Marnie E. Rice, Ph.D., FRSC

1
Violence Risk Assessment How far have we come in
30 years?

• Marnie E. Rice, Ph.D., FRSC
• riceme_at_mcmaster.ca
• www.mhcp-research.com
• Grand Rounds
• McMaster University Dept. of Psychiatry
  Behavioural Neurosciences
• May 12, 2010

2
Violence risk assessment How far have we come
in 30 years?

• Violence Risk Assessment circa 1980
• An early study showing promise for increased
  predictive accuracy
• Research on predictors of violence
• Development of violence risk assessment tools
• Subsequent research using actuarial tools for
  violence risk assessment
• Summary of progress in 30 years
• (Current controversies and unsettled issues)
• Implications for forensic decision-making

3
Components of expertise

• Experts should make different judgments than
  laypersons
• Experts should make more accurate judgments than
  laypersons (the amount of accuracy being limited
  by the amount of agreement shown in their
  judgments)
• Experts should use specialized assessments or
  measurements in arriving at their judgments

After Einhorn,H.J. (1974) Journal of Applied
Psychology, 59, 562-571.
4
History of prediction of violence

• Prediction of violence before mid-80s
• Baxstrom (Steadman, 1973)
• Quinsey Ambtman, 1979
• Pasewark, Bieber, Bosten, Kiser, Steadman, 1982
• Monahan (1981)

Monahan, J. (1981). Predicting violent behavior
An assessment of clinical techniques. Beverly
Hills, CASage Pasewark et al. (1982).
International Journal of Law Psychiatry, 5,
356-374 Quinsey, V.L. Ambtman, R. (1979).
Journal of Consulting and Clinical Psychology,
47, 353-362. Steadman, H.J. (1973).American
Journal of Psychiatry,130, 317-319.
5
An early study showing promise for increased
predictive accuracy

• A Therapeutic Community Program Evaluation
• Participants Psychopaths, Others
• Program Description
• Program Evaluation
• Program Review
• Outcome

Rice, Harris, Cormier (1992). Law Human
Behavior, 16, 399-412.
6
Program review

• This is an exciting program which has the
  hallmark of being right as the final model of
  the DNA molecule looked right to Watson and
  Crick.
• We were satisfied that the patients benefited
  greatly from the Social Therapy Unit experience.
  We are quite sure that the program itself is of
  considerable benefit not only to the patients but
  to the hospital as a whole and the country.

7
Program review

• We were satisfied that the program has.a very
  low recidivism rate
• .developed the techniques thatare the
  most fruitful anywhere in the universe at the
  present time

8
Outcome study

• Treated
  Untreated
• n146
  n146
• Therapeutic
  Prison
• Community

Matched on Age Offense Criminal
History (Psychopathy)
9
Psychopathy treatment

Violent Recidivism

10
Predictors of violent recidivism among mentally
disordered offenders

• Big predictors
• objective risk assessment, antisocial
  personality, violent history, juvenile
  delinquency, age
• Moderate predictors
• nonviolent criminal history, adult criminal
  history, substance abuse, marital status
• Small predictors
• clinical judgement, psychosis, offense
  seriousness

Meta-analysis Bonta, Law Hanson
(1998).Psychological Bulletin, 123,123-142.
11

• Non-predictors of violent recidivism
• Psychological distress
• Remorse
• Insight

Bonta, Law, Hanson, Psychological Bulletin,1998
12
Predictors of violence among psychiatric patients

• Big predictors
• Psychopathy (PCLSV) Adult arrests, Antisocial
  personality disorder, Major mental disorder
  without substance abuse, drug or alcohol abuse,
  anger (Novaco)
• Moderate predictors
• Violent arrests, schizophrenia, child abuse,
  delusions at the time of admission
• Small predictors
• persecutory delusions, hostility, thought
  disturbance, medication nonadherence on admission

MacArthur Risk Study- Steadman, Silver, Monahan
et al.(2000). Law Human Behavior, 24, 83-100.
13
MacArthur Risk Study

• Non-predictors
• mania, depression, any delusions, hallucinations,
  command hallucinations, grandiose delusions

14
Predictors of violent recidivism among sex
offenders

• Big predictors
• Actuarial violence risk assessment, PCL-R
• Medium predictors
• Personality disorder, substance abuse, childhood
  behavior problems, prior criminal history
• Small predictors
• Degree of force used, separation from parents,
  sexually deviant interests, deviant phallometric
  assessment

Meta-analysis Hanson Morton-Bourgon (2005).
Journal of Consulting Clinical Psychology, 73,
1154-1163.
15

• Non-predictors of violent recidivism
• Childhood sexual abuse, general psychological
  functioning, depression, low self-esteem,
  anxiety, severe psychological dysfunction

16
Predictors of recidivism among offenders

• Big predictors
• objective risk assessment, antisocial
  personality, criminal history
• Moderate predictors
• age, juvenile antisocial behavior, substance
  abuse
• Small predictors
• Family structure, family criminality, personal
  distress

Meta-analysis Gendreau, Little Coggin (1996).
Criminology, 34, 575-607.
17
Development of an actuarial tool for the
appraisal of violence risk

• 618 offenders and mentally disordered offenders
• Approximately 50 candidate predictors
• 7 years average time at risk
• 31 committed a new violent offense

Harris, Rice, Quinsey (1993). Criminal Justice
Behavior, 315-335.
18
Violence Risk Appraisal Guide(VRAG)

• Psychopathy Checklist Score
• Elementary school maladjustment
• Age at index offense
• DSM III personality disorder
• Separated from parents ltage 16
• Failure on prior conditional release

19
Violence Risk Appraisal Guide

• History of nonviolent offenses
• Never married
• DSM III schizophrenia
• Victim injury in index offense
• History of alcohol abuse
• Male victim in index offense

20
VRAG- Psychometric properties

• Range of scores -26 to 38, often divided into 9
  bins
• Mean score in construction sample .91 (SD12.9)
• IRR .90
• SEM 4.1 (Means that 95 confidence interval is
  approx. /- 8 or 1 bin)

Quinsey, Harris, Rice, Cormier (1998,
2006).Violent Offenders Appraising and Managing
Risk. Washington, D.C. American Psychological
Association.
21
Performance of the VRAG
N
22
Nonrecidivists Recidivists
N
VRAG Score
23
Receiver Operator Characteristic
AUC .76
24
Replications of VRAG/SORAG (n58)
25
Performance of the VRAG on cross-validation
Harris, Rice, Cormier. (2002). Law Human
Behavior
26
Illustrative ROCs
Chance (AUC.50) Clinicians (AUC.62) Obtained
VRAG(AUC.75) Hypothetical Optimal Actuarial
(AUC.85)
Data from Harris, Rice, Cormier, 2002
27
Accuracy of the VRAG under changed conditions

• Shortterm violence
• Shortest time 12 weeks, ROC area .71
• Outside Canada (US, Europe)
• 17 studies, Mean ROC area .77
• Self-reported violence
• 2 studies, Mean ROC area .73

28
Accuracy of the VRAG whenoutcome is changed

• Institutional violence or misconduct
• 6 studies, Mean ROC area .68
• Sexual re-offense
• 3 studies, Mean ROC area .68
• Any re-offense
• 2 studies, Mean ROC area .76
• Spousal assault recidivism
• 2 studies, Mean ROC area .71

29
Other results

• VRAG predicts
• Number of violent offenses
• Time until first violent offense
• Total months of opportunity
• Severity of total offenses

30
Accuracy of the VRAG among different populations

• Offenderslt 18, lt16
• Developmentally delayed offenders
• Sex offenders (also SORAG)
• Wife assaulters (also ODARA, DVRAG)
• Emergency psychiatric patients

31
Long-term follow-up study

• Method
• 1335 Canadian male offenders from previous
  studies- released in 60s, 70s, 80s, 90s
• New follow-up data gathered in 2000s
• Survival curves using VRAG and PCL-R

Rice Harris, in preparation
32
Predictive accuracy over time

• Examined predictive accuracy of VRAG and PCL-R
  over exact follow-ups ranging from 6 months to
  20 years
• Declared offenders who failed after the follow-up
  interval to be successes for that follow-up
• Successes had to be known to have had at least
  that amount of time at risk or be known to have
  died

33
6 months (N1191)
VRAG
AUC.757
Sensitivity
PCL-R
AUC.748
7 failure rate
34
10 years (N1171)
VRAG
AUC.738
Sensitivity
PCL-R
AUC.673
48 failure rate
35
20 years (N995)
VRAG
AUC.776
Sensitivity
PCL-R
AUC.698
63 failure rate
36
Violence risk appraisals using only current
actuarial instruments

• Considerable expertise
• Different (and more accurate) judgments than
  other experts (and laypersons)
• More reliable than laypersons
• Use special instruments- DSM diagnosis, PCL-R ,
  VRAG, Static-99, DVRAG

37
Meta-analysis of violence risk assessment tools

• Compared measures used to predict violence in
  adults
• 5 instruments with sufficient effect sizes to
  look at them individually
• 4 actuarial tools VRAG, Level of Supervision
  Inventory (LSI-R), Statistical Instrument on
  Recidivism (SIR), PCL-R
• 1 Structured Clinical Judgment tool Historical,
  Clinical and Risk Management Violence Risk
  Management Scheme (HCR-20)
• Actuarial tools better on average
• VRAG significantly better than all other measures
  ( only actuarial tool designed to predict
  violent recidivism)

Campbell, French, Gendreau (2009).Criminal
Justice Behavior, 36, 567-590.
38
Meta-analysis of prediction tools used for sex
offenders

• Sexual Recidivism
• Measures for sexual recidivism ROC
  area
• Actuarial
  .69 (.67-.70)
• Structured professional judgment .62 (.57-.66)
• Violent Recidivism
• Measures for violent recidivism
• Actuarial
  .73 (.70-.75)
• Unstructured professional judgment .58
  (.503-.63)
• Any Recidivism
• Measures for any recidivism
• Actuarial
  .79 (.74- .82)
• Unstructured professional judgment .56 (.52-
  .60)

Hanson Morton-Bourgon (2009). Psychological
Assessment,21, 1-21.
39
So how far have we come in 30 years?

• Many studies of predictors of violence
• Meta-analyses of predictors of violence
• Combining predictors into risk assessment tools
  that show evidence of accuracy better than
  clinicians without tool
• Meta-analyses of risk assessment tools to
  determine best tools for each purpose

40
Unresolved Issues and Current Controversies

• Can we improve on actuarial tools by allowing
  clinical adjustment?
• Can we improve actuarial instruments by including
  dynamic predictors?

41
Can we improve on the VRAG by allowing clinical
adjustment?

• No association between actuarial risk score and
  clinicians opinions even when VRAG score was
  made available
• Recidivism related to VRAG score (r.42), but not
  to clinician recommendation (r.14)

Hilton Simmons (2001). Law and Human Behavior,
25, 393-408.
42
Replication of Hilton Simmons

• No significant association between tribunal
  decisions and actuarially assessed risk of
  violence
• Significant correlation between actuarial risk
  and clinical advice to the tribunal (r .21,
  plt.05)

McKee, Harris, Rice (2007).Behavioral Sciences
and the Law, 25, 485-506.
43
Another example

• Federal judges in U.S. have discretion to
  over-ride federal sentencing guidelines based on
  an actuarial instrument for predicting
  recidivismThe Salient Factor Score
• They are given what the Federal Sentencing
  Guidelines recommend and then can over-ride so
  long as they justify reasons
• Gathered follow-up data after offenders were
  released

Krauss (2004). Behavioral Sciences and the Law,
22, 1-20.
44
ROCsAfter Krauss, 2004, Behavioral Sciences
and the Law

Sentencing guidelines (AUC .57) Judges sentences
(AUC.46)
45
Summary

• No evidence that professional judgment combined
  with actuarial can do a better job than actuarial
  alone and some evidence it can make them worse

46
Can we improve actuarial instruments by including
dynamic predictors?

• VRAG and other actuarial instruments include only
  static variables
• Scores dont change with time or treatment
• Could they be improved by including measures of
  change over time or change due to treatment?

47
A Study of Dynamic Risk

• 151 patients discharged from Dutch security
  hospitals
• Before discharge, clinicians rated their patients
  on a Dynamic Reoffending Risk Scale
• Also rated patients likelihood of violent
  recidivism on a 6 point scale
• Mean follow-up time was 6.2 years

Philipse et al. (2006) Law Human Behavior , 30,
309-327.
48
Examples of Dynamic risk scales

• Empathic acceptance of responsibility for offense
  (11 items)
• Antisocial narcissism (10 items)
• Self- reliance (12 items)
• Treatment compliance (5 items)
• Attainment of treatment goals (7 items)
• Patient avoids others (2 items)

Philipse et al. (2006) Law Human Behavior
49
Results

• In general, dynamic risk indicators were
  related to recidivism in the opposite direction
  to that expected by clinicians
• Avoids contact (-.39)
• Denies offense (-.74)
• Ability to empathize with victim (.27)
• Yielded few plausible treatment targets

Philipse et al. (2006) Law Human Behavior
50
Clinicians ratings of recidivism risk
1
.75
Sensitivity
.5
.25
Chance Clinicians ratings

ROC area .44
0
1
.75
1 -specificity
.25
After Philipse et al., 2006
51
What is a dynamic predictor variable?
Score on variable precedes and is related to
recidivism?
Yes
Score changes/Can be changed?
Yes
Changed variable score predicts better than (or
amount of change adds to) original score?
Yes
Truly dynamic predictor variable
52
Are deviant sexual preferences a dynamic risk
factor?

• Phallometric preferences can be measured before
  recidivism?
• Phallometric preferences related to recidivism?
• Phallometric preferences can be changed with
  treatment?
• Post- treatment preferences predict recidivism
  better than pre-treatment preferences?
• Evidence that they are dynamic risk indicator?

Rice, Quinsey, Harris, 1991
Yes
Yes
Yes
No
No
53
Prospects for Dynamic Prediction

• Dynamic violence risk items
• Shallow affect, lacks empathy or concern for
  others
• Antisocial attitudes
• Poor therapeutic alliance
• Predict WHEN an offender is likely to be violent
• But high VRAG scorers are always at higher risk
  than lower scorers
• Makes most sense to use this primarily for high
  VRAG patients

Quinsey, Coleman, Jones, Altrows (1997),
Journal of Interpersonal Violence, 12, 794-813
Quinsey, Book, Skilling (2004) Journal of
Applied Research in Intellectual Abilities, 17,
243-254 Quinsey, Jones, Book Barr (2006).
Journal of Interpersonal Violence, 21, 1539-1565.
54
Challenges for Dynamic Prediction of Who is
Likely to be Violent

• Static predictors alone yield very high effect
  sizes (ROC areas up to .85)
• SO...Not much room for improvement on static
  predictors given noise in outcome measure unless
  dramatically effective treatments are found
• Dynamic predictors can likely make only a very
  modest contribution to the prediction of who is
  likely to reoffend
• Main hope for dynamic prediction is in predicting
  when recidivism is likely to occur

55
Summary

• No evidence yet for dynamic predictors of who is
  likely to fail violently

56
Implications for policy

• To maximize public safety and minimize
  unnecessary incarceration, mandate the use of
  actuarial tools wherever law specifies risk as
  relevant
• Control the amount they can be over-ridden
• Dispositions for those found not criminally
  responsible by reason of mental disorder
• Parole probation for offenders
• Sentencing
• Long-term offender status
• Dangerous offender status, sexually violent
  person (U.S.), dangerous severe personality
  disorder (UK)

McKee et al., 2007
57
Implications for correctional and forensic
practice

• Use the best actuarial tool for the purpose
• Sexually violent recidivism
• Static-99, Static-2002, MnSOST-R, Risk
  Matrix-2000 sex or SORAG
• Violent recidivism
• VRAG, SORAG, Risk Matrix-2000 violence
• Wife assault recidivism
• DVRAG, ODARA
• Any recidivism
• LSI, SIR

Hilton, N.Z., Harris, G.T., Rice, M.E. (2010).
Risk assessment for domestically violent men
Tools for criminal justice, offender
intervention, and victim services. Washington,
DC American Psychological Association. Phenix,
Hanson, Thornton (2000) CODING RULES FOR THE
STATIC-99 Available from http//ww2.ps-sp.gc.ca/
publications/corrections/CodingRules_e.asp
Hanson Thornton (2003). Notes on the
development of Static-2002. Available
fromhttp//www.publicsafety.gc.ca/res/cor/rep/2003
-01-not-sttc-eng.aspx. Rice, Harris, Lang,
Cormier (2006). Law and Human Behavior, 30,
525-541.
58
Implications for practice (cont.)

• Time spent doing a thorough risk assessment is
  well worth it
• Permits good recommendations about release
• Permits treatment and management interventions to
  be apportioned in accordance with risk

59
Implications for Research

• Continue search for dynamic risk indicators
• Continue to improve actuarial tools
• Evaluate effect on public safety of controlled
  vs. unbridled deviation from actuarial risk score
  in decisions of review boards, parole boards,
  sentencing decisions