Professional Documents
Culture Documents
During the past 20 years, there have been significant developments in the area of offender
assessment. As a result, this knowledge has placed the field in a position to construct guidelines as to what should characterize useful and effective offender assessment instruments. The
authors suggestions as to what constitutes good assessment ranges from the noncontroversial
(e.g., actuarial instruments) to the more contentious (e.g., lessening ones reliance on static
risk scales). Whether the reader agrees with the views expressed, it is hoped that the force of
the empirical arguments will at least provoke some careful consideration rather than summarily
dismissing them.
or those who have to deal with offenders, either directly or indirectly, conducting assessments is simply part of their regular work
routine. Any intervention with an offender requires an assessment of
how the characteristics of the offender and the situation are related to a
relevant outcome. Within a correctional context, there are many outcomes of interest. A correctional officer, for example, may need to
judge whether a depressed inmate is suicidal. Parole board members
consider the likelihood of an inmate adjusting to life in the community. A therapist must assess an offenders progress in treatment. The
list can go on, but these examples are sufficient to demonstrate that
offender assessments are not only common activities but also that the
AUTHORS NOTE: The opinions expressed are those of the author and do not necessarily represent the views of the Ministry of the Solicitor General of Canada. Correspondence concerning this article should be addressed to James Bonta, Corrections
Research, Solicitor General Canada, 340 Laurier Ave. W., Ottawa, Ontario, Canada,
K1A 0P8; e-mail: Bontaj@sgc.gc.ca.
CRIMINAL JUSTICE AND BEHAVIOR, Vol. 29 No. 4, August 2002 355-379
2002 American Association for Correctional Psychology
355
356
Here we are in the 21st century, and one would think that we have
resolved the debate between clinical and actuarial approaches to risk
assessment. Judging from what goes on in the field, I do not think we
have. The results from two recent surveys are the source of my concern. Boothby and Clements (2000) asked 830 correctional psychologists what specific psychological tests they used in their work.
Twenty-three percent reported using the Bender-Gestalt, 20% used
the Rorshach, and 14% used projective drawings. Although there are
scoring systems for these tests, I do not think that they can be considered actuarial. Actuarial measures are structured, quantitative, and
empirically linked to a relevant criterion. To the best of my knowledge, there is no credible literature demonstrating that these tests predict criteria important to corrections (i.e., prison assaults, escape, suicide, and, of course, recidivism).
358
360
In corrections, there are numerous outcomes of importance. Correctional workers may be interested, for example, in whether the individual is likely to reoffend in the community, be a management problem in prison, suffer a psychotic breakdown, or attempt suicide. To my
knowledge, there is no single assessment instrument that can predict
all of these behaviors. On the other hand, I do not think that we need a
specific instrument for each criterion behavior that we are interested
in predicting.
There are two general classes of behaviors that are important to corrections: (a) rule violation (in a very general sense) and (b) psychological instability. Sometimes these behaviors are interrelated (e.g., paranoid delusions and assault on staff), but not always (e.g., severe
depression is unrelated to violent recidivism). For the most part, clinicians are trained to think that poor emotional and psychological health
will lead to deviant behavior. Consequently, recent clinical graduates
enter corrections believing that psychological maladjustment is a
strong correlate of criminal behavior. In addition, they believe alleviating anxiety and feelings of depression and increasing self-esteem
and so forth will help their clients follow a prosocial lifestyle. However, as I try to explain more fully under the next guideline, a focus on
psychological instability is not particularly helpful in reducing the
likelihood of criminal behavior. To make matters worse, the clinical
training received in many of our universities propagate the myth of
psychological disturbance as a correlate of criminal behavior by training psychologists in tests that measure psychological adjustment
(e.g., MMPI, Becks Depression Inventory, etc.).
Studies of the MMPI have not shown the instrument to be a particularly good predictor of rule-violating behaviors. Scale 4 or the old Pd
Scale, which was originally developed for the diagnosis of psychopathy, shows predictive correlation coefficients in the .16 to .21 range
(Gendreau, Little, & Goggin, 1996; Simourd, Bonta, Andrews, &
Hoge, 1991). These correlations were significant but did not exceed
the correlations found with the PCL-R or Goughs Socialization Scale
(r = .29 to .39). Even when the MMPI was subjected to modifications
intended to make it suitable for offender populations, the results have
362
364
tion of psychopaths has facilitated laboratory work on biological, cognitive, and other psychological areas related to the construct.
Somewhat related to the PCL-R is the Violence Risk Appraisal
Guide (VRAG) (Harris et al., 1993). The VRAG was not specifically
derived from theory, but I mention it here because the PCL-R forms
part of the scale. The VRAG, therefore, deserves some recognition as
being influenced by theory. In addition to the PCL-R, the VRAG consists of 11 items, with nearly all of them historical in nature (e.g., age
at index offense, elementary scholastic maladjustment). Some recent
studies of the VRAG have shown large predictive validity estimates
(see Quinsey et al., 1998, for a review). More will be said about the
VRAG later.
The LSI-R is the most theoretically based offender assessment
instrument available today. The instrument consists of 10 subcomponents
that tap the areas outlined in the theoretical model outlined by
Andrews and Bonta (1998a). The Big Four are represented in the
subcomponent areas of criminal history, companions, emotional/
personal, and attitudes/orientation. In addition, a number of lesser but
important factors supported in theory and research are addressed in
the other subcomponents of the LSI-R (e.g., employment/education,
family/marital). The predictive validity of the LSI-R is well established, and reviews can be found elsewhere (Andrews & Bonta, 1995,
1998a; Gendreau et al., 1996; Gendreau, Goggin, & Smith, 2002 [this
issue]). The theoretically based LSI-R has two other important characteristics that make it attractive to staff responsible for managing
offenders: (a) comprehensiveness and (b) the assessment of criminogenic
needs. These are discussed under the next two guidelines.
GUIDELINE 5: SAMPLE MULTIPLE DOMAINS
366
cant advantage over the other scales for the management of offender
risk. These advantages are evident in the next section.
GUIDELINE 6: ASSESS CRIMINOGENIC NEED FACTORS
368
370
Boothby and Clements (2000) found that 87% of correctional psychologists used the MMPI in their work. We do not know why the
MMPI was administeredit is hoped not for assessments of risk for
recidivism. Tests like the MMPI and other personality and cognitive
measures can, however, play a role in offender assessment. They may
be useful with respect to responsivity considerations. The responsivity
principle of offender classification states that the style and mode of
treatment must be matched to the cognitive, personality, and sociocultural characteristics of the individual (Andrews, Bonta, & Hoge,
1990; Bonta, 1995).
Classification instruments such as the I-Level and CM are examples of assessment tools that tap responsivity considerations. For
example, in the I-Level system, offenders are classified along stages
of interpersonal maturity, and interventions are matched to the stage
of development. Thus, offenders in the I-3 stage (manipulator) need to
be dealt with in a way that recognizes and guards against their manipulative interpersonal style. Offenders at different I levels may pose
equal risks for reoffending and have similar criminogenic needs, but
the way the needs and risk level are addressed differ depending on
their interpersonal maturity level.
Although my example treated the assessment of responsivity separately from risk and needs, some measures of responsivity can assess
risk and criminogenic needs. The I-Level, CM, and MMPI demonstrate poor predictive validity with respect to recidivism and therefore
contribute minimally to the assessment of risk and criminogenic needs
(for a review of the research, see Andrews & Bonta, 1998a). However,
the assessment of antisocial personality is an example where information relevant to risk, needs, and responsivity are given. The PCL-R
research can be used to describe what I mean. The PCL-R has shown
good predictive validity for general and violent recidivism; that is,
assessments with the PCL-R yield information on risk. As explained
previously, it has not been used to formulate treatment planning (i.e.,
372
need and responsivity). Yet the PCL-R can provide information for
this purpose. Examination of the individual items in the PCL-R yields
three groupings. There are static risk factors (e.g., early behavior
problems, criminal versatility), dynamic criminogenic needs (e.g.,
pathological lying, poor behavioral controls), and responsivity factors
(e.g., grandiose sense of self-worth, shallow affect). Thus, the construct of antisocial personality or psychopathy assesses not only risk
and needs but also responsivity factors.
Treatment programs for offenders labeled antisocial personality or
psychopathic have recognized the need to deal with these offenders
differently from other offenders. With antisocial personality disordered offenders, the reinforcement contingencies within group sessions must be tightly controlled, extra attention must be paid to motivational issues, and a concerted effort must be made to have them
assume greater responsibility for their behavior (Hare, 1998; Wong,
2000). Assessments of mental disorders (e.g., schizophrenia, manic
depression, etc.) also are important for responsivity consideration.
The static and dynamic risk factors among mentally disordered
offenders are highly similar to the general offender population (Bonta
et al., 1998) but how programming is delivered to offenders with a
mental disorder will certainly influence the success of treatment.
Finally, assessment of responsivity is not limited to the personalitycognitive attributes of offenders. Broader social and cultural factors
also play a role. We do not need tests to assess gender, race, and ethnicity, but program delivery must attend to these factors. Although the
risk assessment research across sociocultural groups is not particularly well developed, it appears that the risk factors for criminal behavior are similar across gender (Simourd & Andrews, 1994) and race and
ethnicity (Bonta, LaPrairie, & Wallace-Capretta, 1997; Lowenkamp
& Latessa, 2000). Once again, for treatment to have an impact, the
style of service must be tailored to the social, cultural, and ethnic characteristics of the offenders.
GUIDELINE 8: USE DIFFERENT METHODS TO ASSESS RISK AND NEEDS
Any test has error associated with it, and therefore, no single test
predicts perfectly. The sources of error vary (e.g., poor conceptualization of the construct being measured, lack of motivation on the part of
the client, etc.). One important source of error is the format used in
assessment. The predictive validity of assessments can improve when
different methods are combined. When different methods are used,
the weakness of one assessment instrument can be compensated by
the strength of another test. There are four common methods used in
offender assessment: (a) paper-and-pencil, (b) interview-based, (c)
behavioral, and (d) file extraction procedures. Each method has
known weaknesses, and many are well known. Paper-and-pencil
questionnaires yield poor results if the client fails to understand the
questions, assessments conducted by interview are open to the personal biases of the examiner, behavioral assessments may be overly
specific and contrived, and file extraction methods are limited by the
quality of the information in the files.
In the area of corrections, all four methods of assessing offender
risk have been used. Paper-and-pencil measures are used in general
offender classification (e.g., Jesness, 1988, I-Level; Megargee &
Bohn, 1979, Megargees MMPI-based system) and for more specialized assessments such as the assessment of antisocial attitudes (e.g.,
Pride in Delinquency [PID] scale) (Shields & Whitehall, 1991;
Simourd & van de Ven, 1999) and the risk for violence (e.g., SelfAppraisal Questionnaire) (Loza & Loza-Fanous, 2000, 2001).
Among the interview-based measures, well-known examples are the
LSI-R, PCL-R, and HCR-20. Behavioral measures include the Adult
Internal Management System (Quay, 1984). Finally, measures using
file extraction procedures are represented by the VRAG and, for the
risk of sexual recidivism, STATIC-99 (Hanson & Thornton, 1999).
Many of the tests illustrated in the previous paragraph sample different domains. Some measure only one domain (e.g., PID: antisocial
attitudes). Others sample two or three domains (e.g., PCL-R: criminal
history, personality, attitudes), whereas some instruments, such as the
LSI-R, sample up to 10 different domains relevant to criminal conduct. These measures also use different methods. The question that I
raise in this section is what would happen if we use different methods
to assess the same domain. For example, if we were interested in
assessing antisocial attitudes, would we gain predictive accuracy if we
were able to assess the construct using multiple methods?
An affirmative answer to this question was provided in a study by
Andrews and his colleagues (Andrews, 1985; Andrews et al., 1985). I
374
At this point, the reader may think that this last guideline was an
afterthought, something added to form a top 10 list. I did not begin
writing this article with a top 10 list in mind but, I must admit that I was
a bit delighted when I wrote Guideline 8 and knew that I still wanted to
write a couple more things. My last guideline reflects my own personal values, and they probably do not agree with everyones views.
I think that the application of offender risk assessments should
adhere to the principle of the least restrictive alternative. The principle
states that the punishment of offenders should use the least intrusive
measures and only to the extent needed to manage their behavior
(Rubin, 1975). Thus, imprisonment should be reserved for the most
recalcitrant of offenders, intermediate sanctions for the nonviolent,
and perhaps minimal or no intervention for the lowest risk offenders.
Obviously, our criminal justice system does not appear to follow very
closely the least restrictive principle. With nearly 1.3 million people in
U.S. prisons, it is hard to imagine that community alternatives are not
appropriate for any of them.
Risk assessment is a double-edged sword. It can be used to justify
the application of severe sanctions or to moderate extreme penalties. It
is easy to sell a risk instrument when the instrument claims to identify the dangerous and justifies the imprisonment of those individuals.
However, the identification of the violent recidivist is not infallible.
We are not at the point where we can achieve a level of prediction that
is free from error. And considering the high levels of incarceration, I
sometimes wonder whether risk instruments serve to identify the
highly dangerous a priori or simply serve to justify decisions already
made.
Using risk instruments to identify lower risk offenders who can be
placed in a noncustodial setting is a tougher sell (albeit the identification of low-risk offenders is also not perfect). I learned this the hard
way. In the early LSI-R research, I tried to convince various jurisdictions that we were overusing incarceration as a sanction and that we
could use the instrument to identify inmates for community placement. That is, there were many offenders who did not need to be in
prison for long periods of time. Many of my arguments fell on deaf
ears. Larry Motiuk, who worked with me on a number of these projects, joked that maybe the name should be changed from Level of Service Inventory to Locked Securely Inside. I am glad that we followed the least restrictive custody principle.
SUMMARY
376
REFERENCES
American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author.
American Psychological Association. (1974). Standards for educational and psychological
tests. Washington, DC: Author.
Andrews, D. A. (1985). Notes on a battery of paper-and-pencil instruments: Part 1. Assessments
of attitudes and personality in corrections. Unpublished report, Carleton University, Department of Psychology, Ottawa, Canada.
Andrews, D. A., & Bonta, J. (1995). The Level of Service InventoryRevised. Toronto, Canada:
Multi-Health Systems.
Andrews, D. A., & Bonta, J. (1998a). The psychology of criminal conduct (2nd ed.). Cincinnati,
OH: Anderson.
Andrews, D. A., & Bonta, J. (1998b). The Level of Service InventoryScreening Version.
Toronto, Canada: Multi-Health Systems.
Andrews, D. A., & Bonta, J. (in press). The psychology of criminal conduct (3rd ed.). Cincinnati,
OH: Anderson.
Andrews, D. A., Bonta, J., & Hoge, R. D. (1990). Classification for effective rehabilitation:
Rediscovering psychology. Criminal Justice and Behavior, 17, 19-52.
Andrews, D. A., & Robinson, D. (1984). The Level of Supervision Inventory: Second report.
Report to Research Services (Toronto), Ontario Ministry of Correctional Services.
Andrews, D. A., & Wormith, J. S. (1984). Criminal sentiments and behaviour (Programs Branch
User Report). Ottawa: Solicitor General Canada.
Andrews, D. A., Wormith, J. S., & Kiessling, J. J. (1985). Self-reported criminal propensity and
criminal behaviour: Threats to the validity of assessment and personality (Programs Branch
User Report). Ottawa: Solicitor General Canada.
378