Transcript Pathways to Desistance

The difficulties of
predicting future violence
Edward P. Mulvey, Ph.D.
Western Psychiatric Institute and Clinic
University of Pittsburgh School of Medicine
[email protected]
Conference on Campus Violence
Columbia Law School
April 4, 2008
Goals

Provide some background about general
methods for making predictions of violence

Identify the inherent challenges to
predicting campus violence

Make some general recommendations about
strategies for addressing this problem
Approaches to Predicting
Incidents of Violence
Actuarial Approach

Same information considered for every
person

Consistent method for combining
information

Factors considered not necessarily “causal”
for that individual
Clinical Approach

Individualized judgment

Range of relevant factors considered is very
broad; ideally, an integrative view

Operates from a theory of how violence
might occur or unfold in the individual’s life
Problems with the Actuarial Approach

All the information has to be available

Generally assumes that all the factors apply
the same to everyone

The risk estimate is devoid of theory
Problems with the Clinical Approach

Depends on the person doing it

Variability in:



Human biases occur



Information gathered
Ways information is combined
Recency
Vividness
Affected by organizational demands


Optimization is assumed
“Satisficing” is more common
Actuarial vs. Clinical Prediction

Old debate
Suicide
 Job/academic success


General findings
Actuarial Approach generally more accurate
 Clinical Approach more versatile


Issue of reliability and validity
Reliable, but not valid
Valid, but not reliable
Actuarial Risk Assessment Tools

General violence/recidivism (mainly in mentally
ill individuals)





Historical-Clinical-Risk Management-20 (HCR-20)
Violence Risk Appraisal Guide (VRAG)
Classification of Violence Risk (COVR)
Violent Offender Risk Assessment Scale (VORAS)
Special purpose instruments

Domestic violence


Risk of sex offense


Spousal Assault Risk Assessment (SARA)
Static-99
Risk of violence among juveniles


Early Assessment Risk List for Boys (EARL-20B)
Manual for the Structured Assessment of Violence Risk in
Youth (SAVRY)
Issues regarding actuarial instruments

Increasingly popular because of technology

Optimization on chance



Shrinkage inevitable

Application on particular sample might not be
appropriate
Effect of context

Information availability

Outcome of decision
Not a replacement for clinical judgment. Integration of
actuarial and clinical information is the goal
Inherent Challenges to Predicting
Campus Violence
Major General John Sedgwick
Problem #1: Low Base Rates
true
positives
Predicted
Violent
Yes
No
false
negatives
Actually violent
false
positives
YesNo
80
180 260
20
720 740 true
negatives
100
900 1,000
Assume that one in ten individuals is actually violent over a given time period
Assume that the instrument correctly identifies 80% of the violent individuals
80% of the nonviolent individuals
Implications

No technology will predict rare events



Secret Service study of school shootings: “There is no
accurate or useful profile of the school shooter”.
“profiles” may be valuable, but not because they are
predictive
The utility of screening and assessment is not to
predict for an individual, but


to identify groups with higher base rates
to focus prevention resources
Problem #1: Low Base Rates
Problem #2: Context matters and
situations change
Problem #2: Context matters and
situations change

Violence is usually
dependent on proximal situational factors
 transactional


Opportunities for violence
may differ substantially across individuals
 can be altered by lifestyle changes

presence of alcohol
 living arrangements

Implications

Risk status may be important, but so are
fluctuations in risk state
Move toward management of high risk
individuals and situations
 Conditional prediction model

“if….then” formulation of risk
 Monitoring and management of “dynamic
predictors”

A research example:
Substance use and violence
Study Design

Select group of individuals who were highly likely to have
frequent involvement in violence

Weekly interviews providing daily reports
 Violent incidents
 Substance use reports at daily level
 Alcohol (number of drinks)
 Marijuana use
 Other drugs (mostly cocaine)

Analyses of the concurrent and lagged relationships of
substance use and violence
Serious
violence
Any
violence
Alcohol (>
3 drinks)
1.4%
3.9%
8.7%
Percent of days
Marijuana Other drugs
20.6%
2.5%
Conditional Probability
SERIOUS VIOLENCE
Conditional Probability
ANY VIOLENCE
P (violence| alcohol only)
2.3 %
7.9%
P (violence| marijuana only )
1.6 %
4.2%
P (violence| other drug only)
2.5 %
7.4%
P (violence| alcohol and mj)
4.4 %
10.5%
P (violence| alcohol and otherdrugs)
4.5 %
10.2%
P (violence| alcohol, mj, otherdrugs)
9.0 %
19.4%
P (alcohol | violence)
21.7 %
20.2%
P (marijuana | violence)
32.4 %
28.0%
P (other drugs| violence)
6.2 %
5.5%
P (alcohol & any drug use)
9.3 %
9.3%
P (alcohol & any drug use| violence)
23.1 %
21.0%
Odds ratios for substance use and violence
one day apart for serious violence
Day After
Serious
Violence
Alcohol
Marijuana
Other
Drugs
Serious Violence
5.4
1.9
1.5
2.1
Alcohol
2.4
9.5
2.1
2.8
Marijuana
1.6
2.3
31.5
1.5
Other Drug
1.5
2.2
1.5
48.1
Day Before
Examples
Case 8
--------|||||||-|----------------------------------------------------|---|-----
Case 2080
-|------|------------|------|---------------|-|------|--------------|---------|--|--------|----|--
Testing the relation of violence and
alcohol use over time
Findings

Evidence for a lagged effect for alcohol use
(greater than three drinks) on violence, but not the
other way around

No significant lagged relationships either way for
marijuana use or other drugs

Even controlling for different types of substance
use, violence on one day predicts for the next day

Use of multiple substances on prior day also
increases likelihood of violence
Problem #1: Low Base Rates
Problem #2: Context matters and
situations change
Problem #3: Late adolescence is all
about change
Problem #3: Late adolescence is all
about change

Late adolescence brings:





Onset period for many mental disorders




Independent social roles
“trying on” lifestyles
ongoing brain development
Different patterns of substance use
Substance abuse: 20 years old
Mood disorders: 30 years old
Schizophrenia: 20 years old (males); 30 years old
(females)
Involvement in violence drops off, even in serious
offenders
Self Report Variety
Score
Self Reported Offending
Serious Adolescent Offenders - males only –
average age 16 at first interview
Group 5
(8.5%)
8
6
Group 4
(15.1%)
4
Group 3
(18.3%)
2
0
Group 2
(33.8%)
0
Group 1
(24.2%)
6
12
18
24
30
Months after Initial Interview
36
Implications

Lack of solid history for use in making judgments
about future violence

Mix of developmental features and valid
symptoms


Impulsiveness, moodiness, feelings of being picked on,
feelings of rejection, tendency to blame others
Diagnostic labels are less valid

Likely to confuse risk markers with risk factors

Assessments have a limited shelf life
Guidelines for structuring risk
assessments
1. Get the right information on
the right individuals

Screen using structured measures, and
assess for risk when warranted

Use all available information consistently

Use relevant assessment tools



Group characteristics
Target behavior of interest
Use actuarial instrument as an “anchor”
Best Bets for individual assessment









history of violence
impulsivity (process from ideation to action)
active ideation (mainly hostility and anger)
drug and alcohol use
psychopathy
perceived threat
plan/access to means
opportunities for violent encounters
coping strategies
2. Consider history in detail


When there is past violence, assess:
 what happened
 what factors explain the incident
 what factors protect against violence
 which risk and protective factors are currently in
effect or likely to be in effect
When there is no past violence, assess:
 “close calls”
 recent changes in life that may exceed coping
capacity
 risk factors for violence, based on appropriate
assessment protocols
3. For prevention efforts, distinguish
between risk markers and risk factors

Key is to focus on causal, dynamic risk factors

Questions




Is this a risk factor for a particular type of violence?
Are there conditions that influence the relationship
between the risk factor and violence?
Does the risk factor play a causal role in violence?
If so, is the risk factor capable of being modified?
4. Take context and risk state
seriously

A single risk assessment is useless without a
management plan

Assess individual factors and conditions
periodically in high risk cases

Create an environment where students trust
authorities enough to share information