Automated Offender Risk Assessment

• Date: 01 February 2017
• Published date: 01 February 2017
• Author: J. Stephen Wormith
• DOI: http://doi.org/10.1111/1745-9133.12277

POLICY ESSAY

RECIDIVISM RISK ASSESSMENT

Automated Offender Risk Assessment

The Next Generation or a Black Hole?

J. Stephen Wormith

University of Saskatchewan

The offender risk assessment enterprise has been moving at warp speed since the

turn of the millennium and shows no signs of slowing. This is good. It reflects the

importance of offender risk assessment in the field of criminal justice, as well as the

energy, thought, and creativity that is being invested in efforts to build a better mousetrap.

It is a vibrant and robust area of research that has a direct application to correctional policy

and practice in numerous ways. Nevertheless, the development of offender risk assessment

began long before the many developments of this modern era. This review describes how

the recently constructed Minnesota Screening Tool Assessing Recidivism Risk (MnSTARR,

2.0), and its related research by Grant Duwe and Michael Rocque (2017, this issue) in

Minnesota, represent a culmination of this history and points out some questions that the

scale’s implementation has for policy and practice.

Offender Risk Assessment Generations and Other Variations

Bonta (1996) documented the early offender risk assessment innovations in what he de-

scribed as a series of progressions or “generations” beginning with traditional clinical judg-

ment (first generation), followed by the movement to actuarial assessments that were based

on static risk factors (second generation), and then by assessments of risk and (dynamic)

criminogenic need factors (third generation), which generate not only an assessment of

offender risk but also give direction for clinicians and correctional staff to pursue in their

work with offenders. By the turn of the century, the results of meta-analyses convincingly

established the superiority of mechanical (second and third generation) over subjective (first

generation) predictions of human behavior in numerous disciplines, but the differences

Disclosure: J. Stephen Wormith receives royalty payments for the Level of Service/Case Management

Inventory from its publisher, Multi-Health Systems, Toronto, Canada. Direct correspondence to J. Stephen

Wormith, Department of Psychology, 9 Campus Drive, University of Saskatchewan, Saskatoon SK, Canada, S7N

5A5 (e-mail: s.wormith@usask.ca).

DOI:10.1111/1745-9133.12277 C2017 American Society of Criminology 281

Criminology & Public Policy rVolume 16 rIssue 1

Policy Essay Recidivism Risk Assessment

were particularly strong in criminal justice and medicine (Grove, Zald, Lebow, Snitz, and

Nelson, 2000). Also by that time, the results of another important meta-analysis on offender

recidivism demonstrated that criminogenic needs, despite their greater difficulty in scoring,

were on average at least as predictive of offender recidivism as were the static (sometimes

called objective) items found in second-generation risk assessment instruments (Gendreau,

Little, and Goggin, 1996). With their capacity to assess risk and inform intervention, third-

generation instruments were considered by many researchers to be an improvement of their

more easily scored predecessors.

It has been suggested that a more recent fourth generation has emerged, one that

explicitly integrates the outcome of a risk–need assessment, and its many details, with treat-

ment planning, case management, and intervention with the offender (Andrews, Bonta,

and Wormith, 2006). The defining and sequencing of these generations provide a con-

venient way to organize and structure the evolution of offender risk assessment and offer

a jumping off point for the future. Nonetheless, they may also be viewed as simplistic,

ignoring the many variations to offender risk assessment that have emerged within or across

these generations. Some of the more prominent innovations are discussed in the following

sections.

Specificity of Offender Risk Assessment Tools. There has been a strong voice for the

development of instruments, or versions of instruments, that are particularly applicable to

demographically identified subsets of the offender population, especially women offenders

(Blanchette and Brown, 2006; Chesney-Lind and Pasko, 2013; Hannah-Moffat, 2009).

This lobby has led to a spirted debate, a variety of solutions, more research, and meta-

analyses (e.g., Andrews et al., 2012; Geraghty and Woodhams, 2015; Smith, Cullen, and

Latessa, 2009; Van Voorhis, Wright, Salisbury, and Bauman, 2010). Although no clear

winner has emerged, consensus does exist that demographics, particularly gender, should

be accommodated in some way, including gendered-based versions (i.e., items, norms, and

prediction formulae) of an otherwise common instrument. This approach has been adopted

by Duwe and Rocque (2017).

The same concern has led to a call for the development and use of local norms

for the “generic” instruments that are used widely by the criminal justice system on a

national or international scale in an effort to accommodate differing legal contexts, offender

populations, and cultures. This call has led to several responses, interesting enough, primarily

in the realm of sexual offender risk assessment (DeClue and Zavodny, 2014; Hanson,

Lunetta, Phenix, Neeley,and Epperson, 2014; McGrath, Hoke, and Lasher, 2013; Phenix,

Helmus, and Hanson, 2012; Wormith, 2014), but it also includes the MnSTARR (Duwe

and Rocque, 2017).

The concern about differing populations and legal contexts has also led to de-

velopment of a proliferation of instruments designed by and for specific correctional

agencies, typically a state or federal jurisdiction. For example, this includes the Admin-

istrative Office of the U.S. Courts with its Post Conviction Risk Assessment (PCRA:

282 Criminology & Public Policy