>HoHuom mem .oHomnouuu

Aconawucou n oanmhv

 

63

Results of Systematic Review

As reported in Table 1, between 1996 and 2006, 23
studies were published with the majority (61%) presented in
the past 5 years. Sample sizes ranged from 35 to 9,692
participants with a mean sample size of 1,498 (SD=2,748).
Of the studies reporting age (20 of 23), the average age of
participants was 14.5. Of the studies reporting gender (18
of 23), sample compositions for males ranged from 54% to
100%, with a mean of 70% (SD=11%). Of the studies reporting
racial demographics (20 of 23), 12% to 100% of samples were
White youth; in the typical study 45% (SD=25%) were Whites.

As demonstrated in Table 1, 16 separate risk
assessment measures were used in these studies. The Youth
Level of Service/Case Management Inventory (YLS/CMI) was
the most widely used with 26% of studies conducted with
this instrument. The Child and Adolescent Functional
Assessment Scale (CAFAS) and North Carolina Assessment of
Risk Instrument (NCARI) were each used twice. Twenty—nine
percent of the instruments had fewer than 10 items, with a
single measure comprised of 60 items. Each of the studies
used multi-domain sampling. It is important to note that no
studies accounted for surveillance levels or macro-system
variables outlined by Bronfenbrenner (1979). Table 1

details the overlap of domains used in the study’s

64

instruments. Education, family, substance abuse and offense
history were assessed in nearly all instruments. The
community and leisure time domains were only found on a
small percentage of risk instruments. The Life Challenges
Questionnaire Risk Assessment Instrument and Alamedo County
Placement Risk Assessments were the only instruments that
used the self—report of the juvenile rather than multiple
sources.

validity and Accuracy

As evidenced in Table 1, researchers used two primary
designs to validate the instruments. Either the study used
a correlation design (n=14) or a means-difference design.
Means-difference designs were between risk level groups
with membership in comparison groups determined by risk
assessment scores of study participants (n=9). In the
correlational studies, researchers examined the level of
association between risk score and delinquency outcome. In
the means differences designs, researchers checked for
significant differences between groups of youth of
different risk levels.

The correlation studies varied in sample size from 35
to 9,672. The follow—up periods ranged in length from 6
months to 60 months with the typical study following the

youth for 19 months subsequent of the initial assessment.

65

For example, Quist and Matshazi (2000) found a significant
correlation (r=.41, p<.05) between risk score on the CAFAS
and recidivist arrest for 35 juveniles from a youth
rehabilitation group home that were observed over a 12
month period post-assessment. It should be noted that
studies merely reporting correlations between scores and
delinquency outcomes have been criticized as being
insensitive to base rates and provide little information as
to the false positive and false negative predictive
accuracy of the instrument (Loeber & Dishion, 1983). While
Quist and Matshazi simply reported the correlation between
recidivist arrest and risk score in validating the CAFAS,
this relationship was reported in a number of ways in other
studies that failed to provide an idea of the correct
classification rate of the validated instrument. For
instance, Hodges and Kim (2000) were unique in that they
constructed a logistic regression model for the
relationship between CAFAS risk score and contact with law
in a study with 1,460 youth that were observed over a six
month period. Controlling for background variables, they
found that the total CAFAS score increased the log—odds of
contact with the law by 1% for every point a youth scored

on the CAFAS. While the CAFAS was concluded to be a valid

66

predictor of delinquency, there was no report of how
accurate the instrument was in making this prediction.

Each of the studies with correlation designs used risk
score as a predictor of a delinquency outcome, thus
reporting the magnitude and direction of relationships
between relevant variables. However, the specific
delinquency outcome varied from study to study. As seen in
Table 2, eleven of the fourteen correlational studies used
recidivism arrest as the predicted delinquency variable
associated with risk score. Contact with law, time to first
arrest and conviction were each respectively used once as
predicted variables in the remaining correlation studies.

There was some variability in the manner in which
these associations were used in concluding that the measure
was valid, which is reflected in the reported validity
coefficients. For example, a single study by Quist and
Matshazi (2000) simply reported the correlation between
risk score and arrest and its significance level in
validating the CAFAS (r=.41, p<.01). However, two studies
(Funk, 1999; Marczyk et al., 2003) reported the amount of
variation in recidivism explained by risk score. The
Marczyk study was the single study that reported that its

respective risk assessment instrument was not valid given

67

the fact that 1% of the variation was accounted for by risk
score.

The remaining ten correlation studies used regressions
models to predict outcomes. The correct classification rate
was reported in determining the accuracy and validity of
the measure. Seven studies reported Area Under the
Curve/Receiver Operator Coefficients. The AUCs ranged in
value from 57% to 92%. In general, the higher the AUC the
better the classification rate. As such, the difference
between the AUC and the base rate of recidivism is the
improvement over chance in classifying juveniles correctly
with that particular instrument. Thus, the FORAI instrument
which was used by Risler and others (2000) is reported as
having the greatest improvement over chance in predicting
recidivist arrest (AUC = .92; improvement over chance =
.42), while Mckinnon (2004) reported that the YLS/CMI had
the lowest predictive accuracy (AUC = .57; improvement over
chance = .07). It should be noted several studies were
conducted in this manner with the YLS/CMI. Schmidt and
others (2005) in a longitudinal study over 60 months with
107 youth reported an AUC of .61. In turn, Catchpole and
others (2003) reported an AUC of .78 in their study over 12
months with 74 youth. Studies reporting AUCs are further

reported in Table 2 and the respective instruments were

68

each at least 10% better than chance at correctly
classifying the juveniles. Of those reporting an AUC, the
average was .68.

In the means-difference testing studies (n=9),
juveniles were grouped by risk level. Recidivism was then
collected during a follow-up period and compared across the
risk level groups. For example, Flores, Travis and Latessa
(2004) found significant differences in recidivist arrest
rates over 12 month period between juveniles classified as
low, moderate and high. In this sample of 1,679 youth, high
risk youth were 44% more likely to be re-arrested than low
risk youth. The remaining studies were similar in nature in
that they looked for differences in recidivism rates
between youth of different risk levels. In these studies,
length of follow-up ranged from 6 months to 12 months. The
delinquency outcomes used were detention (n=1), arrest
(n=5), time to first arrest (n=1), and conviction (n=1).The
sample sizes ranged from 250 to 4,574. As reported in Table
2, each of the instruments validated in this manner had
significant differences in delinquency outcomes between low
and high risk young offenders. There were two studies which
used unique ways to determine the validity of their
instruments. Grinberg and others (2005) took a different

approach in their study with 404 youth validating the Life

69

Challenges Questionaire Risk Assessment Instrument. A
sample of non—detained youth (n=246) and detained youth
(n=87) were compared on the instrument, with significant
differences in scores being found as reported in Table 2.
The researchers consequently concluded that the instrument
was a valid predictor of delinquency. In the Krysik and
Lecroy study (2002) with the National Council of Crime
Delinquency Risk Assessment assessed on 4,754 youth, they
used the DeComo, Wagner and Baird (1994) [PPM] validation
method in which high risk and low risk groups were expected
to differ in recidivism rates by 30% or by a 4 to 1 ratio.
With this instrument the PPM was less than 4, with 76% of
high risk youth offending and 24% of low risk youth
offending. With the exception of that study, each of the
instruments validated were concluded to be valid due to
significant differences in delinquency outcomes between low
and high risk young offenders (Table 2).

Dimensional Identity

As stated earlier, dimensional identity refers to the
consistency in prediction of delinquency outcomes for an
instrument across subpopulations in a sample (e.g. race and
gender). While theoreticians have repeatedly recommended
that validation studies examine the validity of prediction

instruments with girls and minorities, few studies have

70

actually done so. As shown in Table 3, in only 12 of the 23
studies was the differential validity of risk measures by
gender explored. Only five of the 23 studies included an
exploration of differential validity with ethnic and racial
groups. Only a single study (Schwalbe, 2004) reported the
correct classification rates separately by gender and
ethnicity. Further, few even tested for significant
differences in recidivism rates by risk level between
groups. Tables 3 & 4 show the respective sample sizes and
validity coefficients of these groups where reported.
Studies checked for dimensional identity in a number of
ways. Some studies checked for differences in recidivism
rates between groups that shared risk levels but differed
in gender (n=4) or ethnicity (n=2). Others looked for
interactions between gender (n=2)/ethnicity (n=1),
recidivism and scoring in delinquency outcomes. Still
others (gender, n=2; ethnicity, n=1) reported respective
correlations for each group between score and delinquency
outcomes.

Given that only a few studies examined dimensional
identity, it was consistently reported that the higher the
risk (by group) of the youth, the higher the recidivism
rate regardless of race or gender. When this pattern was

further examined, notable differences emerged that merit

71

note and provide a direction for further study. Three of
the nine studies that examined recidivism rates for low
risk juveniles separately by race and gender reported
differences of at least 9% (Schmidt et al., 2005; Turner et
al., 2005 and Flores et al., 2004). However, no '
significance testing was conducted to determine if these
differences were reliable. In each case where there was an
observed difference greater than 9%, girls and non-
minorities recidivated less than boys and minorities
despite being classified in the same risk category.
Moreover, when examining trends with high risk youth, 6 of
the 9 studies showed differences greater than 9% for
girls/minorities and their male/non-minority counterparts.
Again, girls and non-minorities that shared risk
classifications recidivated less than the respective male
minority group.

In the studies reporting correlations, there were
absolute differences in the validity coefficients as a
function of gender and racial groups. Again, the
statistical difference between the coefficients was not
examined. In the two studies where race and gender were
added to regression models, both found race and gender to
be irrelevant. In perhaps the strongest study of

dimensional identity, Schwalbe and others (2004) assessed

72

464 youth with the NCAR and observed their recidivist
arrest over 12 months. While a quarter of the sample was
comprised of girls, more than half were comprised of
minorities. The NCAR showed a significant relationship
between score and recidivist arrest; however the
researchers took a further step in examining this
relationship with regards to race and gender. Performing a
Cox regression they found significant differences in the
time to first arrest and pattern of arrest over time. Boys
reached the base rate of offense at lower scores than
girls. Researchers therefore concluded that the NCAR had
poor dimensional identity.

Based on the literature reviewed here, the practical
appeal of risk assessment instruments appears to far exceed
their scientific foundation. While it is the case that the
research to date provides encouraging results, much more
work needs to be done before the current excitement over
these methods is scientifically justified. A number of key
points are relevant.

First, in the last decade there have been relatively
few scientific investigations of the validity of risk
assessment. Further, the studies summarized used a wide
variety of methodologies and relatively short follow-up

intervals. Of the 23 which this review identifies, it

73

appeared that reasonably representative samples were
employed. Of the 23, 16 different instruments were
employed and only on instrument appeared in more than two
investigations. Further, the average follow—up time period
was only 16 months and only eight studies included follow-
ups. It should also be noted that there was wide variation
in what delinquency outcomes were actually predicted. While
future arrest was quite common, several studies attempted
to predict when such arrests would occur in the future.
However, little rationale was provided for the different
outcomes employed. This is important because there is
little research supporting the validity of risk assessment
measures in predicting arrest, let alone time to first
arrest. In examining the study conducted by Schwalbe and
others (2004) it is clear that the likelihood of arrest can
change over time and that trajectory is not necessarily
linear. As such, the appropriateness of simply adding this
variable to logistic regression models should be considered
when examining this variable. Moreover, no attention was
given to the accuracy of these predictions over time. No
study checked for changes in risk scores over time. The
studies offered no commentary as to how long the

predictions should be considered accurate for the assessed

74

youth; therefore we cannot know how often the assessment
should be given over the course of time.

Second, the multiple instruments demonstrated only
modest agreement about what was important. The domains of
education, family, substance abuse, offense history, and
peers were consistently covered, but community, leisure,
attitude, and personality were present less than half the
time. Essentially, the concern expressed by Bonta (1996)
continues. The risk assessment may well benefit from
conceptual clarity and consensus as the field matures. For
example, the ecological perspective may provide a more
desirable as a basis for developing risk assessment
measures. The failure of oversimplified risk models that
do not account for person-environment fit reflects the
theoretical work of Patterson (1989) and Elliot (1985) on
the developmental pathways of juvenile criminality which
suggests that the undesirable behavior of youth is both
complicated and multifaceted. In some cases delinquency is
a product of psychological pathology, in other cases a
result of detrimental environmental factors or in still
other cases some combination of both, intrinsically
implying that juvenile offenders are a rather heterogeneous
population (Cullen & Agnew, 2006).Thus, while a myriad of

risk factors have been individually correlated with

75

delinquency, it is the combination and cumulative effect of
risk factors that best predict delinquency (Schwalbe,
Fraser, Day & Cooley, 2006).

Third, researchers seemed satisfied to classify
juveniles as low, moderate or high risk and observe a
significant difference in recidivism rates between the
groups. While statistical significance is certainly an
important criterion, strength of association and accuracy
of prediction are more relevant in this case. For example,
only one study used the recommended practice of looking for
a 4 to 1 ratio between recidivism rates of the high and low
groups. In examining the reported recidivism rates by risk
level, it is clear that jurisdictions differ in recidivism
rates that are considered low and those that are considered
high. As seen in Table 3 (Sharkey, 2003), low risk girls
assessed with the OCPDLRA recidivated twice as much as high
risk girls assessed with the SDRRC. What then does it mean
for a juvenile to be low risk if a quarter of this number
will re-offend but half of this group will re—offend
according to another instrument? This highlights the
necessity of validating these measures repeatedly on
different populations in different settings.

Fourth, the dimensional identity of these instruments

is not well established. Research in this area is very much

76

in its infancy. There were 23 studies examining the
validity of risk measures, yet only 12 were conducted on
girls and 5 with non-majority ethnic groups. The success
of these instruments varied from study to study and in fact
the determination of the consistency of the measure across
gender and ethnicity is unclear. Most studies simply
reported correlations and recidivism rates for the
respective groups and then used “inter-ocular t—tests” to
decide if the differences were important. However, given
the small sample sizes and limited replication of results
it was impossible to attribute with any certainty
differences in predictive validity to the dimensional
identity of the instruments. Moreover, there were only two
instruments, the YLS/CMI and NCAR that were validated more
than once with diverse populations. In the case of the
latter, the creator was the single individual validating
the measure. In turn, the instrument with the greatest
overall accuracy, the FORAI, was also only studied once and
did not report its accuracy and validity with girls and
diverse populations. Rather than determining that these
instruments were valid and accurate with diverse
populations, these studies indicate the necessity of
further research with better samples and consistent

methodology.

77

This review was able to provide us with a glimpse of
what an ideal validation study should look like. Studies
should have large enough samples that classification based
on risk level can be made not only on the general populace
but with both gender groups and relevant racial groups in
mind. While checking for differences in delinquency rates
between risk levels is useful, studies should also check
that there is a 4 to 1 ratio between high and low risk
groups. Most importantly, researchers should use their risk
models on a case by case basis to predict relevant outcomes
for their samples reporting the correct classification
rates for the different racial and gender groups. More
attention should also be given to the specific delinquency
outcome of time to first offense. Also, when validating
each of these instruments other researchers should be
encouraged to follow-up these studies periodically with

checks in different populations.

78

Chapter 5

Rationale for Validating the YLS/CMI

There is some evidence in the literature supporting
perspectives that assert risk assessments will under-
predict recidivism for African American youth due to
exclusion of relevant factors at various levels of a
youth’s life like exposure to policing, resiliency, socio-
economic status. This evidence also supports perspectives
that assert girls are relatively minor offenders and
offense rates are lower for female offenders in spite of
sharing risks in common with their male counterparts.
Consequently, instruments are highly likely to over-predict
offending for this group.

The focus then of this study was the predictive
validity of a widely used and popular risk measure, the
Youth Level of Service Case Management Inventory, across
gender groups and with Whites and Blacks in a particular
juvenile justice system. The YLS/CMI is a 42 item risk
assessment instrument with eight subscales [Education,
Family, Drug Involvement or Substance Abuse, Personality,
Attitude, Peer Association, Free-Time use, and Offense
History]. Seven of the eight subscales are dynamic risk

factors that have been shown in previous research to be

79

associated with offending. Offenders scoring between 0 and
8 are considered low risk, 9 and 22 are moderate risk, and
23 and 42 are high risk for re-offense.

In two previous studies, validating the YLS/CMI as a
predictor of recidivism for minority groups and girls, the
instruments received mixed but generally positive ratings.
In the one study Jung and Rawana (1999) tracked 250 youth
that were administered the YLS/CMI over six months,
ultimately correlating their risk scores with their
recidivism rates. They compared the risk-recidivism
relationship of boys and girls; in addition to that of
Native Canadians and non-Native Canadians. They found that
the instrument differentiated between recidivists and non-
recidivists for all subgroups. When we turn our attention
to girls and Blacks, there have been relatively few studies
that examined risk-recidivism relationships across groups.
For instance, our previous validation work with the YLS/CMI
with a one year follow-up suggested the possibility that
the YLS/CMI under-predicted recidivism for African-American
boys; however, given limited sample sizes for the various
subgroups conclusions about the instrument’s dimensional
identity were tentative at best, necessitating larger

sample sizes and a longer follow-up period.

80

Since then Flores, Travis, and Latessa (2004) further
explored these differences in predictive validity between
gender groups and racial groups on the YLS/CMI in one of
the largest studies to date. Their sample was comprised
1,671 youth in total, with 79% of the sample being male and
21% being female. In turn 67% were white and the remaining
portion reported simply as “nonwhite.” A significant
correlation between YLS/CMI score and re-arrest was found
for each group with girls having the highest validity
coefficient, r=.318, p<.05; and nonwhites having the lowest
validity coefficient, r=.208, p<.05. However, when the
actual recidivism rate by risk level analysis was performed
there was some indication that differential predictive
validity existed for some groups despite shared
directionality of the relationship between risk score and
recidivism for each group. A chi—square analysis revealed
that the recidivism rate by risk level distribution was
substantially different for the respective groups. In
other words the rate of re-arrest for “low” risk white
juveniles (8%) was substantially lower than that of “low”
risk juveniles (40%) in the nonwhite sample. Moreover, the
recidivism rate of “moderate” risk boys was greater than
that of “very high" risk girls. Flores and others concluded

that the YLS/CMI adequately predicts recidivism for the

81

various subgroups, but requires re—norming of the risk
levels for those various subgroups.

Thus the YLS/CMI has received the greatest amount of
attention from researchers and has proven to be relatively
accurate in identifying offenders with a great likelihood
of repeat contact with the juvenile justice system.
However, despite the amount of attention it has received
few studies have examined its validity specifically with
girls and African Americans. In that context, this study
answered two fundamental questions:

21.How well does the YLS/CMI predict recidivism for

African Americans and girls?

:2.Are there differences in the risk-recidivism
relationship between these understudied groups and the

traditional sample?

82

Chapter 6

Methodology
Setting

This study took place in a medium sized Midwestern
county. The United States Census estimated the population
of this county to be 277,000 in 2006. There are 60,940
youth under the age of 18 in the county. Female youth
comprise 51% of the population, while African Americans
comprise 11% of the population.

The Juvenile Court in this county serves youth in a
number of programs including substance abuse, residential
services, day treatments, sex offender services, diversion,
anger management, family services, detention, probation,
and truancy services.

Sampling and Data Collection

This study drew upon archival data from a pooled
sample of two divisions in the court from March 2004 to
March 2006. Juveniles from the Intake Division were
assessed on the YLS/CMI upon entry into the system; while
juveniles from the Probation Division comprised a cross-
sectional sample in that they represented all juveniles
receiving probation services in that time-period. The two

samples were selected to represent youth entering the

83

juvenile justice system and those already under the
jurisdiction of the court. There were no refusals.

Demographic Information. Ultimately this sample
included 968 youth between the ages of 9 and 18; with a
mean age of 14 years 6 months. There were 274 girls (28%)
and 694 boys (72%). There were 509 White youth and 459
Black youth.

Data Sources. Demographic variable data was collected
by the court through self-report of participants.
Participants reported their age, address, school grade,
gender, and race/ethnicity. Participants had the option of
selecting African-American, White, Hispanic, or other.

YOuth Level of Service/case Management Inventory. The
YLS/CMI has 42 items which comprise 8 subscales: Offense
History, Family Circumstances, Education, Peer Group,
Substance Abuse, Leisure Activity, Attitude, and
Personality. Previous research (Schmidt, Hoge & Gomes,
2005) has demonstrated that the YLS/CMS has high internal
consistency and reliability, which has been confirmed by
this researcher and is reported elsewhere. Each item is
scored dichotomously with a “yes" response receiving a
single point and a “no” response receiving no points. These
scores can be tallied for a risk score corresponding to

each subscale and a total score. The creators of the

84

instrument have offered recommended cut scores for low,
moderate and high classifications. Scores below 9 are low,
below 24 are moderate and above 23 high.

Subscales. Items on each subscale of the YLS/CMI had
dichotomously scored items [no=0, yes=1]. An examination of
the subscale structure is presented below, but the eight
original subscales (Hoge, Andrews, & Leschied, 2002) and
their item configuration are listed here.

a) Prior and Current Offenses, Adjudications. Five

items focusing on official criminal history (e.g. item

I-a, “Three or More Prior Convictions”).

b) Leisure/Recreation. Three items focusing on of

free time (e.g. item II-b, “Could Make Better Use of

Time”).

c) Education/Employment. Seven items focusing on the

school behavior and performance (e.g. item III-a,

“Disruptive Classroom Behavior”).

d) Peer Relations. Four items focusing on the

characteristics of friends and acquaintances (e.g.

item IV-c, “Lack of Positive Acquaintances”).

e) Substance Abuse. Five items focusing on pattern of

alcohol and illegal drug use and abuse (e.g. item V-d,

“Substance Abuse Interferes with Life”).

85

f) Family Circumstances/Parenting. Six items assessing
experience of parental discipline and relationships
(e.g. item VI-b, “Difficulty in Controlling
Behavior”).
9) Attitudes/Orientation. Five items reflecting
criminogenic and prosocial tendencies (e.g. item VII-
c, “Actively Rejecting Help”)
h) Personality/Behavior. Seven items reflecting
disruptive patterns of behavior (e.g. item VIII-g,
“Verbally Aggressive, Impudent”).
The item-total correlation for the sample is provided
below.
Table 4.
Corrected Item—Total Correlations for the

Intake and Probation Samples

 

 

 

Corrected
Item-Total
Correlation
Item
Offense History a=.81
Three or More Prior
Adjudications 0.31
Two or More Failures to Comply 0.40

86

(Table 4 Continued)
Prior Probation
Prior Detention
Three or More Current

Adjudications

.51

.49

.00

 

Family Circumstances;

Inadequate Supervision
Difficulty in Controlling
Behavior

Inappropriate Discipline
Inconsistent Parenting

Poor Relations/Father-Child

Poor Relations/Mother-Child

.65

.28

.65

.28

.42

.30

.31

 

Education

Disruptive Classroom Behavior
Disruptive Behavior on School
Property

Low Achievement

Problems With Peers

Problems With Teachers
Truancy

Unemployed/Not Seeking

Employment

.66

.46

.32

.56

.33

.41

.47

.27

 

87

(Table 4 Continued)

Peer Relationships

Some Delinquent Acquaintances
Some Delinquent Friends

No or Few Positive
Acquaintances

No or Few Positive Friends

.70

.32

.48

.44

.40

 

Substance Use

Occasional Drug Use

Chronic Drug Use

Chronic Alcohol Use

Substance Abuse Interferes With
Life

Substance Abuse Linked to

Offense(s)

.75

.44

.47

.17

.47

.30

 

Leisure

Limited Organized Activities
Could Make Better Use of Time

No Personal Interests

.79

.46
.50

.45

 

Personality

Inflated Self-Esteem
Physically Aggressive

Tantrums

88

.68

.07

.44

.42

(Table 4 Continued)

 

Short Attention Span 0.38
Poor Frustration Tolerance 0.53
Inadequate Guilt Feelings 0.42
Verbally Aggressive/Impudent 0.44
Attitude on: . 67
Pro-Criminal Attitudes 0.54
Not Seeking Help 0.51
Actively Rejecting Help 0.19
Defies Authority 0.60
Callous 0.48

variables. Demographic variable data was collected
through self-report of participants. Recidivism was defined
as any new criminal petitions up to 24 months following the
first YLS/CMI administration. Recidivism data was derived
from the data management system of the court. Tickets and
truancy violations were not counted. When referring to re-
offense or recidivism, this actually a reference to new
petitions a juvenile may have received. Two pieces of
information are included on the recidivism collection form:
recidivism (yes or no), and total number of charges. The

form is completed by court staff using the court’s data

89

management system and entered into a database to be used
for analysis.

Procedure. Court personnel had four full days (32
hours total) of training in the YLS/CMI prior to
administration. This training consisted of explaining
definitions and scoring criteria for each items, practice
cases and coding, and inter-rater reliability checks.
During training 369 pairs of interviewers rated 36
different cases. The resulting percent agreement was 90%.
Following training, one-twentieth of all coded cases were
randomly checked by a second interviewer from an audio tape
of the interview (N = 27). In all cases inter-rater

reliability exceeded 90%.

9O

Chapter 7

Results

Descriptive Statistics

The mean YLS/CMI for young offenders was 11.7
(SD=8.4), which was in the moderate risk range. Table 5
provides the item by item frequencies for each racial and
gender group. On average, 28% of youth exhibited risk on
any given item. The following items had the greatest
likelihood of a youth exhibiting risk: poor relationship
with father (51%), disruptive behavior in classroom (40%),
disruptive behavior on school grounds (43%), low
achievement (54%), some delinquent acquaintances (55%),
some delinquent friends (42%), limited organized activities
(57%), could make better use of time (61%), short attention
span (41%), and poor frustration tolerance (49%).

Table 5. Item Frequencies (% Yes)

 

 

 

African

American Whites
Item Boys Girls Boys Girls
Offense History
Prior Convictions 8% 5% 5% 5%
Failure to Comply* 15% 6% 10% 8%

91

(Table 5 Continued)

 

 

Prior Probation 17% 14% 17% 16%
Prior Custody 18% 20% 17% 14%
Current Convictions 3% <1% <1% 3%
Family Circumstances

Inadequate Supervision* 32% 14% 27% 18%
Difficulty in

Controlling behavior 40% 34% 40% 38%
Inappropriate Parental

Discipline 15% 13% 23% 19%
Inconsistent Parenting 431% 24% 37% 32%
Poor Relationship with

Father* 60% 47% 46% 51%
Poor Relationship with

Mother* . 26% 32% 28% 31%
Education

Disruptive Behavior in

Classroom* 50% 41% 41% 28%
Disruptive Behavior of

School Grounds* 46% 49% 39% 38%
Low Achievement* 60% 45% 56% 53%
Problems with Peers* 30% 46% 31% 27%
Problems with Teachers* 38% 31% 36% 26%

92

(Table 5 Continued)

 

 

 

Truancy* 41% 34% 33% 43%
Unemployed 6% 3% 6% 6%
Peer Group

Some Delinquent

Acquaintances 60%' 49% 55% 55%
Some Delinquent

Friends* 47%' 41% 35%' 45%
Some Positive

Acquaintances* 36% 19% 30% 23%
Some Positive Friends* 30% 20% 26% 16%
Drug Involvement/Substance Abuse

Occasional Drug Use* 34% 19% 39% 40%
Chronic Drug Use* 21% 8% 20% 20%
Chronic Alcohol Use* 2% <1% 9% 9%
Substance Use

Interferes with Life* 18% 8% 22% 23%
Substance Abuse Linked

to Offense* 9% 2% 17% 19%
Free time Use

Limited Organized

Activities 56% 62% 56% 53%
Better Use of Free Time 66% 62% 57% 60%

93

(Table 5 Continued)

 

 

No Personal Interests 25% 28% 21% 20%
Personality

Inflated Self Esteem* 7% 2% 6%’ .2%
Physically Aggressive 34% 41% 34% 30%
Tantrums* 24% 23% 30% 38%
Short Attention Span* 45% 30% 55% 32%
Poor Frustration

Tolerance 50% 44% 52% 51%
Inadequate Guilt 21% 14% 19% 13%
Verbally Aggressive 27% 35% 26% 32%
Attitudes

Pro-criminal Attitudes* 26% 16% 18% 13%
Not Seeking Help* 23% 12% 17% 16%
Rejecting Help 11% 6% 8% 8%
Defies Authority* 24% 14% 18% 12%
Callous* 15% 6% 10% 5%

Group Differences on Frequency of Exhibited Risk

Mean scores for African American boys
White boys and girls are detailed in Table
(M=10.6) scored significantly lower on the

boys (M=12.1), t(968)=-2.6, p<.01, however

and girls, and

6. Girls

YLS/CMI than

there were no

significant differences in scores for African Americans

94

(M=11.8) and Whites (M=11.5). Within the two racial groups
there were no significant differences in mean score for
boys or girls.

There were however, significant differences in
distribution of exhibited risk on individual items by race
and gender group as seen in Table 5. Boys exhibited risk at
a significantly greater rate than girls on the following
items: failure to comply, inadequate supervision, poor
relationship with father, disruptive behavior in the
classroom, low achievement, problems with teachers,
positive friends, occasional drug use, chronic drug use,
chronic alcohol use, substance abuse inferring with life,
inflated self-esteem, short attention span, pro—criminal
attitudes, not seeking help, and defying authority.

Racial differences in distribution of exhibited risk
were limited to drug involvement items. Table 6 shows that
with the exception of the drug involvement and attitude
subscales, there was little variation in mean score across
subgroups. Moreover African American female youth exhibited
significantly lower risk for re-offense on the drug

involvement subscale than the other subgroups.

95

Table 6. Subscale scores and ROC by race and gender

 

African Americans Whites

 

Boys AUC Girls AUC Boys AUC Girls AUC

 

 

 

 

 

 

 

 

 

Offense History 0.6 n.s 0.5 n.s 0.5 0.57 0.5 n.s
Family

Circumstances 2 n.s 2 0.66 2 n.s. 1.9 0.75
Education 2.7 0.59 2.5 0.61 2.4 0.59 2.2 0.67
Peer Association 1.7 0.6 1.3 n.s. 1.5 0.62 1.4 0.66
Drug Involvement 0.8 0.58 0.4 n.s. 1.1 0.59 1.1 n.s
Free time 1.5 n.s 1.5 n.s. 1.3 0.62 1.3 n.s
Personality 2.1 n.s 1.9 0.65 2.2 0.58 2 0.69
Attitude 1 0.55 0.5 n.s. 0.7 n.s. 0.5 0.61
Overall 12.4 0.6 10.6 0.675 11.7 0.63 10.9 0.72

 

As seen in Table 7, this trend is reflected in the
distribution of subgroup members by risk level. Overall,
45% of the sample was low risk, 42% moderate risk and 14%
high risk. African American female offenders and African
American male offenders exhibited significant differences
in risk score level with twice as large a proportion of
boys (18%) than girls (6%) rated as high risk. Girls
overall were under-represented in the high risk group
compared to boys.

In turn, for each group nearly half of their members
were rated as low risk for re-offense, while no more than

one in five exhibited a high risk of recidivism. When

96

examining mean scores on each of the subscales by gender

and racial group few differences emerged.

 

Table 7. Descriptive Statistics

 

 

Mean
YLS/CMI
Score
N (SD) Risk Level and Recidivism Rate
Low RR Moderate RR High RR
0
4.)
E 11.5 X“2=34.8,
3 509 (8 4) 46% 23% 42% 42% 13% 59% p<.05
10.9
Girls 146 (8.4) 48% 11% 42% 34% 10% 60%
11.7
Boys 363 (8.4) 45% 28% 42% 44% 14% 59%
x
O
3 11.8 X“2=21.6,
m 459 (8.4) 44% 36% 41% 56% 15% 61% p<.05
10.2
Girls 128 (7.3) 56% 23% 38% 41% 6% 63%
12.5
Boys 331 (8.5) 40% 43% 43% 61% 18% 61%
H
(U
8 11.7 X“2=54.9,
B 968 (8.4) 45% 29% 42% 48% 14% 60% p<.05
12.1
Boys 694 (8.5) 42% 35% 42% 52% 16% 60%
10.6
Girls 274 (8.0) 52% 17% 40% 37% 8% 61%

Significant Differences between Subgroups in Risk Level-Recidivism

Relationship

97

(Table 7 Continued)

a. White Low Risk Boys versus Girls

x*2=7.8, p<.05

b. Black Low Risk Boys versus Girls

x*2=8.2, p<.05

c. Black Moderate Risk Boys versus Girls

x‘2=6.o, p<.05

d. Low Risk Girls versus Boys

x‘2=14.5, p<.05

e. Moderate Risk Girls versus Boys

x‘2=7.3, p<.05

f. Black Low Risk Boys versus White Low Risk Boys
x‘2=6.8, p<.05

9. Black Moderate Risk Boys versus White Moderate Risk Boys
x*2=8.1, p<.05

Question 1. HOw well does the YLS/CMI predict recidivism
for African Americans and girls?

Over a two year follow-up period, 41% of assessed
youth re—offended (n=968). There were substantial
differences in re-offense rates by groups (see table 7).
For instance, boys (46%) re-offended at a significantly
higher rate than girls (29%). Furthermore, over half of
African American male youth in the study recidivated (54%).
This rate was followed in size by White male youth in which
39% of this group re-offended. African American female and
White female youth had lower recidivism rates with 32% and
26% respectively re-offending over the twenty-four month
period following assessment.

Risk-Recidivism Relationship

Overall, the YLS/CMI demonstrated its ability to
differentiate between non-recidivists and re-offenders over

a two year period moderately well (see table 8). In each

98

group, recidivists scored significantly higher than non-
recidivists on the YLS/CMI (p<.01). The difference in mean
scores between recidivists and non-recidivists was as
follows: African American males (3.5), African American
females (5.1), White males (4), and White females (6).
Logistic regression allows us to examine the
relationship between the continuous variable, YLS score,
and the dichotomous variable, re-offense. Moreover, the
statistical technique can provide information on
incremental percentage increase in likelihood of offense
for each point scored on the YLS/CMI. For example in each
of the groups, there was approximately a 1.06 percent
increase in the likelihood of re-offense for each point
scored on the YLS/CMI. From 3% to 13% of the variation in
re-offense rate was explained by YLS/CMI, thus the effect
size for each group was small but significant. Overall, 7%
[Nagelkerke R22] of the variation in offense rates was

explained by total risk score.

99

Table 8. Hierarchical Logistic Regression for YLS/CMI score

and New Offenses

 

Correct
95% C.I. for Nagelkirke Classification

Group) Exp(B) Exp(B) R‘2 Rate

 

African American

Boys 1.04 1.01 to 1.07 .03 60%
Girls 1.09 1.03 to 1.1 .10 69%
Whites

Boys 1.05 1.02 to 1.08 .06 62%
Girls 1.09 1.03 to 1.1 .13 75%

In practical terms this resulted in 30% of re-
offenders being correctly identified with logistic
regression and 81% of non-recidivists being correctly
identified (see table 8). The overall correct
classification rate was 64%. African American boys had the
lowest correct classification rate with 60% correctly
identified. Girls had the highest correct classification
rate with 75% of offenders correctly identified. So while
the relationship between YLS/CMI score and recidivism had a
relatively small effect size, predictions based on these
relationships produced a wide range of correct

classification rates based on group membership. African

100

American boys were incorrectly classified at the greatest
rate with 47% of boys mis-identified based on YLS/CMI
score. Both girl groups were correctly classified over 65%
of the time.

Given the recommendation by methodologist that
Receiver Operating Characteristics and its corresponding
AUC coefficients better reflect correct classification
rates in a binary classification system by controlling for
base rate of re-offense, ROC curve analysis was also
conducted. The AUC refers to the sensitivity of the
instrument versus the specificity. Sensitivity is the
proportion of true positives; that it to say the number of
offenders predicted to re-offend that did in fact re-
offend. Specificity is the number of true negatives; that
is to say the number of offenders predicted to not re-
offend that did not in fact re-offend.

The YLS/CMI had an AUC of .641, p<.05 overall,
indicating it performed 14% better than chance in correctly
classifying offenders by risk score. The YLS/CMI produced
similar AUCs for each of the subgroups (Girls, AUC=.68;
Boys, AUC=.61; White, AUC=.66; Black, AUC=.63) (see table
7).

Furthermore, there was a significant difference in the

distribution of re-offenders by risk level (X‘2=54.9,

101

p<.05) with high risk offenders being more than twice as
likely to re-offend as low risk offenders. When examining
distribution of re-offenders by risk level for each of the
gender and racial groups, there was positive relationship
between offense level and offense rate. This can be seen in
Table 7 with recidivism rates becoming increasingly higher
as risk level increases for each of the gender and racial
subgroups.

Question 2. Are there differences in the risk-recidivism
relationship between these understudied groups and the
traditional sample?

Due to the small overall effect and sample size
constraints, differences between groups were not able to be
detected using logistic regression. Furthermore, an
examination of the relationship between cumulative risk
score and re-offense did not reveal any significant
differences in AUC for the gender groups (Girls, AUC=.68;
Boys, AUC=.61) or racial groups (White, AUC=.66; Black,
AUC=.63). Despite non-significant differences in over-all
classification rates, there were substantial practical
differences in the distribution of correctly identified
recidivists versus non-recidivists. There was a significant
difference in true negatives between White girls, White

boys and African American boys. Using a binomial proportion

102

difference test, we examined whether these proportions were
significantly different. There was a significant difference
between true positives for African American boys and each
of the other groups as reflected in the overall
classification rates found in table 8.

There were several significant differences within risk
levels by subgroup as shown in Table 7 [Chi-Square
significant difference test coefficients are reported at
the bottom of Table 7]. For instance, White low risk boys
and African American low risk girls were twice as likely to
re-offend as White low risk females. Moreover, African
American low risk boys were four times as likely to re-
offend as White low risk girls. In turn, low risk boys
exhibited greater re-offense rates than girls overall. When
examining subgroup difference in recidivism within the
moderate risk level, similar differences are apparent.
African American moderate risk male youth had substantially
higher rates of re-offense than the other subgroups with
nearly 61% re-offending, while White moderate risk female
youth had the lowest recidivism rate of that risk level
(34%), which was even lower than that of Black low risk
male youth. It should be noted however that there were no
significant differences in recidivism rates between Black

female youth and White female youth. When comparisons were

103

made of high risk subgroups, there are no identifiable
differences in recidivism rates by race or gender.
Generally, two-thirds of high risk offenders recidivated
over the two year period.

Recidivism-Subscale Relationship

Given the differences in re-offense rates within risk
levels by racial and gender subgroups, further analysis was
performed examining the correct classification rates for
each of the subscales by gender and racial groups. Table 6
shows that White female youth had five subscales [family,
education, peer, personality, attitudes] with AUCs that
were significantly better than chance in correctly
classifying non-recidivists and recidivists. White male
youth in turn had six subscales where this was the case
[offense history, education, peer group, drug involvement,
free time, and personality]. Black female youth had the
fewest subscales that predicted re-offense better than
chance with family, education, and personality subscales
meeting this criteria. Blackmale youth in turn had four
subscales [education, peer, substance abuse/use, and

attitude].

104

Chapter 8

Discussion

Juvenile justice systems have a vested interest in
correctly predicting which young offenders are most likely
to come in repeated contact with the legal system. There is
considerable evidence that repeat offenders commit the
majority of offenses, despite comprising a numerical
minority of all offenders. Thus treating all offenders as
though they have a similar likelihood of recidivating can
quickly exhaust system resources rehabilitating and
incapacitating the majority of young offenders who are not
likely to re—offend regardless of system response to their
first offense. This issue compounded by differential
offending rates amongst racial and gender groups that some
paradigms attribute to differential prosecution and
policing. Thus the current practice of formal risk
assessment is highly recommended because in theory, 1) risk
assessments correctly predict recidivist based on risk
level, and 2) risk assessments can help reduce
disproportionate minority contact through assessing risk
based on criminogenic dynamic risk factors.

The problem with this approach lies in the fact that

few risk assessments have been validated across racial and

105

gender groups and these instruments very well may suffer
from poor dimensional, resulting in over and under-
prediction of re-offense depending on the subgroup. This is
possible due to the inclusion and exclusion of risk factors
that are relevant for some groups and irrelevant for other
groups. For instance, no risk prediction instruments check
for risk due to surveillance level and zero—tolerance
policing policies that exist in some neighborhoods as
opposed to others. As race and class stratification
continue to permeate U.S. neighborhoods, it can be
postulated that the re-offense rates of Black males for
instance should be under-predicted by risk assessments due
to the exclusion of macro-level policing factors on risk
assessments. Moreover, while risk assessments are supposed
to solicit objective and verifiable information on youth,
there is still some degree of judgment and latitude that
assessors have in scoring youth on the instruments. Thus,
if some theories are correct biases can still conceivably
trickle into the scoring on these measures. Thus it is
highly recommended by proponents of the risk assessment
model of delinquency prevention that researchers validate
these instruments on a wide cross—section of juveniles,

especially with regards to African Americans and girls.

106

In that context this study offered some examination of
the dimensional identity of the YLS/CMI, a relatively
popular risk assessment measure with juveniles. In keeping
with other findings reported in the literature on this
instrument, this study found the YLS/CMI does moderately
well in identifying young re-offenders over a two year
period. Moreover, as risk level increases, so do actual re—
offense rates. High risk offenders, re-offended at twice
the rate of low risk offenders. This result was also
encouraging because there were three times as many low risk
juveniles as high risk juveniles, allowing for the
possibility to divert or refract nearly 45% of the offender
population based on risk level. This would reduce caseload
sizes and free up programming resources for moderate and
high risk offenders. The risk prediction was shown to be
dependable over-all by the ROC analysis, which indicated
that the YLS/CMI was 14% better than chance at classifying
youth. While the amount of variation in offense rates
explained by the YLS/CMI was relatively small, the over-all
correct classification rate was over 60%. It can be
concluded that in examining the YLS/CMI's performance with
the over-all offender population, it is a powerful
instrument deserving of its popularity with justice

officials. However, this over-all conclusion says nothing

107

for its performance when parceling out subgroups that may
have had different risk-recidivism relationships. There are
then two questions that immediately stem from this original
analysis. Does the YLS/CMI predict re-offense with
subgroups like African Americans and girls? Are their
differences in these risk prediction relationships? While
the questions appear similar, we see that it is possible
for the instrument predict re-offense for these groups, but
have differences with regards to the degree for which those
predictions are correct. Moreover, the risk factors that
contribute to those classifications may differ by subgroup
with regard to predictive validity.

Question 1. HOW well does the YLS/GMT predict recidivism
for.African Americans and girls?

When examining these predictions by race and gender,
we see that each group had a significant relationship
between risk and recidivism. Recidivists for both girls and
boys of both racial groups exhibited significantly higher
YLS/CMI scores than non-recidivists. The amount of
variation in offending explained by YLS/CMI score ranged
from 3% for African American boys to 13% for White girls.
While this effect size was clearly small for each subgroup,
the correct classification rate of offenders was still

greater than 50% for each subgroup. Despite the small

108

effect size for African American boys, 60% of these
offenders were correctly classified as non-re-offenders or
re—offenders. The ROC analysis for each subgroup also
indicated the YLS/CMI was a good predictor of re-offense
for each group, with AUCs ranging from .61 with boys and
.68 for girls. Finally, for each group, re-offense rates
significantly and substantially increased by risk level.

Again, these findings support the YLS/CMI as a
predictor of recidivism with young offenders. However, it
is also clear from considering the results that were
possible differences in the validity coefficients for each
subgroup. Thus, if case management decisions are made by
risk score or risk level differences in rates of
delinquency can still exist, which brings us to the second
question of this study.

Question 2. Are there differences in the risk-recidivism
relationship between African Americans, girls and the
traditional sample?

In examining differences between the various groups,
the first round of differences emerges from scoring on the
instrument itself. Girls on averaged scored two points
lower on the YLS/CMI than boys. There were no significant
differences in scoring based on race. The item by item

frequency analysis revealed that were a substantial number

109

of YLS/CMI items that boys were more likely exhibit risk on
than girls. With the exception of drug involvement and
attitudes, the variation in scores by subgroup did not
aggregate in overall subscales.

These small differences in item by item exhibited risk
did result in larger proportions of some groups being rated
as high risk. More than twice as many African American boys
were classified as high risk than African American girls.
However, there were no significant differences in the
distribution of low risk offenders by subgroups. Moreover,
at least four in ten young offenders were considered low
risk. Thus, each group has nearly half of its population
which could be eligible to be refracted or diverted from
the system.

There were further differences in the recidivism rates
by subgroup. Nearly twenty percent more boys than girls re-
offended. African American males had the highest re-offense
rate, while White females had the lowest re-offense rates.
The literature suggests that the YLS/CMI typically has a
correlation coefficient of .20 to .30. At the .05 alpha
level, to have a 60% chance of detecting a difference for
correlations between group, this would require sample sizes
for each group in excess of 1,300 assessed youth.

Therefore, while a regression was conducted to answer the

110

first question, the effect size produced by this
regression was not used in determining whether differences
between groups exist. Thus due to sample size constraints,
significant differences could not be detected using
regression, differences in the effect sizes which ranged
from 3% to 13% were not found to be significant. However,
when examining the correct classification rates by subgroup
using proportion difference test, significant differences
were detected between African American boys, White boys and
the two girl groups. The YLS/CMI performed much better with
girls than with boys. This may be due to the low recidivism
rate of girls and corresponding large group of offenders
classified as low risk. While both boy groups had a larger
re—offense rate but similarly sized low risk groups,
leading to a larger false negative rate for boys than
girls. This meant that the YLS/CMI under-predicted re-
offense for African American boys in particular.
Furthermore, compared to the other groups the YLS/CMI over-
predicted re-offending for White girls.

The YLS/CMI under-predicted recidivism for Black low
risk male youth, while over—predicting recidivism for White
low risk female youth. This trend of over and under
prediction for these groups was also true of moderate risk

groups. Only high risk youth had similar rates of re-

111

offense regardless of racial or gender group. It should be
noted however that within each subgroup, re-offense rates
incrementally increased by risk level suggesting that
weighting scores by race and gender grouping could improve
the dimensional identity of the YLS/CMI.

Given these findings, further analysis of the
predictive validity of each subscale by subgroups revealed
that while the subgroups scored similarly on the respective
subscales, there were differences between groups in correct
classification rates by subscale. The majority of YLS/CMI
subscales could individually predict re-offense for White
male youth whereas as few as three subscales on the
instrument individually predicted re-offense for Black
female youth. In turn, education was the single
criminogenic risk factor that could predict re-offense for
each of the four subgroups, while offense history only had
predictive validity with White male youth. Both subgroups
of girls demonstrated a unique predictive risk relationship
for family circumstances and personality.

These different predictive relationships between
criminogenic factors and recidivism by race and gender
suggest the necessity of thoroughly examining the factor
structure of the YLS/CMI by group and comparing alternative

risk models. Moreover, for some groups there is a strong

112

possibility that relevant risk factors are being excluded,
which merely weighting YLS/CMI scores would fail to
address.
Study Limitations

This study did not control for socio-economic status
and other macro—level variables that mitigate and
exacerbate delinquency risk. Moreover, this study only
examined the risk-recidivism relationship for African
American and White young offenders. While risk assessment
data was available for Hispanic youth, sample size
constraints did not allow for validating the instrument
with this group or other ethnic and racial groups.
Future Study

A future direction of research should include
accounting for the multi-level nature of criminogenic risk
by accounting for possible differences in structural
factors that naturally increase likelihood of identified
recidivism like policing levels, surveillance, and criminal
justice policy. By conducting hierarchical linear modeling
with neighborhood variables this could be accomplished.
Moreover clustering youth across the subscales by subgroup
can also identify differences in patterns of risk that
could explain differing recidivism rates by group within

risk levels. Research in this regard will reduce over and

113

under-predictions of criminal activity for female and

African-American subgroups.

114

Chapter 9

Conclusions

The YLS/CMI is a powerful instrument capable of
classifying youth based on likelihood of re-offense. The
instrument has both dynamic and static risk factors which
are generally accepted as strong predictors of re-offense.
Moreover, in classifying youth the instrument allows for
identifying youth that would benefit from diversion or
being completely refracted from the system. In making these
classification nearly half of the offender population can
be diverted. Using this classification system benefits the
juvenile justice system because it can result in lower
case—load sizes, freeing up of resources that can be better
used with moderate and high risk offenders, and provides
specific criminogenic areas in a youth life that can be
targeted with interventions leading better program efficacy
and program planning. Consequently, this study further
evidences why juvenile justice officials are so excited
about the potential of risk assessments as a tool of
delinquency prevention and rehabilitation.

That said, justice officials are also excited about
risk assessments potential to reduce disparities in

formally adjudicated crime due to system biases against

115

certain racial and gender groups. This study, however,
shows that such excitement is premature. While there is a
general risk-recidivism trend for African Americans,
Whites, boys and girls where recidivism rates increase as
risk level increases; the degree to which those increases
occur differ by gender and racial group. Girls are less
likely to offend and are less likely to be classified as
high risk offenders. However, general trends in juvenile
justice could mitigate the benefits of risk assessment with
this group by widening the adjudication net for girls. For
example, by expanding the use of truancy court and
targeting status offenses a larger proportion of girls will
fall under the pUrview of the court for first time and
minor offenses. Thus, increasing the over—all pool of one-
time offenders as compared to that of boys. The YLS/CMI in
that case over-predicts offending in this group in that the
low risk group of offenders is actually different than the
low risk group of offenders for boys. One can then question
whether there are any iatrogenic effects in exposing girls
to moderate and high risk offenders in the system.
Furthermore, this study indicated the YLS/CMI under-
predicted re-offense for boys, especially African American
boys. African American low risk boys were comparable in

proportion to other low risk groups; however they re-

116

offended at a substantially higher rate than the other
groups. One can conclude one of two things from this trend.
Either African American boys truly offend more than other
groups or they are under greater surveillance than other
groups due to first time offending putting them under the
purview of the juvenile court. This study did not attempt
to determine differences in actual deviance rates and
formally prosecuted delinquency rates. Instead it
determined whether the youth exhibited similar risk levels
across racial and gender groups. The results indicated that
African American boys did in fact exhibit similar levels of
low criminogenic risk.

In exhibiting similar risk but dissimilar offending
rates, item by item analysis and subscale ROC analyses,
indicated that there were substantial differences in how
youth were assigned to the respective risk levels. Boys
quite simply scored higher on the YLS/CMI items than girls.
Moreover, each subgroup differed on the set of subscales
that could individually predict re-offense for their
subgroup. This suggests the possibility that a number of
factors were included for some groups that were not
relevant. Further, several factors were excluded for some
that were relevant. We can conclude than that simply

weighting the YLS/CMI by race or gender would not improve

117

case—management or program planning due to
incomplete/faulty information derived from the YLS/CMI.
Instead the justice officials would benefit from tracking
the pathway from deviance to formally adjudicated
delinquency. Differences in those pathways to contact with
law enforcement can have considerable impact on the rates
of “crime” and exhibited risk. Moreover, the juvenile
justice system would also benefit from expanding assessment
beyond micro-level and individual based risk factors to
macro-level and system based risk factors, which contribute
offense rates. This can be accomplished by examining

neighborhood and school effects in multi-level analysis.

118

Literature Cited

Akers, R. (2000). Criminological Theories: Introduction,
Evaluation, and Application. Los Angeles: Roxbury.

Albonetti, C. A. (1987). Prosecutorial discretion: The
effects of uncertainty. Law & Society Review, 21, 291—
313.

Andrews, D., Kiessling, J., Robinson, D.,& Mickus, S.
(1986). The risk principle of case classification: An
outcome evaluation with young adult probationers.
Canadian Journal of Criminology, 28, 377-384.

Archwamety, T., & Katsiyannis, A. (1998). Factors related
to recidivism among delinquent females at a state
correctional facility. Journal of Child and Family
Studies, 7, 59-67.

Bandura, A., & McDonald, F. (1963). The influence of social
reinforcement and the behavior of models in shaping
children’s moral judgments. Journal of Abnormal and
Social Psychology, 67, 274.

Barnoski, R. Assessing Risk for Re-offense: Validating the
Washington State Juvenile Court Assessment. Olympia:
Washington State Institute for Public Policy, 2004.

Belknap, J. (1996). The Invisible Woman: Gender, Crime, and
Justice. Cincinnati: Wadsworth.

Bonta, J. (1996). Risk-needs assessment and treatment. In
A. T. Harland (Ed.), Choosing correctional options
that work: Defining the demand and evaluating the
supply (pp. 18-32). Thousand Oaks, CA: Sage.

Braithwaite, John. (1989). Crime, Shame, and Reintegration.
Cambridge: Cambridge University Press.

Bronfenbrenner, U. (1979). The Ecology of Human
Development: Experiments by Nature and Design.
Cambridge, MA: Harvard University Press.

Burgess, E.W. (1928.) “Factors Determining Success or
Failure on Parole.” In A.A. Bruce (ed.). The Working
of the Indeterminate Sentence Law and the Parole
System in Illinois (pp. 205—249). Springfield, IL:

119

Illinois (xmmdttee (Mi Indeterminate-Sentence Ixnv and
Parole.

Burke, R.(2001) An Introduction to Criminological Theory.
Willan Publishing, Devon.

Catchpole, R. E. & Gretton, H. M. (2003). The predictive
validity of risk assessment with violent young
offenders: A 1-year examination of criminal outcome.
Criminal Justice and Behavior, 30(6), 688—708.

Chesney-Lind, M. & Shelden, R. G (2004). Girls,
Delinquency and Juvenile Justice (3rd ed.).

Belmont,CA: Wadsworth.

Bureau of Justice Statistics. Sourcebook Criminal Justice
Statistics, 2002.30th Ed. (Pub NCJ 203301) Washington
DC: BJS, 2004. Available at www.albany.edu/sourcebook.

 

Cottle, C.C., Lee, R.J., & Heilbrun, K. (2001). The
prediction of criminal recidivism in juveniles.
Criminal Justice and Behavior, 28, 367-394.

Courtwright, D. T. (1996). The drug war’s hidden toll.
Issues in Science and Technology, 73.

Cullen, F. T., & Agnew, R. (2006). Criminological theory
past to present: essential readings. (Third edition).
LA: Roxbury Publishing Company.

DeComo, R. E., Wagner, D., & Baird, S. C. (1994).
validation of the Oregon risk assessment tools used
for classification of offenders on supervision. San
Francisco: National Council on Crime and Delinquency.

Dembo, R., Schmeidler, J., Nini-Gough, 8., Sue, C. C.,
Borden, P.,&Manning, D. (1998). Predictors of
recidivism to a juvenile assessment centerzAthree year
study. Journal of Child and Adolescent Substance
Abuse, 7, 57-77.

Duncan, R., Kennedy,W., & Patrick, C. (1995). Four-factor
model of recidivism in male juvenile offenders.
Journal of Clinical Child Psychology: 24, 250—257.

Eck, J. E., & Weisburd, D. (1996). Crime in Place. Monsey:
Criminal Justice Press.

120

Elliot, D., Huizinga, D., & Ageton, S. (1985). Explaining
delinquency and drug use. Beverly Hills. CA: Sage.

Erikson, E. H. (1963). Childhood and society. Norton 2d.ed.

Fellner, J., & Mauer, M. (October 1998). “Losing the Vote:
The Impact of Felony Disenfranchisement Laws in the
United States," Human Rights Watch and The Sentencing
Project.

Flores, A., Travis, L. F., and Latessa, E. J. (2004). Case
Classification for Juvenile Corrections: An
assessement of the YLS/CMI. NCJRS Report.

Frank, S. J., Paul, J. 8., Marks, M., and Van Egeren, L.
(2000). JOurnal American Academy of Child Adolescent
Psychiatry, 39(10), 1300-1308.

Freudenberg, N. (2002). Adverse effects of US jail and
prison policies on the health and well—being of women
of color. American Journal of Public Health, 92, 1895—
1899.

Funk, S. (1999). Risk assessment for juveniles on
Probation: a focus on gender. Criminal Justice and
Behavior, 26, 1, 44-68.

Gottfredson, S. D., & Moriarity, L. J. (2006). Statistical
risk assessment: Old problems and new applications.
Crime & Delinquency, 52, 1, 178-200.

Grinberg, I., Dawkins, M., Dawkins, M. P., and Fullilove,
C. (2005). Adolescents at risk for violence: an
initial validation of the Life Challenges
Questionnaire and Risk Assessment Index. Adolescence
,40, 159.

Grove, W. M. and Meehl, P. E. (1996). Comparative
efficiency of informal (subjective, impressionistic)
and formal (mechanical algorithmic) prediction
procedures: The Clinical-Statistical Controversy.
Psychology, Public Policy, & Law. 2(2):293-323.

Guevara, L., Herz, D., & Spohn, C. (2006). Gender and

juvenile justice decision making: What role does race
play? Feminist Criminology, l, 4, 258 — 282.

121

Harlow CW. Profile of Jail Inmates—1996. Washington, DC: US
Dept of Justice; 1998:1-15. Bureau of Justice
Statistics special report NCJ 164620.

Harris, P. (2006). What Community Supervision Officers Need
to Know About Actuarial Risk Assessment and Clinical
Judgment. Federal Probation, 69 (2), 8—14.

Richard D Hartley, Sean Maddan, Cassia C Spohn. (2007).
Prosecutorial Discretion: An Examination of
Substantial Assistance Departures in Federal Crack-
Cocaine and Powder-Cocaine Cases. Justice Quarterly
JQ, 24(3), 382-407.

Heider, F. (1958). The psychology of inter-personal
relations. New York: Wiley.

Hishinuma, E. S., Johnson, R. C., Kim, P. S., Nishimura, S.
T., Makini, G. K., Andrade, N. N., Yates, A., Goebert,
D. A., Mark, G. Y., Mayeda, D. T., and Revilla, L. A.
(2005). Prevalence and correlates of misconduct among
ethnically diverse adolescents of Native
Hawaiian/Part—Hawaiian and Non-Hawaiian Ancestry.
International Journal of Social Psychiatry 51(3), 242-
258.

Hodges, K., and Kim, C. (2000). Psychometic study of the
child and adolescent functional assessment scale:
Prediction of contact with the law and poor school
attendance. Journal of Abnormal Child Psychology,
28,3, 287-297.

Howell, J. C. (2003). Preventing & reducing juvenile
delinquency: A comprehensive framework. Thousand Oaks,
CA: Sage.

Hubbard, D. J. & Pratt, T. (2002). A meta-analysis of the
predictors of delinquency amongst girls. Journal of
Offender Rehabilitation, vol. 34(3), 2002. Pp. 1—13.

Huff, D., and Prell, L. The Proposed Iowa Juvenile Court
Intake Risk Assessment. Division of Criminal and
Juvenile Justice Planning and Statistical Analysis
Center.

Jackson, P. 1., & Carroll, L. (1981). Race and the war on
crime: The socio-political determinants of municipal

122

police expenditures in 90 non-Southern US cities.
American Sociological Review, 46, 290-305.

Jensen, Gary P. 2003. Social Disorganization Theory.
Encyclopedia of Criminology. Richard A. Wright
(Editor). Fitzroy Dearborn Publishers.

Jimerson, S. R., Sharkey, J. D., O’brien, K., Furlong, M.
(2004). The Santa Barbara Assets and Risks Assessment
to predict recidivism among male and female juveniles:
An investigation of inter-rater reliability and
predictive validity. Education and Treatment of
Children 27, 4, 353-373.

Jung, S., and Rawana, E. (1999). Risk and Need Assessment
of Juvenile Offenders. criminal JUstice and Behavior,
26, 1, 69-89.

Kent, S.L., & Jacobs, D. (2005). Minority threat and police
strength from 1980 to 2000: A fixed-effects analysis
of nonlinear and interactive effects in large cities.
Criminology, 43, 731-760.

Kerstetter, W. (1990). Gateway to justice: Police and
prosecutorial response to sexual assaults against
women. Criminology, 81, 267-313.

Kliewer,W., Fearnow, M. D.,&Walton, M. N. (1998).
Dispositional, environmental, and context-specific
predictors of children’s threat perceptions in
everyday stressful situations. JOurnal of Youth and
Adolescence, 27, 83-100.

Kliewer, W., & Kung, F. (1998). Family moderators of the
relation between hassles and behavior problems in
inner-city youth. Journal of Clinical Child
Psychology, 27, 278-292.

Kliewer,W., Lepore, S. J., Oskin, D., & Johnson, P. D.
(1998). The role of social and cognitive processes in
children’s adjustment to community violence. Jburnal
of consulting and Clinical Psychology, 66, 199-209.

Kraska, Peter B. (2006). Criminal Justice Theory: Toward
Legitimacy and an Infrastructure. JUstice Quarterly,
23 (2), 0741-8825. Retrieved September 13, 2007, from
http://www.informaworld.com/10.1080/07418820600688735

123

Krisberg, B. & Austin, J. (1978) The children of Ishmael.
Palo Alto, CA: .Mayfield.

Krysik, J. and Lecroy, C. W. (2002). The empirical
validation of an instrument to predict risk of
recidivism among juvenile offenders. Research on
Social Wbrk Practice, 12(1), 71-81.

Lahey, B. B., Goodman, S. H.,Waldman, I. D., Bird, H.,
Canino, G., Jensen, P., et a1. (1999). Relationship of
age of onset to the type and severity of child and
adolescent conduct problems. Jburnal of Abnormal Child
Psychology, 27, 247-260.

Leiber, M. J. & Fox., K. C. (2005). Race and the impact of
detention on juvenile justice decision making. Crime
Delinquency, 51, 470 — 497.

Lipsey, M.W. (2005). What works with juvenile offenders:
Translating research into practice. Presented at the
Adolescent Treatment Issues Conference, Tampa, FL.

Liska, A. (1987). Perspectives on Deviance, 2nd ed.,
Prentice-Hall.

Loeber, R., & Dishion, T. (1983). Early predictors of male
delinquency: A Review. Psychological Bulletin, 93, 68-
99.

Loeber, R., & Farrington, D. P. (Eds.). (1999). Serious and
violent juvenile offenders. London: Sage.

Loftin, C., & McDowall, D. (1982). The police, crime, and
economic theory: An assessment. American Sociological
Review, 47, 393-401.

Lowenkamp, C. T., Holsinger, A., & Latessa, E. (2001).
Risk/need assessment, offender classification, and the
role of childhood abuse. Criminal JUstice and
Behavior, 28(5), 543-563.

Maguire, E. R. (2001). Research evidence on the factors
influencing police strength in the United States. In
Hiring and retention Issues in Police Agencies:
Readings on the determinants of police strength,
hiring and retention of officers and the federal cops
program (pp. 7-25). A report to the National Institute

124

of Justice by Christopher S. Koper, Edward R. Maguire,
and Gretchen E. Moore. Retrieved from http://
www.urban.org/pdfs/Hiring—and—Retention.pdf

 

Marczyk, G. R., Heilbrun, K., Lander, T., and DeMatteo, D.
(2003). Predicting juvenile recidivism with the
PCL:YV, MAYSI, and YLS/CMI. International Jburnal of
Forensic.Mental Health. 2(1), 7-18.

Mauer, M. (1999). Race to incarcerate. New York: New Press.

Maxfield, M.G., and C.S. Widom, “The Cycle of Violence:
Revisited Six Years Later,” Archives of Pediatrics and
Adolescent Mbdicine 150 (1996): 390-395.

May, L. & Sharratt, S. (1994). Applied Ethics: A
.Multicultural Approach, Englewood Cliffs, NJ:
Prentice-Hall.

McKinnon, L. J. (2004). Predicting risk of violence in a
young offender population: The predictive validity of
the PCL:YV and the YLS/CMI. Dissertation:Lakehead
university (canada), 54.

Mcreynolds, L. S. (2005). Prediction of juvenile
recidivism: Influence of gender, psychiatric disorder,
and neighborhood disadvantage.Dissertation Abstracts
International Section A: Humanities and Social
Sciences Vol 66(1-A): 358.

Moffit, T. E. (1993). Adolescent-limited and life-course-
persistent antisocial behavior: A developmental
taxonomy. Psychological Review, 100, 674-701.

Myers, H. F., Taylor, S., Alvy, K. T., Arrington, A., &
Richardson, M. A. (1992).Parental and family
predictors of behavior problems in inner-city Black
children. American Jburnal of Community Psychology,
20, 557-576.

Myner, J., Santman, J., Cappelletty, G.,& Perlmutter, B.
(1998).Variables related to recidivism among juvenile
offenders. International JOurnal of Offender Therapy
and Cbmparative Criminology, 42, 65-80.

The National Center on Addiction and Substance Abuse (CASA)
at Columbia University. (2004, October). Criminal

125

neglect: substance abuse, juvenile justice, and the
children left behind. New York: The National Center on
Addiction and Substance Abuse (CASA at Columbia
University). Retrieved May 15, 2006 from
http://www.nicic.org/Library/020178

 

National Council on Crime and Delinquency. (2001). Alameda
County Placement Risk Assessment Validation. Executive
Summary.

Nowicka, M. (2005). The Level of Service Inventory-Ontario
Revision: A recidivism follow-up study within a sample
of male young offenders. Dissertation Abstracts
International: Section B: The Sciences and
Engineering, 66(5-B),2835.

Nunnally, J. C. (1978). Psychometric Theory. Mcgraw-Hill.
NY.

Office of Juvenile Justice and Delinquency Prevention
(2007). “OJJDP DMC, Reduction Best Practices
Databases”. Retrieved from:
http://mpg.dsgonline.com/dmc default.aspx

 

Onifade, E., Livsey, S., Davidson, W., Turke, G., Campbell,
C., Mackey, M., & Malinowsky, J. (2007). Predicting
the Who, What and When of Recidivism with the Ybuth
Level of Service case Management Inventory. Paper
submitted for publication

Patterson, G. R., Debaryshe, B. D, & Ramsey, E. (1989). A
developmental perspective on antisocial behavior.
American Psychologist, 44, 329-335.

Phillips 8, Bloom B. In whose best interests? The impact of
changing public policy on relatives caring for
children with incarcerated parents. Child Welfare.
1998;77:531-541.

Pope, C. and W. Feyerherm. (1992)..Minorities and the
Juvenile JUstice System. Rockville, MD: U.S.
Department of Justice, Office of Juvenile Justice and
Delinquency Prevention, Juvenile Justice
Clearinghouse.

Quist, R. M. and Matshazi, D. G. (2000). The Child and
Adolescent Functional Assessment Scale (CAFAS): A

126

dynamic predictor of juvenile recidivism. Adolescence,
35(137), 181-192.

Rappaport, J. (1986). In praise of paradox. In E. Seidman
and J. Rappaport. (eds) Redefining social problems.
New York: Plenum Press

Richie BE. Challenges incarcerated women face as they
return to their communities: findings from life
history interviews. crime Delinquency. 2001;47:368-
389.

Risler, E., Sutphen, R., & Shields, J. (2000). Preliminary
validation of the juvenile first offender risk
assessment index. Research on Social Wbrk Practice,10,
1, 111-126.

Schmidt, F., Hoge, R. D., and Gomes, L. (2005). Reliability
and validity analyses of the youth level of
service/case management inventory. Criminal Jastice
and Behavior 32, 3, 329-344.

Schrantz, D., & McElroy, J. (2000). Reducing Racial
Disparity in Criminal JDstice: A Manual for
Practitioners and Policmakers .Sentencing Project.
Washington, D.C.

Schumacher. M., & Kurz-Gwen, A. (2000). The 8% solution:
preventing serious, repeat juvenile crime. Thousand
Oaks: Sage Publications.

Schwalbe, C.S. (2007). Risk Assessment for juvenile
justice: A meta-analysis. Law and HUman Behavior, 31,
449-462.

Schwalbe, C. S., Fraser, M. W., Day, S. H., and Cooley, V.
(2006). Classifying juvenile offenders according to
risk of recidivism:Predictive validity,
race/ethnicity, and gender. Criminal JUstice and
Behavior, 33(3), 305-324.

Schwartz, 1. M., Steketee, M. W. & Schneider,V. W. (1989).
The Incarceration of Girls: Paternalism or Crime
Control? Ann Arbor, MI: Center for the Study of Youth
Policy

127

Seidman, E. (1987). Toward a framework for primary
prevention research. In J.A. Steinberg & M.M.
Silverman (Eds.), Preventing mental disorders: A
research perspective (pp. 2-19). Washington, DC: U.S.
Government Printing Office.

Sever, B. (2003). The minority population/police strength
relationship: Exploring past research. Criminal
JUstice Studies, 16, 153—171.

Sharkey, J. D., Furlong, M. J., Jimerson, S. R., and
O’brien, K. O. (2003). Evaluating the utility of a
risk assessment to predict recidivism among male and
female adolescents. Education and Treatment of
Children, 26, 4, 467-494.

Sickmund, M (2004). Juveniles in corrections. JUvenile
Offenders and Victims, National Report Series Bulletin
(June).

Siegel, D. J. (1999). The Developing.Mind: waa d.A
Neurobiology of Interpersonal Experience.1-22, 276-82.

Simourd, L., & Andrews, D. (1994). Correlations or
delinquency: A look at gender differences. Forum on
Corrections Research, 6, 26-31.

Smith, W. R. and Aloisi, M. F. (1999). Prediction of
recidivism among second timers in the juvenile justice
system: Efficiency in screening chronic offenders.
American JOurnal of Criminal JUstice, 23(2), 201-222.

Snyder, H. N., & Sickmund, M. (2006). Juvenile Offenders
and Victims: 2006 National Report. Washington, DC:
U.S. Department of Justice, Office of Justice
Programs, Office of Juvenile Justice and Delinquency
Prevention.

Steffensmeier, D. J., & Steffensmeier, R. (1980). Trends in
female delinquency: An examination of arrest, juvenile
court, self-report, and field Data.” criminology 18:
62-85.

Streib, V. 1987. Death Penalty for JUveniles. Bloomington,
IN: Indiana University Press.

128

Stucky, T.D. (2005). Local politics and police strength.
Justice Quarterly, 22, 139-169.

Faye Taxman, James M Byrne, April Pattavina. (2005).
Racial Disparity and the Legitimacy of the Criminal
Justice System: Exploring Consequences for
Deterrence. Journal of Health Care for the Poor and
Underserved: Reducing HIV/AIDS and Criminal Justice
Involvement in..., 16(4), 57-77.

Thomas D Stucky (2005). Local Politics and Police
Strength. JUstice Quarterly : JQ, 22(2), 139-169.

Torbert, P. (1996). Juvenile Probation: The Workhorse of
the Juvenile. JUstice System. Juvenile JUstice
Bulletin. Office of Juvenile Justice and Delinquency
Prevention

Turner, S., and Fain, T. (2003). Validation of the Los
Angeles County probation department’s risk and needs
assessment instruments. Rand Institute Report.

VanVoorhis, P., & Brown, K. (1997). Risk classification in
the 19903.Washington, DC: National Institute of
Corrections.

Von Eye, A., & Bergman, L. R. (2003). Research strategies
in developmental psychopathology: Dimensional identity
and the person-oriented approach. Development and
Psychopathology, 15, 553-580.

Walsh, A. (2000). Behavior genetics and anomie strain
theory. Criminology, 38(4).

Walters, G. D. (2006). Risk appraisal versus self report in
the prediction of criminal justice outcomes. A meta-
analysis. Criminal Justice and Behavior.33(3), 279-
304. '

R.L. Weisman, J.S. Lamberti, N. Price. (2004).
Integrating Criminal Justice, Community Healthcare,
and Support Services for Adults with Severe Mental
Disorders. Psychiatric Quarterly; 75(1), 71-85.

Wiebush, R.C., Baird, C., Krisberg, B., and Onek, D.

(1995). Risk assessment and classification for
serious, violent, and chronic juvenile offenders. In

129

J.C. Howell, B. Krisberg, J.D. Hawkins, and J.J.
Wilson (Eds.), A Sourcebook: Serious, Violent, and
Chronic Juvenile Offenders. Thousand Oaks, CA: Sage
Publications.

Wilson, J. Q. (1968). Varieties of police behavior: The
management of law and order in eight communities.
Cambridge, MA: Harvard University Press. '

Wolfgang, M., Figlio, R. M., and Sellin, T. (1972).
Delinquency in a birth cohort. Chicago. University of
Chicago Press.

Wordes, M., Bynum, T. S., & Corley, C. Locking Up Youth:
The Impact of Race on Detention Decisions. (1994)
JOurnal of Research in Crime and Delinquency, 31, 149-
165. '

130

 

hlrl Iii
i ill II.

 

fl)" I ll lull

3 1293 02956 6845