Assessing Risk for Violence

1
Assessing Risk for Violence

• Stephen D. Hart
• Simon Fraser University

2
Violence

• Actual, attempted, or threatened physical harm
  that is deliberate and nonconsenting
• Includes violence against victims who cannot give
  full, informed consent
• Includes fear-inducing behavior, where threats
  may be implicit or directed at third parties

3
Violence as a Choice

• The proximal cause of violence is a decision to
  act violently
• The decision is influenced by a host of
  biological, psychological, and social factors
• Neurological insult, hormonal abnormality
• Psychosis, personality disorder
• Exposure to violent models, attitudes that
  condone violence

4
We Cant Predict Violence

• Violence is too rare to predict with any
  accuracy, by any means
• Professional decisions are particularly bad
• Predictions of violence necessitate a
  deterministic view of behavior

5
Or, Can We?

• The issue of prediction is moot
• Dont predict, evaluate risk (e.g., suicide)
• Regardless, predictions made by professionals are
  reliably better than chance
• The scientific literature on violence is large
  and growing
• Any choice can be predicted

6
What is Risk Assessment?

• Process of understanding a hazard to limit its
  potential negative impact
• Hazard identification (which events occur?)
• Hazard accounting (how frequently?)
• Scenarios of exposure (under which conditions?)
• Risk characterization (conditions present?)
• Risk management (which interventions?)

7
Nature of Violence Risk

• Violence risk is a multi-faceted construct
• Nature what kinds of violence might occur?
• Severity how serious might the violence be?
• Frequency how often might violence occur?
• Imminence how soon might violence occur?
• Likelihood what is the probability that violence
  might occur?

8
Risk is Context-Specific

• We never know a persons risk for violence we
  merely estimate it assuming certain conditions
• Assuming institutionalization, assuming release
  with supervision, assuming release without
  treatment for substance use
• Consequently, relative or conditional risk
  judgments are more useful than absolute or
  probabilistic risk judgments

9
Goals of Risk Assessment

• To make better decisions
• Improve consistency
• Protect public safety
• Guide intervention
• Protect clients rights
• Liability management

10
Risk CharacterizationContent Issues
11
What to Include?

• Three primary criteria
• Empirical (predictive accuracy)
• Professional (practical utility)
• Legal (fairness and reasonableness)

12
Problems With Empirical Criterion

• Not everything that is important has been proven
  or validated scientifically
• Can lead to exclusion of good but rare or
  difficult-to-assess risk factors
• Prediction ? cause, explanation, or intervention
• Can lead to inclusion of bad but common or
  easy-to-assess factors

13
Example The SIEVE

• Age Young is bad
• Sex Male is bad
• Facial hair Dense is bad
• Foot size Big is bad

14
Problems With Professional Criterion

• Focus on dynamic factors may bias risk
  assessments
• Can lead to exclusion of good but static or
  easy-to-ignore factors
• Conventional wisdom of professionals may be plain
  wrong
• Can lead to inclusion of bad but vivid or
  dramatic factors

15
Example Clinical Intuition

• Depression Present is good
• Anxiety Present is good
• Intelligence High is good
• Rorschach Seeing viscera is bad

16
Problems With Legal Criterion

• Useful for excluding risk factors, but not for
  including them
• It can be argued that almost any risk factor is
  unfair or unreasonable in some respect

17
Risk CharacterizationProcedural Issues
18
Conventional Approaches

• Professional judgment
• Unstructured or clinical
• Anamnestic (see Melton et al., 1997)
• Structured (e.g., HCR-20, SVR-20)
• Actuarial decision-making
• Psychological tests (e.g., MMPI-2, PCL-R)
• Risk scales (e.g., VRAG, RRASOR)

19
Professional Judgment

• Most commonly-used method for violence risk
  assessment
• Familiar to professionals
• Familiar to courts and tribunals
• General strengths of method
• Flexible (easy administration)
• Requires limited training and technology
• Person-centered (idiographic)

20
Unstructured Features

• No constraints on evaluation
• Any information can be considered
• Information can be gathered in any manner
• No constraints on decisions
• Information can be weighted and combined in any
  manner
• Results can be communicated in any manner

21
Unstructured Limitations

• No systematic empirical support
• Low agreement (unreliable)
• Low accuracy (unvalidated)
• Foundation is unclear (unimpeachable)
• Relies on charismatic authority
• Decisions are broad bandwidth
• Focus is on culpability, not action

22
Anamnestic Features

• Imposes minor structure on evaluation
• Must consider, at a minimum, nature and context
  of past violence
• Action-oriented
• Logically related to development of risk
  management strategies
• Consistent with relapse prevention or harm
  reduction approaches

23
Anamnestic Limitations

• Unknown reliability
• Unknown validity
• Assumes that history will repeat itself
• Violent careers are static
• Violent people are specialists

24
Structured Features

• Imposes major structure on evaluation
• Must consider, at a minimum, a fixed and explicit
  set of risk factors
• Specifies process for information-gathering
• Imposes minor structure on decision
• Specifies language for communicating findings
• Action-oriented

25
Structured Limitations

• Requires retooling of evaluation process
• Systematized information-gathering
• New training and technology
• Justification for imposing structure requires
  inductive logic (faith)
• What works elsewhere will work here
• Professional discretion is appropriate

26
HCR-20

• HCR-20, version 2
• Webster, Douglas, Eaves, Hart (1997)
• Designed to assess risk for violence in those
  with mental or personality disorders
• 10 Historical, 5 Clinical, and 5 Risk Management
  factors

27
Applications

• Assess clinical evaluations of violence risk
  across a broad range of populations and settings
• Civil and forensic psychiatric, correctional
• Institution, community
• Monitor clinical and situational factors that may
  be relevant to violence
• Guide risk management strategies

28
Conceptual Basis

• Intended to bridge clinical and empirical domains
  and knowledge bases
• Evidence-based risk assessment
• Content determined rationally
• Based on reviews of scientific and professional
  literatures
• Not optimized on a particular sample

29
Temporal Organization
30
Historical Factors

• Previous violence
• Young age at first violence
• Relationship instability
• Employment problems
• Substance use problems

• Major mental illness
• Psychopathy
• Early maladjustment
• Personality disorder
• Prior supervision failure

31
Clinical Risk Management Factors

• Lack of insight
• Negative attitudes
• Active symptoms of major mental illness
• Impulsivity
• Unresponsive to treatment

• Plans lack feasibility
• Exposure to destabilizers
• Lack of personal support
• Noncompliance with remediation attempts
• Stress

32
Actuarial Decision-Making

• Commonly-used adjunctive method for violence risk
  assessment
• Familiar to some professionals (psychologists)
• Somewhat familiar to courts and tribunals
• General strengths of method
• Highly structured/systematic (objective)
• Empirically-based (scientific)

33
Psychological Tests Features

• Measure some disposition that predicts violence,
  according to past research
• Reliability and validity of test-based decisions
  has been evaluated
• Imposes major structure
• On some part of the evaluation process
• On some part of the decision-making process

34
Psychological Tests Limitations

• Require professional judgment
• Which tests to use
• How to interpret scores
• Justification of use requires inductive logic
• Our population is like theirs
• Our use of the test is like theirs

35
PCLSV

• Symptom construct rating scale
• requires clinical / expert judgment
• based on all data
• Data obtained from two primary sources
• review of case history (required)
• interview / observation (recommended)

36
PCLSV Items

• Part 1
• Superficial
• Grandiose
• Deceitful
• Lacks remorse
• Lacks empathy
• Doesnt accept responsibility

• Part 2
• Impulsive
• Poor behavioral controls
• Lacks goals
• Irresponsible
• Adolescent antisocial behavior
• Adult antisocial behavior

37
PCL Summary 1

• The correlation between the PCL and violent
  recidivism averages about .35
• Regardless of length of follow-up
• Even in sex offenders, forensic patients, women,
  delinquents even in nonviolent, nonpsychopathic
  offenders
• Association is quasi-linear (positive and
  monotonic)

38
PCL Summary 2

• Among psychopaths (e.g., PCL-R gt 30) released
  from prison, the 5-year violent reoffense rate is
  about 70
• Versus about 30 in low group (lt 20) and 50 in
  medium group (21-29)
• Versus low group, psychopaths are at very high
  risk for reoffense (rate ratio 2x to 3x odds
  ratio 5x to 10x)

39
PCL Summary 3

• Psychopaths not only commit more violence, they
  commit different kinds of violence
• The violence of psychopaths often has unusual or
  atypical motivations
• Instrumentality/gain
• Impulsivity/opportunism
• Sadism

40
PCL Conclusions

• Psychopathy must be assessed as part of
  comprehensive violence risk assessments
• The presence of psychopathy compels a conclusion
  of high risk
• The absence of psychopathy does not compel a
  conclusion of low risk
• Psychopathy must be assessed by trained
  professionals using adequate procedures

41
Decision Tree
Homicidal/suicidal?
YES
High Risk
NO
Sexual sadism?
High Risk
YES
NO
Psychopathic? (e.g., PCL-R gt 30)
High Risk
YES
NO
Assess other factors (e.g., HCR-20)
42
Risk Scales Features

• Designed solely to predict an outcome
• High-fidelity
• Optimized for specific outcome, time period,
  population, and context
• Impose rigid structure
• On all of the evaluation process
• On all of the decision-making process

43
Risk Scales Limitations

• Still require professional judgment
• Which scales to use
• How to interpret scores
• Justification of use still requires induction
• Our population is like theirs
• Our use of the test is like theirs
• Results may be easily misinterpreted
• Pseudo-objective, pseudo-scientific

44
VRAG

• Violence Risk Appraisal Guide
• Quinsey et al. (1998)
• Constructed in adult male patients assessed or
  treated at a maximum security hospital
• 12 items weighted according to ability to
  postdict violence over 7 year follow-up
• Total scores divided into 9 bins, with estimated
  p(violence) from 0 to 100

45
VRAG Items

• PCL-R score
• Elem. school problems
• Personality disorder
• Age ()
• Separated from parents under age 16
• Failure on prior conditional release

• Nonviolent offense history
• Never married
• Schizophrenia ()
• Victim injury ()
• Alcohol abuse
• Female victim ()

46
VRAG Potential Problems

• The VRAG is, in essence, a history lesson
• What if patient profile changes?
• What if p (violence) changes?
• What if the assessment context changes?

47
VRAG Actual Problems?

• Paul Bernardo is a convicted serial murderer (3
  sexual homicides) and serial rapist (75 known
  rapes)
• Currently serving life imprisonment for murder,
  and an indeterminate sentence for the rapes
• VRAG completed on the basis of case history data

48
Bernardos VRAG Results

• PCL-R score 4
• Elem. school -1
• Personality disorder 3
• Age 0
• Separated from -2
• Failure on prior 0
• Nonviolent offense -2
• Marital status -2
• Schizophrenia 1
• Victim injury -2
• Alcohol abuse 1
• Female victim -1

Total -1 Bin 4 p(viol) 17 - 31
49
Problems With All

• Focus on negative characteristics
• Sticky labels
• What about strengths (resources, buffer
  factors)?
• Risk assessment ¹ risk management
• What to do with high-risk individuals?
• Quality control
• Who will assess risk, and how?

50
Risk References

• Boer, D. P., Hart, S. D., Kropp, P. R.,
  Webster, C. D. (1997). Manual for the Sexual
  Violence Risk-20 Professional guidelines for
  assessing risk of sexual violence. Burnaby,
  British Columbia Simon Fraser University.
• Grove, W. M., Meehl, P. E. (1996). Comparative
  efficiency of informal (subjective,
  impressionistic) and formal (mechanical,
  algorithmic) prediction procedures The
  clinical-statistical controversy. Psychology,
  Public Policy, and Law, 2, 293-323.
• Kropp, P. R., Hart, S. D., Webster, C.W.,
  Eaves, D. (1995). Manual for the Spousal Assault
  Risk Assessment Guide, 2nd ed. Vancouver, BC
  British Columbia Institute on Family Violence.
• Meehl, P. E. (1996). Clinical versus statistical
  prediction A theoretical analysis and a review
  of the literature. Northvale, NJ Jason Aronson.
  (Original work published in 1954.)

51
Risk References (cont.)

• Melton, G. B., Petrila, J., Poythress, N. G.,
  Slobogin, C. (1997). Psychological evaluations
  for the courts A handbook for mental health
  professionals and lawyers (2nd ed.). New York
  Guilford.
• Quinsey, V. L., Rice, M. E., Harris, G. T.,
  Cormier, C. (1998). Violent offenders Appraising
  and managing risk. Washington, DC American
  Psychological Association.
• Webster, C. D., Douglas, K. S., Eaves, D.,
  Hart, S. D. (1997). HCR-20 Assessing risk for
  violence, version 2. Burnaby, British Columbia
  Simon Fraser University.

52
But How Accurate AreCharacterizations of Risk?
53
A Complex Phenomenon

• Studying the accuracy of risk characterizations
  is difficult due to the complexity of
• The characterizations
• The violence
• The follow-up
• Indexes of accuracy

54
The Characterizations

• Evaluator Professional vs. researcher novice
  versus expert
• Process Clinical vs. actuarial, contextual vs.
  context-free
• Timing Admission vs. discharge, static vs.
  dynamic
• Metric Uni- vs. multi-dimensional, categorical
  vs. continuous

55
The Violence

• Direction Self vs. others, acquaintances vs.
  strangers
• Nature Instrumental vs. reactive
• Severity Threats vs. battery vs. homicide
• Frequency Single vs. multiple events
• Context Situational precipitants or co-factors

56
The Follow-Up

• Data source Patient vs. collaterals vs. records
• Time at risk Weeks vs. months vs. years
• Interventions Dynamic factors, life events
• Monitoring Continuous vs. endpoint

57
Indexes of Accuracy

• Comparison group Other patients vs. normals
• Statistic Uni- vs. multi-variate, time
• Weighting of errors Equal vs. differential
• Interpretation Chance vs. status quo vs.
  perfection

58
Science Responds to Complexity
59
Example Psychopathy
Harris, Rice, Cormier (1991)
60
Example (cont.)

• In this study...
• Accuracy of positive predictions is 77
• Accuracy of negative predictions is 79
• Overall accuracy is 78
• Chance-corrected agreement is 53
• Correlation is .53
• Odds ratio is 12.5

61
Example (cont.)

• So, how did we do?
• Relative to chance Great!
• Relative to perfection Awful!
• Relative to the status quo ???

62
What is the status quo?

• Predictions of violence using the PCL-R typically
  have an effect size (r) of about .35 the average
  effect size for psychosis is about .30
• An effect size of .40 may be the forensic sound
  barrier
• But what is the status quo in other human
  endeavors?

63
Meta-Meta-Analysis

• Lipsey Wilson (1993) reviewed 302 meta-analyses
• Determined typical effect sizes for
  psychological, educational, and medical
  interventions

64
Psychological Interventions

• CBT ? depression .44
• Psychotherapy any ? .39
• Correctional programs (youths) .23 any ?
• Diversion (youths) ? recidivism .20
• Correctional treatment (adults) .12 any ?

65
Educational Interventions

• Small classes ? class climate .26
• Tutoring ? grades .20
• Small classes ? grades .10
• Media campaigns ? seatbelt use .06

66
Medical Interventions

• Speech therapy ? stuttering .54
• Bypass surgery ? angina pain .37
• Cyclosporine ? organ rejection .15
• Bypass surgery ? mortality .07
• ASA ? heart attack .04

67
Violence Predictions in Context

• Speech therapy ? stuttering .54
• CBT ? depression .44
• Bypass surgery ? angina pain .37
• Psychopathy ? violence .35
• Psychosis ? violence .30
• Small classes ? class climate .26

68
Conclusions

• Violence predictions, on the whole, are just as
  good as most other human prognostications
• Consider other fields...
• Stock analysts
• Meteorologists
• Fire inspectors
• Structural engineers

69
Risk AssessmentReports and Testimony
70
General Issues

• Acknowledge professional qualifications
• But, less important than approach used
• Acknowledge limits of information base
• But, comprehensiveness depends on context
• Explain risk factors considered
• Empirical, professional, legal justification
• Acknowledge uncertainty of predictions

71
Do Dont

• Use multiple approaches
• Make relative or conditional risk judgments
• Make detailed risk management recommendations

• Rely on a single approach
• Make absolute or probabilistic risk judgments
• Ignore the issue of what can or should be done

72
Special Issues Professional

• Report should tell a story
• Summarize the circumstances of past violence and
  any recent changes in them
• Describe the likely nature and context
  (scenarios) of future violence
• Identify factors that may increase risk and
  therefore serve as flags for re-assessment
• Recommend, evaluate, and prioritize risk
  management strategies

73
Cross-Exam Professional

• What, if any, is the basis of your expertise in
  the assessment of violence risk?
• What is the scientific basis for your decision
  (not) to follow these procedures?
• What is the scientific basis for your decision
  (not) to consider these factors?
• Can you state with any reasonable degree of
  scientific certainty the likelihood that X will
  be violent?

74
Special Issues Actuarial

• Report should provide full interpretation of test
  scores
• Be familiar with research supporting the tests
  reliability and validity
• Discuss general limitations of the test
• Discuss limitations of the test in this case

75
Cross-Exam Actuarial

• Isnt it arbitrary to consider risk factors such
  as A, B, and C, but to ignore D, E, and F?
• Is there any scientific evidence that scale Z
  predicts outcome Y in patients at this clinic?
• How do you know that predictions using scale Z
  are accurate for Mr. X?
• By using scale Z, arent you simply relying on
  statistical profile evidence?

76
Contact Information

• Stephen D. Hart, Ph.D.Department of
  PsychologySimon Fraser UniversityBurnaby,
  British ColumbiaCanada V5A 1S6
• Tel [phone] / Fax [phone]E-mail
  shart_at_arts.sfu.caURL www.sfu.ca/psychology/group
  s/faculty/hart