Predicting Juvenile Reentry Success: Developing a Global Risk Score and Risk Classification Levels Using the Residential Positive Achievement Change Tool

Published date01 July 2019
AuthorKevin T. Wolff,Michael T. Baglivio
DOI10.1177/1541204018804870
Date01 July 2019
Subject MatterArticles
Article
Predicting Juvenile Reentry
Success: Developing a Global
Risk Score and Risk
Classification Levels Using
the Residential Positive
Achievement Change Tool
Michael T. Baglivio
1
and Kevin T. Wolff
2
Abstract
Recent decades have seen the proliferation of risk assessment implementation and, subsequently,
validation studies. Most assessments are used at the front end, at arrest or postdisposition for
juvenile offenders. The current study develops an overall risk score and risk classification levels from
a tool developed specifically for use with youth within residential placement in efforts to predict
reentry success. A diverse, 4-year statewide sample of serious juvenile offenders (N¼15,078) is
examined. The total risk score and classification schematic development are outlined, predictive
validity assessed, and the ability of the classification to partition youth into meaningful subgroups of
overall risk to reoffend analyzed. Results show predictive validity on par with prominent juvenile
tools, which is noteworthy given the sample exclusivity: serious offenders in residential placement.
Further, the tool classifies youth into five distinct groups with meaningful dispersion across groups
with differing recidivism base rates. Implications for juvenile reentry are discussed.
Keywords
juvenile offenders, recidivism, risk assessment, reentry
A reported 45,567 youth were in juvenile justice residential placements in the United States on a
given day in 2016, according to the Office of Juvenile Justice and Delinquency Prevention (Puzzan-
chera, Hockenberry, Sladky, & Kang, 2018). Current best-practice standards dictate that only the
highest risk juvenile offenders be recommended for placement in such residential programs, and
only after community-based alternatives have been exhausted (Baglivio, Greenwald, & Russell,
2014; Wilson & Howell, 1993). The proliferation of risk assessments as structured decision-making
1
TrueCore Behavioral Solutions, LLC, Tampa, FL, USA
2
John Jay College of Criminal Justice, City University of New York, New York, NY, USA
Corresponding Author:
Michael T. Baglivio, Analytic Initiatives, LLC, 12344 Rangeland Blvd., Odessa, FL 33556, USA.
Email: mbaglivio@analyticinitiatives.com
Youth Violence and JuvenileJustice
2019, Vol. 17(3) 241-268
ªThe Author(s) 2018
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/1541204018804870
journals.sagepub.com/home/yvj
tools, and studies examining their predictive validity, aids in the advancement of this endeavor
(Schwalbe, 2007, 2008; Yang, Wong, & Coid, 2010). These reform efforts, centered on the now
dominant risk–need–responsivity model (Andrews & Bonta, 2003; Andrews, Zinger, Hoge, Bonta,
Gendreau, & Cullen, 1990; Lipsey, 2009; Lowencamp & Latessa, 2005), in conjunction with a
system of graduated sanctions (Howell, 2009) address how best to advantageously position offen-
ders and target resources to these highest risk youth.
However, far less work has examin ed the application of these the oretical (and practical) mod els
on the back end of the juvenile justice system, based on the likelihood of reoffending upon return
to the community of deep-end residential placement youth. Specifically, extensive work has
examined the risk principle (Andrews & Bonta, 2003; Andrews & Kiessling, 1980; Andrews
et al., 1990; Bonta, Wallace-Capretta, & Rooney, 2000; Lipsey, 2009; Lowenkamp, Latessa, &
Holsinger, 2006) for prioritizing resources and more restrictive placements to higher risk youth,
thereby elucidating the need for valid risk/need assessment. Recent work has even examined the
best level of placement for a given offense and risk profile within a juvenile justice service
continuum (Baglivio et al., 2014). However, there is a dearth of empirical work investigating
tools designed to assess risk to reoffend on the back end, for youth completing deep-end residen-
tial placements and beginning the reentry process.
One such tool, the Residential Positive Achievement Change Tool (R-PACT) developed specif-
ically for residential youth, is the subject of the current analysis. While prior work has leveraged
R-PACT items in examining juvenile recidivism, and one study has assessed the validity of the R-
PACT domain-level subscores, no study has yet examined the development of an overall risk to
reoffend classification for the tool. As the R-PACT is used statewide in at least one U.S. state, and in
several other jurisdictions across the nation, the development of an overall risk classification is
needed. The current study contributes to that discourse by examining the factors most related to
recidivism among serious juvenile offenders released from residential placements, and the validity
of an actuarial overall risk score developed to predict those negative outcomes and potentially guide
reentry policy.
This study has three central aims. First, we develop an actuarial tool from the items most
predictive of recidivism from the R-PACT risk/needs assessment administered to every youth in
residential placement across 4 years of releases in an entire diverse state. Second, we examine the
predictive validity of the developed total risk score and risk classifications (e.g., low risk, high risk).
Finally, we examine the potency of the tool by analyzing whether there is meaningful dispersion of
base rate recidivism and sample youth across risk classifications. Prior to examining these questions,
we provide a brief overview of juvenile risk assessments, as well as detail the prior research using
the R-PACT, and assessment of the predictive validity of its domain-level scores.
Juvenile Risk Assessments
The use of “generations” terminology to classify correctional and juvenile risk to reoffend assess-
ment instruments is now common lexicon (Andrews, Bonta, & Wormith, 2006), though much to the
disdain of some (Baird, 2009). First-generation tools refer to clinical or professional judgment,
second generation to actuarial tools composed of static factors, while third-generation tools incor-
porated dynamic/changeable “criminogenic needs,” and fourth-generation assessments added
responsivity factors and are distinguished by the results of the tool driving case planning. More
recently, nonregression base d, more sophisticated methodol ogies (such as neural networks and
machine learning) have been suggested as a fifth generation (Schaffer, Kelly, & Lieberman,
2011). Those having issue with the “generations” terminology caution the inherent assumption that
successive generations outperform prior generations; the verdict of which is empirically ambiguous
to date (see Baird, 2017; Hamilton, Neuilly, Lee, & Barnoski, 2015), with the exception being the
242 Youth Violence and Juvenile Justice 17(3)

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT