PSYCH 3CC3 Lecture 8: Assessment of Violence Risk
19 Jun 2018
School
Department
Course
Professor
Done typically after trial.
-
Relevant to post-trial decisions
Assessment of sentencing, parole
○
Parole decisions rely on assessment of violence risk
○
Decisions made within the prison system (i.e., counselling, programming,
holding for this particular individual)
○
-
Canadian psychologists have made the major contributions to the literature -
bulk from Ontario.
-
Baxtrom v Herold (1996) -U.S. Supreme Court
Baxtrom was a prisoner in the NY State prison system.
-
6 months before he was to be released, he was found to be in need of mental
health support and was moved to a mental health facility for criminals.
-
They held him in this facility long past the time of his release.
-
They didn't release him because they feared that he would be a risk to society.
-
Baxtrom sued on the basis of high rights being violated
He was hospitalized beyond his prison term and without a dangerousness
assessment.
○
-
The Supreme Court agreed that he should be released.
This released ~ 1000 prisoners moved to low security hospitals.
○
20% subsequently violent - much smaller than what would have been
predicted
○
18% discharged within a year
○
1% were later readmitted to secure hospitals because they committed a
violent criminal act.
○
-
Over 4.5 years:
50% of patients were released.
○
< 3% returned to secure hospitals.
○
-
Interesting considering the recidivism rate (30%).
Much lower recidivism rate for these people who were held against their
will on the basis of violent concerns, mental health concerns.
○
-
Thornberry and Jacoby (1979)
586 patients released from Pennsylvania institution (Dixon v Pennsylvania)
-
3-year return/recidivism rate = 23.7%
-
4-year violent arrest record = 15%
-
Quinsey & Ambtman (1979)
Wanted to see how good forensic psychiatrists were at assessing violence risk.
9 high school teachers and 4 forensic psychiatrists.
-
Evaluate recidivism risk, release, for 30 patients:
11 property offenders
○
9 child molesters
○
11 serious adult offenders (murder)
○
-
3 types of patient data:
Offense description
○
Patient history
○
Psychiatric assessments (IQ, MMPI, Rorschach, etc.)
○
-
Asked them to then predict reoffending, release, etc.
-
Inter-rater reliability low for both psychiatrists and teachers.
-
Inter-rater reliabilities similar for psychiatrists and teachers.
-
Psychiatrists rate offenders as more likely to commit offense than do teachers.
-
Little use of psychiatric assessments made by either group.
-
Summary:
"these date question the usefulness of psychiatric examinations in the
prediction of dangerousness. Perhaps psychiatric examinations would be
best restricted to determining whether an offender is treatable and
should not address the issue of dangerousness."
○
-
Violence of Assessment in the 1980s
Monahan (1981): "psychiatrist and psychologists are accurate in no more than
one out of three predictions of violent behaviour."
Book stopped research into violence assessment - he was extremely
critical of violence assessments.
○
-
Barefoot v.Estelle (1983):
Barefoot was a defendant incarcerated with mental health issues and he's
been denied release on the basis of the fact that he will be a risk to the
public.
○
He sues to have the psychiatric evaluation removed from his record since
we known psychiatrists are not good at assessing violence risk.
○
The APA agreed with him!
○
"we are not convinced that the view of the APA should be converted into
a constitutional rule barring an entire category of expert testimony …
neither petitioner nor the Association suggests that psychiatrists are
always wrong with respect to future dangerousness, only most of the
time."
○
Basically, they say that even though psychiatrists are terrible at assessing
future violence risk, we can still use it???? WHY
○
-
Violence Risk Assessment Research
Actuarial Methods: ignored the inaccurate clinician, and rely on statistics data.
Re: how insurance companies select your rates.
○
Determine how likely it is that people with the person's characteristics are
to violently offend.
○
-
Structured Professional Judgment: increase clinician's accuracy by standardizing
the assessment process.
-
Measures of Accuracy (binary decision making?)
Disorder = violence risk, presence of future/current/past disorder, etc.
-
Positive Predictive Power
Based on test results
○
Hits/(Hits + False Alarms)
○
Of the total of people we said had the disorder, what proportion actually
have it.
○
-
Negative Predictive Power
Correct Rejections/(Correct Rejections + Misses)
○
Of the total people we said did not have the disorder, what proportion
actually don't have it.
○
-
Sensitivity
S = H/(H+M)
○
Of all of the people with the disorder, what proportion did we say actually
had the disorder.
○
-
Specificity
Sp = R/(R + FA)
○
Of all of the people without the disorder, what proportion actually don't
have it.
○
-
Overall Accuracy
= R + H / (R + H + FA + M)
○
-
Heavily influenced to the base rate - frequency of occurrence in the population.
Number of correct choices we make is heavily affected by the base rate.
○
-
PPP: Sensitivity to Base Rate
Example #1:
Assume 95% accuracy
○
Assume 10,000 people, disorder has 50% prevalence
○
Grayed out numbers represent the actual people in each group
○
PPP = 95%, pretty good
○
-
Example #2:
Assume 95% accuracy again
○
Assume 10,000 people, disorder has 5% prevalence
○
PPP = 50% , not as good this time even though your test appears to be
95% accurate
Half of the people you say have the condition, don't.
§
○
-
Example #3:
Assume 95% accuracy again
○
Assume 10,000 people, disorder has 1% prevalence
○
PPP = 16%, totally crap
80% of the cases where we say "this person has the disorder" were
wrong
§
○
-
The lower the probability of having a disorder, the more errors a test of
identification would have.
-
Example:
Prevalence of prostate cancer - 10 cases in every 10,000 males
○
Huge number will be falsely identified and go through a lot of
procedures/treatment for a disease they don't have
○
-
Sensitivity to Base Rate
** watch the podcast for this slide.
ROC Measures of Accuracy
ROC = Sensitivity/(1-Sensitivity)
-
ROC = H/FA
-
As response criterion changes, the curve changes.
We want a curve as close to the diagonal line as possible.
○
Ideally you want AUC to be high because it's a direct measure of an
individual's ability to determine whether someone has something.
○
-
AUC Measure of Accuracy
Large populations of criminals, you're going to give them a test to see if they are
at risk of violence.
-
We can follow them to see which ones became violence vs which ones didn't
-
AUC equivalent to the probability that you randomly selected a person who was
violent and randomly selected a non-violent individual, the violent individual
would have a higher test score.
-
AUC = p (V > NV)
-
Violence Risk Appraisal Guide (VRAG)
Harris, Rice & Quinsey (1993)
Predict violence among offenders with prior violent episodes.
-
685 violent or sexually violent Oak Ridge offenders.
-
Insanity acquittals and matched sample in Oak Ridge for one day between 1975
and 1978.
-
Measure: any criminal charge or chargeable.
If they had, they were judged to have violently re-offended.
○
-
Actuarial method - therefore, no interview with individual.
Just went through their file and looked at a variety of variables.
○
-
Childhood History:
Highest school grade
○
Teen alcohol abuse score
○
Elementary school maladjustment
○
Socioeconomic status
○
Childhood aggression
○
Behavioural problems
○
Suspended or expelled
○
Arrested under age 16
○
Separated from parents under age 16
○
Parental crime
○
Parental psychiatric history
○
Parental alcoholism
○
** all of these are risk factors for adult violence/aggression.
○
-
Adult Adjustment:
Longest employment (months)
○
Admissions to corrections
○
Psychiatric admissions
○
Alcohol abuse score
○
Impulsivity score
○
Property offense history
○
Violent offense history
○
Never married (higher risk than married)
○
Previous violent offense
○
Ever fired
○
Escaped from an institution
○
Failure on prior conditional release
○
-
Index Offense:
Age at index offense
○
Victim injury
○
Seriousness of index offense
○
Violent offense
○
Victim knew offender
○
Female victim
○
Weapon used
○
Sexual motive
○
Alcohol involved
○
-
Assessment Results:
Intelligence (IQ)
○
Level of supervision inventory (LSI)
○
Psychopathy Checklist (PCL)
○
Elevation on MMPI scale 4 (Pd)
○
DSM-III Schizophrenia (less likely to reoffend with schizophrenia)
○
DSM-III Personality Disorder
○
Pro-criminal values
○
Attitude unfavourable to convention
○
-
Variables chosen for VRAG:
Those most highly associated/correlated with violent reoffending.
○
Psychopathy checklist (0.34)
○
Elementary school maladjustment (0.31)
○
DSM-III Personality Disorder (0.26)
○
Age at index offense (-0.26)
○
Separated from parents under age 16 (0.25)
○
Failure on prior conditional release (0.24)
○
Property offense history (0.20)
○
Never married (0.18)
○
DSM-III Schizophrenia (-0.17)
○
Victim injury in index offense (-0.16)
The higher the rate of injury to the original victim, the lower the risk
of reoffending
§
○
Alcohol abuse history (0.13)
○
Female victim in index offense (-0.11)
○
-
SORAG
Quinsey et al (1998):
Extension of VRAG to assess risk o violence among past sexual offenders.
-
10 items from VRAG
-
4 additional items specific to sex offending
-
Strongly correlated with VRAG
-
Better at predicting violent (generic) recidivism than sexual recidivism.
True of all measures of violence.
○
-
AUC (anything about > 0.7 good, < 0.3 poor)
VRAG Validity
Harris et al (1993): Original Paper
0.44 between VRAG score and violent recidivism
-
AUC = 0.76 (0.60 for sexual) - really good
-
Rice and Harris (1997): 159 sex offenders, violence and sexual violence
Correlation = 0.47 for VRAG vs violence; corr = 0.20 for sexual
-
AUC = 0.77 for violent recidivism (0.62 for sexual)
-
Harris et al (2003): Four Canadian forensic samples x 4 instruments
VRAG vs SORAG vs RRASOR vs Static-99 = VRAG best
-
VRAG AUC = 0.73 for violent recidivisim (0.65 for sexual)
-
Critique of Actuarial Approach
Definitional differences across instruments:
"violence" defined differently
○
"sexual" offenses ranger even more, from rape to voyeurism and
exhibitionism.
○
-
Method gives risks for groups, not individual.
They assess risks for groups of individuals with matching
variables/characteristics.
○
Even if you, as an individual, share these characteristics, you might not
necessarily be more likely to offend - simply the group on average,
overall, has an increased risk.
○
-
Focus on 'static', long-term risk.
Historical variables won't change (i.e., past)
○
But there are things that can change, dynamic, that aren't considered.
○
-
Ignores dynamic, short-term risk factors.
-
Weak on info re: amelioration
Amelioration - what can we do - now that we know the risk factors
present - to prevent the person from violently offending.
○
-
Iterative Classification Tree
Steadman et al (2000):
Designed to try to tackle some of the concerns of the actuarial approach.
-
Why don't clinicians use actuarial methods?
Linear regression implies one method fits all.
○
Improvement in prediction not clinically significant.
○
There are two thresholds, not one: one for high risk, one for low risk.
Below the low threshold - low/no risk
§
Above the high threshold - high risk
§
○
There is a section between the high and low cutoffs that we cannot really
know.
Pretty large number of people.
§
○
-
ICT risk factors
Psychopathy - high predictability of violent reoffending
○
** don't need to know the rest of the risk factors - there are many.
○
-
How do we use these risk factors?
Started with the total sample (knew which ones were violent)
○
Divided them into high/low psychopathy based on PCL-R scores.
○
Split psychopathy into two groups based on whether they have a high or
low score with respect to child abuse.
○
From those with child abuse , divide into low/high drug abuse scores.
○
Go through the entire list of variables.
○
-
End up with a very low risk group and very high risk group.
But you'll also get a group of people who aren't assessed as either high or
low risk.
○
You will put those people back through the program.
○
-
How good is it?
Steadman et al (2000): 76% classified as high or low risk.
Left a 24% middle/fuzzy area
§
AUC = 0.82
§
○
Monahan et al (2005): 700 civi psychiatric patients
76% classified as high or low risk
§
AUC = 0.70
§
○
-
Method is new, not widely used because of computing demands.
Now, computers are very powerful and widely available, so it might be
more popular.
○
-
Structured Professional Judgement
Purposes:
Standardize how evaluations are conducted.
○
Standardize how variables are weighted.
○
Allow for modification by clinical judgement.
Once you've got a score, clinician still has to make a decision based
on experience/interviews.
§
○
-
Two main vehicles/instruments for this:
Historical/Clinical Risk Management - 20 (HCR-20)
And Sexual Violence Risk - 20 (SVR-20)
§
○
Level of Service Inventory - Revised (LSI-R)
○
-
Historical/Clinical Risk Management - 20
Webster, Douglas & Hart (Simon Fraser University)
Derek Eaves (B.C. Forensic Psychiatric Services)
20 variables organized into 3 scales:
Historical/scale (10 static variables)
○
Clinical Scale (5 items; current mental state)
○
Risk Management Scale (5 items; future environmental factors)
○
-
Variables:
Historical Scale (10)
Previous violence
§
Young age at first violence incident
§
Relationship instability
§
Employment problems
§
Substance use problems
§
Major mental illness
§
Psychopathy
§
Early maladjustment
§
Personality disorder
§
Prior supervision failure
§
○
Clinical Scale (5 variables)
Lack of insight
§
Negative attitudes
§
Active symptoms of major mental illness
§
Impulsivity
§
Unresponsive to treatment
§
○
Risk Management Scale (5 variables)
Plans lack feasibility
§
Exposure to destabilizers
§
Lack of personal support
§
Noncompliance with remediation attempts
§
Stress
§
○
-
Accuracy?
Douglas et al (2003): Violence risk for 100 patients released into
community
Inter-rater reliability (kappa) = 0.61
§
AUC (physical violence) = 0.70
§
AUC (any violence) = 0.67
§
○
Douglas et al (2005): Compared HCR-20, VRAG, PCL-R, VORA on predicting
violent reoffending.
HRC-20 and VRAG useful
§
PCL-R and VORA just OK (alone) - and inter-correlated
§
○
-
Level of Service Inventory-Revised (LSI-R)
Canadian, late 1970s
-
54 items grouped rationally into 10 scales
Developed for the reason of arranging appropriate systems for people
who are on parole/coming out of jai, but it does a really good job of
assessing violence risk.
○
Clinical assessment tool
○
-
Scale 1: Criminal History
Any prior convictions
○
Escape from a correctional institution
○
Punished for institutional misconduct
○
-
Scale 2: Education/Employment
-
Scale 3: Financial problems?
-
Scale 4: Family/Marital
-
Scale 5: Accommodations
-
Scale 6: Leisure/Recreation
-
Scale 7: Companions
-
Scale 8: Alcohol/Drug Problems
-
Scale 9: Emotion/Personal
-
Scale 10: Attitudes/Orientation
-
Assessment of Violence Risk
Friday, March 23, 2018
8:32 AM
Done typically after trial.
-
Relevant to post-trial decisions
Assessment of sentencing, parole
○
Parole decisions rely on assessment of violence risk
○
Decisions made within the prison system (i.e., counselling, programming,
holding for this particular individual)
○
-
Canadian psychologists have made the major contributions to the literature -
bulk from Ontario.
-
Baxtrom v Herold (1996) -U.S. Supreme Court
Baxtrom was a prisoner in the NY State prison system.
-
6 months before he was to be released, he was found to be in need of mental
health support and was moved to a mental health facility for criminals.
-
They held him in this facility long past the time of his release.
-
They didn't release him because they feared that he would be a risk to society.
-
Baxtrom sued on the basis of high rights being violated
He was hospitalized beyond his prison term and without a dangerousness
assessment.
○
-
The Supreme Court agreed that he should be released.
This released ~ 1000 prisoners moved to low security hospitals.
○
20% subsequently violent - much smaller than what would have been
predicted
○
18% discharged within a year
○
1% were later readmitted to secure hospitals because they committed a
violent criminal act.
○
-
Over 4.5 years:
50% of patients were released.
○
< 3% returned to secure hospitals.
○
-
Interesting considering the recidivism rate (30%).
Much lower recidivism rate for these people who were held against their
will on the basis of violent concerns, mental health concerns.
○
-
Thornberry and Jacoby (1979)
586 patients released from Pennsylvania institution (Dixon v Pennsylvania)
-
3-year return/recidivism rate = 23.7%
-
4-year violent arrest record = 15%
-
Quinsey & Ambtman (1979)
Wanted to see how good forensic psychiatrists were at assessing violence risk.
9 high school teachers and 4 forensic psychiatrists.
-
Evaluate recidivism risk, release, for 30 patients:
11 property offenders
○
9 child molesters
○
11 serious adult offenders (murder)
○
-
3 types of patient data:
Offense description
○
Patient history
○
Psychiatric assessments (IQ, MMPI, Rorschach, etc.)
○
-
Asked them to then predict reoffending, release, etc.
-
Inter-rater reliability low for both psychiatrists and teachers.
-
Inter-rater reliabilities similar for psychiatrists and teachers.
-
Psychiatrists rate offenders as more likely to commit offense than do teachers.
-
Little use of psychiatric assessments made by either group.
-
Summary:
"these date question the usefulness of psychiatric examinations in the
prediction of dangerousness. Perhaps psychiatric examinations would be
best restricted to determining whether an offender is treatable and
should not address the issue of dangerousness."
○
-
Violence of Assessment in the 1980s
Monahan (1981): "psychiatrist and psychologists are accurate in no more than
one out of three predictions of violent behaviour."
Book stopped research into violence assessment - he was extremely
critical of violence assessments.
○
-
Barefoot v.Estelle (1983):
Barefoot was a defendant incarcerated with mental health issues and he's
been denied release on the basis of the fact that he will be a risk to the
public.
○
He sues to have the psychiatric evaluation removed from his record since
we known psychiatrists are not good at assessing violence risk.
○
The APA agreed with him!
○
"we are not convinced that the view of the APA should be converted into
a constitutional rule barring an entire category of expert testimony …
neither petitioner nor the Association suggests that psychiatrists are
always wrong with respect to future dangerousness, only most of the
time."
○
Basically, they say that even though psychiatrists are terrible at assessing
future violence risk, we can still use it???? WHY
○
-
Violence Risk Assessment Research
Actuarial Methods: ignored the inaccurate clinician, and rely on statistics data.
Re: how insurance companies select your rates.
○
Determine how likely it is that people with the person's characteristics are
to violently offend.
○
-
Structured Professional Judgment: increase clinician's accuracy by standardizing
the assessment process.
-
Measures of Accuracy (binary decision making?)
Disorder = violence risk, presence of future/current/past disorder, etc.
-
Positive Predictive Power
Based on test results
○
Hits/(Hits + False Alarms)
○
Of the total of people we said had the disorder, what proportion actually
have it.
○
-
Negative Predictive Power
Correct Rejections/(Correct Rejections + Misses)
○
Of the total people we said did not have the disorder, what proportion
actually don't have it.
○
-
Sensitivity
S = H/(H+M)
○
Of all of the people with the disorder, what proportion did we say actually
had the disorder.
○
-
Specificity
Sp = R/(R + FA)
○
Of all of the people without the disorder, what proportion actually don't
have it.
○
-
Overall Accuracy
= R + H / (R + H + FA + M)
○
-
Heavily influenced to the base rate - frequency of occurrence in the population.
Number of correct choices we make is heavily affected by the base rate.
○
-
PPP: Sensitivity to Base Rate
Example #1:
Assume 95% accuracy
○
Assume 10,000 people, disorder has 50% prevalence
○
Grayed out numbers represent the actual people in each group
○
PPP = 95%, pretty good
○
-
Example #2:
Assume 95% accuracy again
○
Assume 10,000 people, disorder has 5% prevalence
○
PPP = 50% , not as good this time even though your test appears to be
95% accurate
Half of the people you say have the condition, don't.
§
○
-
Example #3:
Assume 95% accuracy again
○
Assume 10,000 people, disorder has 1% prevalence
○
PPP = 16%, totally crap
80% of the cases where we say "this person has the disorder" were
wrong
§
○
-
The lower the probability of having a disorder, the more errors a test of
identification would have.
-
Example:
Prevalence of prostate cancer - 10 cases in every 10,000 males
○
Huge number will be falsely identified and go through a lot of
procedures/treatment for a disease they don't have
○
-
Sensitivity to Base Rate
** watch the podcast for this slide.
ROC Measures of Accuracy
ROC = Sensitivity/(1-Sensitivity)
-
ROC = H/FA
-
As response criterion changes, the curve changes.
We want a curve as close to the diagonal line as possible.
○
Ideally you want AUC to be high because it's a direct measure of an
individual's ability to determine whether someone has something.
○
-
AUC Measure of Accuracy
Large populations of criminals, you're going to give them a test to see if they are
at risk of violence.
-
We can follow them to see which ones became violence vs which ones didn't
-
AUC equivalent to the probability that you randomly selected a person who was
violent and randomly selected a non-violent individual, the violent individual
would have a higher test score.
-
AUC = p (V > NV)
-
Violence Risk Appraisal Guide (VRAG)
Harris, Rice & Quinsey (1993)
Predict violence among offenders with prior violent episodes.
-
685 violent or sexually violent Oak Ridge offenders.
-
Insanity acquittals and matched sample in Oak Ridge for one day between 1975
and 1978.
-
Measure: any criminal charge or chargeable.
If they had, they were judged to have violently re-offended.
○
-
Actuarial method - therefore, no interview with individual.
Just went through their file and looked at a variety of variables.
○
-
Childhood History:
Highest school grade
○
Teen alcohol abuse score
○
Elementary school maladjustment
○
Socioeconomic status
○
Childhood aggression
○
Behavioural problems
○
Suspended or expelled
○
Arrested under age 16
○
Separated from parents under age 16
○
Parental crime
○
Parental psychiatric history
○
Parental alcoholism
○
** all of these are risk factors for adult violence/aggression.
○
-
Adult Adjustment:
Longest employment (months)
○
Admissions to corrections
○
Psychiatric admissions
○
Alcohol abuse score
○
Impulsivity score
○
Property offense history
○
Violent offense history
○
Never married (higher risk than married)
○
Previous violent offense
○
Ever fired
○
Escaped from an institution
○
Failure on prior conditional release
○
-
Index Offense:
Age at index offense
○
Victim injury
○
Seriousness of index offense
○
Violent offense
○
Victim knew offender
○
Female victim
○
Weapon used
○
Sexual motive
○
Alcohol involved
○
-
Assessment Results:
Intelligence (IQ)
○
Level of supervision inventory (LSI)
○
Psychopathy Checklist (PCL)
○
Elevation on MMPI scale 4 (Pd)
○
DSM-III Schizophrenia (less likely to reoffend with schizophrenia)
○
DSM-III Personality Disorder
○
Pro-criminal values
○
Attitude unfavourable to convention
○
-
Variables chosen for VRAG:
Those most highly associated/correlated with violent reoffending.
○
Psychopathy checklist (0.34)
○
Elementary school maladjustment (0.31)
○
DSM-III Personality Disorder (0.26)
○
Age at index offense (-0.26)
○
Separated from parents under age 16 (0.25)
○
Failure on prior conditional release (0.24)
○
Property offense history (0.20)
○
Never married (0.18)
○
DSM-III Schizophrenia (-0.17)
○
Victim injury in index offense (-0.16)
The higher the rate of injury to the original victim, the lower the risk
of reoffending
§
○
Alcohol abuse history (0.13)
○
Female victim in index offense (-0.11)
○
-
SORAG
Quinsey et al (1998):
Extension of VRAG to assess risk o violence among past sexual offenders.
-
10 items from VRAG
-
4 additional items specific to sex offending
-
Strongly correlated with VRAG
-
Better at predicting violent (generic) recidivism than sexual recidivism.
True of all measures of violence.
○
-
AUC (anything about > 0.7 good, < 0.3 poor)
VRAG Validity
Harris et al (1993): Original Paper
0.44 between VRAG score and violent recidivism
-
AUC = 0.76 (0.60 for sexual) - really good
-
Rice and Harris (1997): 159 sex offenders, violence and sexual violence
Correlation = 0.47 for VRAG vs violence; corr = 0.20 for sexual
-
AUC = 0.77 for violent recidivism (0.62 for sexual)
-
Harris et al (2003): Four Canadian forensic samples x 4 instruments
VRAG vs SORAG vs RRASOR vs Static-99 = VRAG best
-
VRAG AUC = 0.73 for violent recidivisim (0.65 for sexual)
-
Critique of Actuarial Approach
Definitional differences across instruments:
"violence" defined differently
○
"sexual" offenses ranger even more, from rape to voyeurism and
exhibitionism.
○
-
Method gives risks for groups, not individual.
They assess risks for groups of individuals with matching
variables/characteristics.
○
Even if you, as an individual, share these characteristics, you might not
necessarily be more likely to offend - simply the group on average,
overall, has an increased risk.
○
-
Focus on 'static', long-term risk.
Historical variables won't change (i.e., past)
○
But there are things that can change, dynamic, that aren't considered.
○
-
Ignores dynamic, short-term risk factors.
-
Weak on info re: amelioration
Amelioration - what can we do - now that we know the risk factors
present - to prevent the person from violently offending.
○
-
Iterative Classification Tree
Steadman et al (2000):
Designed to try to tackle some of the concerns of the actuarial approach.
-
Why don't clinicians use actuarial methods?
Linear regression implies one method fits all.
○
Improvement in prediction not clinically significant.
○
There are two thresholds, not one: one for high risk, one for low risk.
Below the low threshold - low/no risk
§
Above the high threshold - high risk
§
○
There is a section between the high and low cutoffs that we cannot really
know.
Pretty large number of people.
§
○
-
ICT risk factors
Psychopathy - high predictability of violent reoffending
○
** don't need to know the rest of the risk factors - there are many.
○
-
How do we use these risk factors?
Started with the total sample (knew which ones were violent)
○
Divided them into high/low psychopathy based on PCL-R scores.
○
Split psychopathy into two groups based on whether they have a high or
low score with respect to child abuse.
○
From those with child abuse , divide into low/high drug abuse scores.
○
Go through the entire list of variables.
○
-
End up with a very low risk group and very high risk group.
But you'll also get a group of people who aren't assessed as either high or
low risk.
○
You will put those people back through the program.
○
-
How good is it?
Steadman et al (2000): 76% classified as high or low risk.
Left a 24% middle/fuzzy area
§
AUC = 0.82
§
○
Monahan et al (2005): 700 civi psychiatric patients
76% classified as high or low risk
§
AUC = 0.70
§
○
-
Method is new, not widely used because of computing demands.
Now, computers are very powerful and widely available, so it might be
more popular.
○
-
Structured Professional Judgement
Purposes:
Standardize how evaluations are conducted.
○
Standardize how variables are weighted.
○
Allow for modification by clinical judgement.
Once you've got a score, clinician still has to make a decision based
on experience/interviews.
§
○
-
Two main vehicles/instruments for this:
Historical/Clinical Risk Management - 20 (HCR-20)
And Sexual Violence Risk - 20 (SVR-20)
§
○
Level of Service Inventory - Revised (LSI-R)
○
-
Historical/Clinical Risk Management - 20
Webster, Douglas & Hart (Simon Fraser University)
Derek Eaves (B.C. Forensic Psychiatric Services)
20 variables organized into 3 scales:
Historical/scale (10 static variables)
○
Clinical Scale (5 items; current mental state)
○
Risk Management Scale (5 items; future environmental factors)
○
-
Variables:
Historical Scale (10)
Previous violence
§
Young age at first violence incident
§
Relationship instability
§
Employment problems
§
Substance use problems
§
Major mental illness
§
Psychopathy
§
Early maladjustment
§
Personality disorder
§
Prior supervision failure
§
○
Clinical Scale (5 variables)
Lack of insight
§
Negative attitudes
§
Active symptoms of major mental illness
§
Impulsivity
§
Unresponsive to treatment
§
○
Risk Management Scale (5 variables)
Plans lack feasibility
§
Exposure to destabilizers
§
Lack of personal support
§
Noncompliance with remediation attempts
§
Stress
§
○
-
Accuracy?
Douglas et al (2003): Violence risk for 100 patients released into
community
Inter-rater reliability (kappa) = 0.61
§
AUC (physical violence) = 0.70
§
AUC (any violence) = 0.67
§
○
Douglas et al (2005): Compared HCR-20, VRAG, PCL-R, VORA on predicting
violent reoffending.
HRC-20 and VRAG useful
§
PCL-R and VORA just OK (alone) - and inter-correlated
§
○
-
Level of Service Inventory-Revised (LSI-R)
Canadian, late 1970s
-
54 items grouped rationally into 10 scales
Developed for the reason of arranging appropriate systems for people
who are on parole/coming out of jai, but it does a really good job of
assessing violence risk.
○
Clinical assessment tool
○
-
Scale 1: Criminal History
Any prior convictions
○
Escape from a correctional institution
○
Punished for institutional misconduct
○
-
Scale 2: Education/Employment
-
Scale 3: Financial problems?
-
Scale 4: Family/Marital
-
Scale 5: Accommodations
-
Scale 6: Leisure/Recreation
-
Scale 7: Companions
-
Scale 8: Alcohol/Drug Problems
-
Scale 9: Emotion/Personal
-
Scale 10: Attitudes/Orientation
-
Assessment of Violence Risk
Friday, March 23, 2018 8:32 AM
Done typically after trial.
-
Relevant to post-trial decisions
Assessment of sentencing, parole
○
Parole decisions rely on assessment of violence risk
○
Decisions made within the prison system (i.e., counselling, programming,
holding for this particular individual)
○
-
Canadian psychologists have made the major contributions to the literature -
bulk from Ontario.
-
Baxtrom v Herold (1996) -U.S. Supreme Court
Baxtrom was a prisoner in the NY State prison system.
-
6 months before he was to be released, he was found to be in need of mental
health support and was moved to a mental health facility for criminals.
-
They held him in this facility long past the time of his release.
-
They didn't release him because they feared that he would be a risk to society.
-
Baxtrom sued on the basis of high rights being violated
He was hospitalized beyond his prison term and without a dangerousness
assessment.
○
-
The Supreme Court agreed that he should be released.
This released ~ 1000 prisoners moved to low security hospitals.
○
20% subsequently violent - much smaller than what would have been
predicted
○
18% discharged within a year
○
1% were later readmitted to secure hospitals because they committed a
violent criminal act.
○
-
Over 4.5 years:
50% of patients were released.
○
< 3% returned to secure hospitals.
○
-
Interesting considering the recidivism rate (30%).
Much lower recidivism rate for these people who were held against their
will on the basis of violent concerns, mental health concerns.
○
-
Thornberry and Jacoby (1979)
586 patients released from Pennsylvania institution (Dixon v Pennsylvania)
-
3-year return/recidivism rate = 23.7%
-
4-year violent arrest record = 15%
-
Quinsey & Ambtman (1979)
Wanted to see how good forensic psychiatrists were at assessing violence risk.
9 high school teachers and 4 forensic psychiatrists.
-
Evaluate recidivism risk, release, for 30 patients:
11 property offenders
○
9 child molesters
○
11 serious adult offenders (murder)
○
-
3 types of patient data:
Offense description
○
Patient history
○
Psychiatric assessments (IQ, MMPI, Rorschach, etc.)
○
-
Asked them to then predict reoffending, release, etc.
-
Inter-rater reliability low for both psychiatrists and teachers.
-
Inter-rater reliabilities similar for psychiatrists and teachers.
-
Psychiatrists rate offenders as more likely to commit offense than do teachers.
-
Little use of psychiatric assessments made by either group.
-
Summary:
"these date question the usefulness of psychiatric examinations in the
prediction of dangerousness. Perhaps psychiatric examinations would be
best restricted to determining whether an offender is treatable and
should not address the issue of dangerousness."
○
-
Violence of Assessment in the 1980s
Monahan (1981): "psychiatrist and psychologists are accurate in no more than
one out of three predictions of violent behaviour."
Book stopped research into violence assessment - he was extremely
critical of violence assessments.
○
-
Barefoot v.Estelle (1983):
Barefoot was a defendant incarcerated with mental health issues and he's
been denied release on the basis of the fact that he will be a risk to the
public.
○
He sues to have the psychiatric evaluation removed from his record since
we known psychiatrists are not good at assessing violence risk.
○
The APA agreed with him!
○
"we are not convinced that the view of the APA should be converted into
a constitutional rule barring an entire category of expert testimony …
neither petitioner nor the Association suggests that psychiatrists are
always wrong with respect to future dangerousness, only most of the
time."
○
Basically, they say that even though psychiatrists are terrible at assessing
future violence risk, we can still use it???? WHY
○
-
Violence Risk Assessment Research
Actuarial Methods: ignored the inaccurate clinician, and rely on statistics data.
Re: how insurance companies select your rates.
○
Determine how likely it is that people with the person's characteristics are
to violently offend.
○
-
Structured Professional Judgment: increase clinician's accuracy by standardizing
the assessment process.
-
Measures of Accuracy (binary decision making?)
Disorder = violence risk, presence of future/current/past disorder, etc.
-
Positive Predictive Power
Based on test results
○
Hits/(Hits + False Alarms)
○
Of the total of people we said had the disorder, what proportion actually
have it.
○
-
Negative Predictive Power
Correct Rejections/(Correct Rejections + Misses)
○
Of the total people we said did not have the disorder, what proportion
actually don't have it.
○
-
Sensitivity
S = H/(H+M)
○
Of all of the people with the disorder, what proportion did we say actually
had the disorder.
○
-
Specificity
Sp = R/(R + FA)
○
Of all of the people without the disorder, what proportion actually don't
have it.
○
-
Overall Accuracy
= R + H / (R + H + FA + M)
○
-
Heavily influenced to the base rate - frequency of occurrence in the population.
Number of correct choices we make is heavily affected by the base rate.
○
-
PPP: Sensitivity to Base Rate
Example #1:
Assume 95% accuracy
○
Assume 10,000 people, disorder has 50% prevalence
○
Grayed out numbers represent the actual people in each group
○
PPP = 95%, pretty good
○
-
Example #2:
Assume 95% accuracy again
○
Assume 10,000 people, disorder has 5% prevalence
○
PPP = 50% , not as good this time even though your test appears to be
95% accurate
Half of the people you say have the condition, don't.
§
○
-
Example #3:
Assume 95% accuracy again
○
Assume 10,000 people, disorder has 1% prevalence
○
PPP = 16%, totally crap
80% of the cases where we say "this person has the disorder" were
wrong
§
○
-
The lower the probability of having a disorder, the more errors a test of
identification would have.
-
Example:
Prevalence of prostate cancer - 10 cases in every 10,000 males
○
Huge number will be falsely identified and go through a lot of
procedures/treatment for a disease they don't have
○
-
Sensitivity to Base Rate
** watch the podcast for this slide.
ROC Measures of Accuracy
ROC = Sensitivity/(1-Sensitivity)
-
ROC = H/FA
-
As response criterion changes, the curve changes.
We want a curve as close to the diagonal line as possible.
○
Ideally you want AUC to be high because it's a direct measure of an
individual's ability to determine whether someone has something.
○
-
AUC Measure of Accuracy
Large populations of criminals, you're going to give them a test to see if they are
at risk of violence.
-
We can follow them to see which ones became violence vs which ones didn't
-
AUC equivalent to the probability that you randomly selected a person who was
violent and randomly selected a non-violent individual, the violent individual
would have a higher test score.
-
AUC = p (V > NV)
-
Violence Risk Appraisal Guide (VRAG)
Harris, Rice & Quinsey (1993)
Predict violence among offenders with prior violent episodes.
-
685 violent or sexually violent Oak Ridge offenders.
-
Insanity acquittals and matched sample in Oak Ridge for one day between 1975
and 1978.
-
Measure: any criminal charge or chargeable.
If they had, they were judged to have violently re-offended.
○
-
Actuarial method - therefore, no interview with individual.
Just went through their file and looked at a variety of variables.
○
-
Childhood History:
Highest school grade
○
Teen alcohol abuse score
○
Elementary school maladjustment
○
Socioeconomic status
○
Childhood aggression
○
Behavioural problems
○
Suspended or expelled
○
Arrested under age 16
○
Separated from parents under age 16
○
Parental crime
○
Parental psychiatric history
○
Parental alcoholism
○
** all of these are risk factors for adult violence/aggression.
○
-
Adult Adjustment:
Longest employment (months)
○
Admissions to corrections
○
Psychiatric admissions
○
Alcohol abuse score
○
Impulsivity score
○
Property offense history
○
Violent offense history
○
Never married (higher risk than married)
○
Previous violent offense
○
Ever fired
○
Escaped from an institution
○
Failure on prior conditional release
○
-
Index Offense:
Age at index offense
○
Victim injury
○
Seriousness of index offense
○
Violent offense
○
Victim knew offender
○
Female victim
○
Weapon used
○
Sexual motive
○
Alcohol involved
○
-
Assessment Results:
Intelligence (IQ)
○
Level of supervision inventory (LSI)
○
Psychopathy Checklist (PCL)
○
Elevation on MMPI scale 4 (Pd)
○
DSM-III Schizophrenia (less likely to reoffend with schizophrenia)
○
DSM-III Personality Disorder
○
Pro-criminal values
○
Attitude unfavourable to convention
○
-
Variables chosen for VRAG:
Those most highly associated/correlated with violent reoffending.
○
Psychopathy checklist (0.34)
○
Elementary school maladjustment (0.31)
○
DSM-III Personality Disorder (0.26)
○
Age at index offense (-0.26)
○
Separated from parents under age 16 (0.25)
○
Failure on prior conditional release (0.24)
○
Property offense history (0.20)
○
Never married (0.18)
○
DSM-III Schizophrenia (-0.17)
○
Victim injury in index offense (-0.16)
The higher the rate of injury to the original victim, the lower the risk
of reoffending
§
○
Alcohol abuse history (0.13)
○
Female victim in index offense (-0.11)
○
-
SORAG
Quinsey et al (1998):
Extension of VRAG to assess risk o violence among past sexual offenders.
-
10 items from VRAG
-
4 additional items specific to sex offending
-
Strongly correlated with VRAG
-
Better at predicting violent (generic) recidivism than sexual recidivism.
True of all measures of violence.
○
-
AUC (anything about > 0.7 good, < 0.3 poor)
VRAG Validity
Harris et al (1993): Original Paper
0.44 between VRAG score and violent recidivism
-
AUC = 0.76 (0.60 for sexual) - really good
-
Rice and Harris (1997): 159 sex offenders, violence and sexual violence
Correlation = 0.47 for VRAG vs violence; corr = 0.20 for sexual
-
AUC = 0.77 for violent recidivism (0.62 for sexual)
-
Harris et al (2003): Four Canadian forensic samples x 4 instruments
VRAG vs SORAG vs RRASOR vs Static-99 = VRAG best
-
VRAG AUC = 0.73 for violent recidivisim (0.65 for sexual)
-
Critique of Actuarial Approach
Definitional differences across instruments:
"violence" defined differently
○
"sexual" offenses ranger even more, from rape to voyeurism and
exhibitionism.
○
-
Method gives risks for groups, not individual.
They assess risks for groups of individuals with matching
variables/characteristics.
○
Even if you, as an individual, share these characteristics, you might not
necessarily be more likely to offend - simply the group on average,
overall, has an increased risk.
○
-
Focus on 'static', long-term risk.
Historical variables won't change (i.e., past)
○
But there are things that can change, dynamic, that aren't considered.
○
-
Ignores dynamic, short-term risk factors.
-
Weak on info re: amelioration
Amelioration - what can we do - now that we know the risk factors
present - to prevent the person from violently offending.
○
-
Iterative Classification Tree
Steadman et al (2000):
Designed to try to tackle some of the concerns of the actuarial approach.
-
Why don't clinicians use actuarial methods?
Linear regression implies one method fits all.
○
Improvement in prediction not clinically significant.
○
There are two thresholds, not one: one for high risk, one for low risk.
Below the low threshold - low/no risk
§
Above the high threshold - high risk
§
○
There is a section between the high and low cutoffs that we cannot really
know.
Pretty large number of people.
§
○
-
ICT risk factors
Psychopathy - high predictability of violent reoffending
○
** don't need to know the rest of the risk factors - there are many.
○
-
How do we use these risk factors?
Started with the total sample (knew which ones were violent)
○
Divided them into high/low psychopathy based on PCL-R scores.
○
Split psychopathy into two groups based on whether they have a high or
low score with respect to child abuse.
○
From those with child abuse , divide into low/high drug abuse scores.
○
Go through the entire list of variables.
○
-
End up with a very low risk group and very high risk group.
But you'll also get a group of people who aren't assessed as either high or
low risk.
○
You will put those people back through the program.
○
-
How good is it?
Steadman et al (2000): 76% classified as high or low risk.
Left a 24% middle/fuzzy area
§
AUC = 0.82
§
○
Monahan et al (2005): 700 civi psychiatric patients
76% classified as high or low risk
§
AUC = 0.70
§
○
-
Method is new, not widely used because of computing demands.
Now, computers are very powerful and widely available, so it might be
more popular.
○
-
Structured Professional Judgement
Purposes:
Standardize how evaluations are conducted.
○
Standardize how variables are weighted.
○
Allow for modification by clinical judgement.
Once you've got a score, clinician still has to make a decision based
on experience/interviews.
§
○
-
Two main vehicles/instruments for this:
Historical/Clinical Risk Management - 20 (HCR-20)
And Sexual Violence Risk - 20 (SVR-20)
§
○
Level of Service Inventory - Revised (LSI-R)
○
-
Historical/Clinical Risk Management - 20
Webster, Douglas & Hart (Simon Fraser University)
Derek Eaves (B.C. Forensic Psychiatric Services)
20 variables organized into 3 scales:
Historical/scale (10 static variables)
○
Clinical Scale (5 items; current mental state)
○
Risk Management Scale (5 items; future environmental factors)
○
-
Variables:
Historical Scale (10)
Previous violence
§
Young age at first violence incident
§
Relationship instability
§
Employment problems
§
Substance use problems
§
Major mental illness
§
Psychopathy
§
Early maladjustment
§
Personality disorder
§
Prior supervision failure
§
○
Clinical Scale (5 variables)
Lack of insight
§
Negative attitudes
§
Active symptoms of major mental illness
§
Impulsivity
§
Unresponsive to treatment
§
○
Risk Management Scale (5 variables)
Plans lack feasibility
§
Exposure to destabilizers
§
Lack of personal support
§
Noncompliance with remediation attempts
§
Stress
§
○
-
Accuracy?
Douglas et al (2003): Violence risk for 100 patients released into
community
Inter-rater reliability (kappa) = 0.61
§
AUC (physical violence) = 0.70
§
AUC (any violence) = 0.67
§
○
Douglas et al (2005): Compared HCR-20, VRAG, PCL-R, VORA on predicting
violent reoffending.
HRC-20 and VRAG useful
§
PCL-R and VORA just OK (alone) - and inter-correlated
§
○
-
Level of Service Inventory-Revised (LSI-R)
Canadian, late 1970s
-
54 items grouped rationally into 10 scales
Developed for the reason of arranging appropriate systems for people
who are on parole/coming out of jai, but it does a really good job of
assessing violence risk.
○
Clinical assessment tool
○
-
Scale 1: Criminal History
Any prior convictions
○
Escape from a correctional institution
○
Punished for institutional misconduct
○
-
Scale 2: Education/Employment
-
Scale 3: Financial problems?
-
Scale 4: Family/Marital
-
Scale 5: Accommodations
-
Scale 6: Leisure/Recreation
-
Scale 7: Companions
-
Scale 8: Alcohol/Drug Problems
-
Scale 9: Emotion/Personal
-
Scale 10: Attitudes/Orientation
-
Assessment of Violence Risk
Friday, March 23, 2018 8:32 AM
Document Summary
Parole decisions rely on assessment of violence risk. Decisions made within the prison system (i. e. , counselling, programming, holding for this particular individual) Canadian psychologists have made the major contributions to the literature - bulk from ontario. Baxtrom v herold (1996) - u. s. supreme court. Baxtrom was a prisoner in the ny state prison system. 6 months before he was to be released, he was found to be in need of mental health support and was moved to a mental health facility for criminals. They held him in this facility long past the time of his release. They didn"t release him because they feared that he would be a risk to society. Baxtrom sued on the basis of high rights being violated. He was hospitalized beyond his prison term and without a dangerousness assessment. The supreme court agreed that he should be released. This released ~ 1000 prisoners moved to low security hospitals.
