Lecture 9: Risk Assessment Flashcards
HCR-20
An example of a structured professional judgment instrument for carrying out risk assessment. Twenty risk factors are coded across 3 domains - historical (static) factors, clinical (dynamic) factors, and risk management factors. Based on a patient’s combined score across these factors the patient is assigned to a general level of risk - low, moderate, or high.
risk assessment
A concept involving two components: (1) risk prediction - assessing the risk that people will commit violence in the future, and (2) risk management - developing effective intervention strategies to manage that risk.
ROC analysis
Stands for receiver operating characteristics analysis. Used as a procedure for measuring the accuracy of risk predictions. The height of the ROC curve, as measured by the area under the curve (AUC), indicates accuracy.
two by two contingency table
A method for recording the frequency of possible outcomes that can occur when making two alternative decisions, like predicting whether a patient will be violent or not. The decision outcomes included in this table include hits (true positives), false alarms (false positives), misses (false negatives), and correct rejections (true negatives).
VRAG
An example of an actuarial technique for carrying out risk assessment. Twelve static risk factors are coded, and based on a patient’s combined score across these weighted factors, the patient is assigned to a specific level of risk.
why should we care about risk assesment?
Risk assessment informs sentencing, classification, treatment needs, treatment intensity, parole decisions, level of supervision, notification decisions, release conditions, etc.
risk prediction
assess the risk that people will commit violence in the future
risk management
develop effective intervention strategies to manage that risk
goals of risk assessment
- Improve accuracy
- Improve transparency
- Improve consistency
what do we consider in a risk assessment?
risk factors
risk factor
a variable that is related to recidivism
static risk factors
- Fixed and unchanging
- Most convenient
- Most frequently used
- Can be reliably measured
- Are very predictive
examples of static risk factor
demographic variables, history of criminal behaviour, history of mental disorder
dynamic risk factors
- Change with time
- Less convenient and reliable
- Less frequently used
- Sensitive to change
- With intervention, we can change the level of risk
2 main types of dynamic risk factors
stable & acute dynamic
acute dynamic
rapidly fluctuating
stable dynamic
persistent and change slowly, if at all
big 4 risk factors
- Criminal history
- Procriminal personality (impulsive, aggressive)
- Procriminal attitudes
- Procriminal associates
not risk factors
- Low SES
- Personal distress/psychopathology
- Includes low self-esteem or depression
- Fear of punishment
- Verbal intelligence
- Remorse/empathy
- Offence severity
approaches to risk assessment
- Unstructured clinical judgment
- Actuarial tools
- Structured professional judgment
unstructured clinical judgment
- Subjectively select, analyze, and interpret risk factors
- No longer used today
advantages of unstructured clinical judgment
- Flexible
- Idiographic
disadvantages of unstructured clinical judgment
- Inconsistent
- Low accuracy
actuarial tools
Collect pre-specified risk factors and enter them into a statistical model that combines and weights them
advantages of actuarial tools
- Consistent
- High accuracy
disadvantages of actuarial tools
- Nomothetic
- Validity across different samples
improving accuracy
we want our predictions to be correct (i.e. they were released and did not re-offend or were retained and would have re-offended)
improving transparency
the offender and the victim deserve to know why the decision to release to retain the offender was made
improving consistency
use systematic methods for determining who gets let out and who doesn’t
VRAG
- Consists of 12 weighted (according to their predictive power) static risk factors
- Added together to give overall probability of risk
12 VRAG factors
- PCL-R score (+)
- Elementary school problems (+)
- Personality disorder (+)
- Separated from parents (+)
- Failure on prior release (+)
- Alcohol abuse (+)
- Nonviolent offence history (+)
- Never married (+)
- Schizophrenia (-)
- Victim injury (-)
- Female victim (-)
- Age (-)
administration of the VRAG
- Code the presence of risk factors
- Total the score
- Assign the individual to 1 of 9 bins (based on the probability of offending)
- 1 is the lowest risk and 9 is the highest
- Estimate the probability of violence
structured professional judgment
- Collect pre-specified risk factors while adding in any case-specific details
- The final assessment of risk is clinical judgment (informed by empirical risk factors)
advantages of structured professional judgment
- Flexible
- Nomothetic-idiographic
disadvantages of structured professional judgment
- Moderate accuracy (clinical judgment)
- Less consistent than actuarial
SPJ Example: HCR-20
- 10 historical factors
- 5 clinical factors
- 5 risk management factors
- Any other case-specific factors
example of a historical factor
previous violence
example of a clinical factor
lack of insight
example of a risk management factor
Plans lack feasibility
Administration of the HCR-20
- Code the presence of risk factors
- Code case-specific risk factors
- Subjectively decide on the level of risk
risk ratings of the HCR-20
- Low risk: monitor and intervene with low priority and intensity
- Mid risk: monitor and intervene with some priority and intensity
- High risk: monitor and intervene with high priority and intensity
what is the most important factor when evaluating risk assessment tools?
predictive accuracy: Does it predict recidivism? Do the high-risk offenders re-offend more than the low-risk offenders?
other important factors when evaluating risk assessment tools
easy to use easy to train people on, how much time it takes, construct validity, and consistent (inter-rater reliability)
Receiver operating characteristic (ROC) analysis
a technique for measuring the accuracy of risk assessments by examining false positives and true positives across decision thresholds
decision outcomes of a risk assessment
- true positive (truth = violent & prediction = violent)
- false positive (truth = not violent & prediction = violent)
- false negative (truth = violent & prediction = not violent)
- true negative (truth = not violent & prediction = not violent)
a ROC graph
- For each possible cutoff value (score), plot false positive rate (x-axis) as a function of true positive rate (y-axis)
- Connect the dots to get a curve
- We can then measure the area under that curve to get an overall measure of predictive accuracy
ROC interpretation
- AUC ranges from 0.50 (chance accuracy) to 1.00 (perfect accuracy)
- The probability that a randomly selected recidivist will have a higher risk score than a randomly selected non-recidivist
strength of the ROC
It’s the only procedure that allows researchers to summarize accuracy in a way that is not biased by decision thresholds (i.e. scores on an assessment tool)
levels of predictive accuracy of the ROC
- Clinical judgments: AUC =0.55
- Actuarial tools: AUC = 0.68-0.80
- Structured professional judgment: AUC = 0.62-0.75
thresholds and screening devices
The strength of whatever tool you choose to use varies as a function of the chosen threshold
