PSYC 3402: Risk Assessment

Contents

1. 1. Why Should We Care?
2. 2. Risk Assessment
   1. 2.1 Goals of Risk Assessment (CAT)
   2. 2.2 Three Major Approaches to Risk Assessment
   3. 2.3 Evaluating Risk Assessment Tools

1. Why Should We Care?

• The public cares.
• The media cares: recent news items.
• Risk assessment informs: sentencing (especially dangerous offender hearings), security classification, treatment needs and intensity, parole decisions, level of supervision in the community or prison, notification decisions, release conditions, etc.

2. Risk Assessment

Nowadays, we asses both risk prediction and management – prediction and management are correlated. Before, we used to stop at prediction (people were either dangerous or not). We have also expanded on our assessment instruments: different types of risks, how risks change in different circumstances etc. Risk potential can change over time.

Risk Prediction: assess the risk that people will commit violence in the future.
Risk Management: develop effective intervention strategies to manage that risk (prevention).

2.1 Goals of Risk Assessment (CAT)

The following is a partial list of the goals of risk assessment in a parole board (in addition to treating offenders as individuals):

1. Improve Consistency: risk assessors should come up with the same decision looking at the same sorts of cases. This is a legally, ethically important variable.
2. Improve Accuracy: want to be confident that the person they’re releasing is a low risk and the person they’re not releasing is a high risk.
3. Improve Transparency: their procedures must be as transparent as possible (to the victim, public, federal government, CJS, etc). Describing why you used the procedure you did is the toughest question to answer; it’s much easier to explain how and what procedures you used.
   • Accountability: clinicians want to feel like they’ve made the right decision about offenders.

In Canada, we use the follow 2 types of risk factors to assess risk:

1. Dynamic Risk Factors: variables that are related to recidivism (dynamic, criminogenic needs). These change with time, less convenient and less reliable, less frequently used but they are sensitive to change (with a level of intervention we can change the level of risk).
   • Acute Factors: change very rapidly, such as current level of substance abuse; your social group; getting divorced; losing your job.
   • Stable Variables: they change slowly over time, such as levels of impulsivity; attitude towards violence; anything related to inherent dispositions, traits and attitudes.
2. Static Risk Factors: these are variables that are fixed and unchanged (they simply exist in the offenders lives), not hard to measure (convenient), frequently used in risk assessment, can be reliably measure and are very predictive.
   • Examples of static variables: three categories – demographic variables, history of criminal behaviour, history of mental disorder. The presence of these doesn’t necessarily increase the risk of re-offence. (Examples: The time at which you first offended, family history of alcohol abuse, gender of the offender.)

Big 4 Risk Factors:

• Criminal History: static
• Antisocial Personality: stable dynamic; impulsivity, aggression
• Antisocial Attitudes: stable dynamic
• Antisocial Associates: dynamic acute and dynamic stable

Not Risk Factors:

• Low SES (socio-economic status)
• Personal distress/psychopathology (low self-esteem and depression)
• Fear of punishment
• Verbal intelligence
• Remorse/empathy
• Offence severity

Unique Risk Factors: below we list specific risk factors related to particular circumstances…

• Domestic Violence: jealousy, number of step-children, hostility towards women.
• Sexual Offending: sexual preoccupations, deviant sexual interests, intimacy deficits, emotional identification with children, stranger victims

2.2 Three Major Approaches to Risk Assessment

1. Unstructured clinical judgment: this is still sometimes used; where clinicians assess the offender with variables they think are important – subjectively selecting, analyzing and interpreting risk factors. Illusionary correlation – where the clinician interprets two variables as correlated but they’re not.
   AUC = 0.55
   • Advantages: flexible (variables can be adjusted to the offender, the context, etc) and idiographic (case-by-case – personalized rather than generalized).
   • Disadvantages: inconsistent (different variables might be used and analyzed in different ways) and has low accuracy (cognitively, we can’t process and identify risk predictors in a sensible way, beliefs, schemas about offenders).
2. Actuarial (statistical) tools: where we collect pre-specified risk factors and enter them into a statistical model that combines and weights them (the opposite of unstructured clinical judgment).
   AUC = 0.68 to 0.80
   • Advantages: consistent (everyone is assessed exactly the same way) and highly accurate.
   • Disadvantages: nomothetic (generalized – not taking into account the individual… especially does not include uncommon, rare events) and validity across different samples (sometimes people don’t do cross-evaluations so they don’t know if the tool is valid and sometimes the tool isn’t valid at all).
3. Structured professional judgment: where we collect pre-specified risk factors while adding in any case specific details (a hybrid of the previous two approaches); the final assessment is a clinical judgment that’s informed by empirical risk factors.
   AUC = 0.62 to 0.75
   • Advantages: flexible (allows individual factors to be important) and nomothetic-idiographic.
   • Disadvantages: moderate accuracy (clinical judgment) and less consistent than actuarial (because it’s idiographic).

2.3 Evaluating Risk Assessment Tools

Short answer: yes, we’re pretty good at it. We are primarily interested in accuracy (do they re-offend and do higher risk offenders re-offend more than low-risk offenders) but it has to be easy-to-use, consistent and gives us a good, fuzzy feeling.

ROC (Receiver Operating Characteristic) Analysis: gives you a particular accuracy measure (AUC – area under the curve) – the probability that an outcome (re-offence) will occur.

• Where we measure the accuracy of risk assessment across decision thresholds (where thresholds are determined by the tool you’re using to assess risk); there are four possible outcomes when predicting risk:
1. True Positive (Hits): correct prediction of re-offending.
2. True Negative (Correct Rejections): correct prediction of not re-offending.
3. False Positive (False Alarms): offender was predicted to re-offend but did not.
4. False Negative (Misses): offender was predicted not to re-offend but did.
• Advantage: ROC is the only procedure that allows researchers to establish accuracy scores that are not biased by decision thresholds (i.e., scores on an assessment tool)  or impacted by base rate (like the correlation)
• Base Rates: the percentage of people within a given population who engage in a specific behaviour or have a mental disorder; it tends to be easier to predict frequent events than non-frequent events.
• Plotting the ROC Graph: plot the hits (y) and false alarm rates (x); the area under curve (AUC) gives you the overall measure of accuracy (more area = more accurate) which is not threshold specific.

Methodological Issues in Predicting Risk: three main issues have been identified…

• Limited number of risk factors being studied
• (How the variable is measured) Official records underestimate the number of crimes actually committed (undiscovered true positives).
• (How the variable is defined) In addition to whether or not they committed a violent act, we need to include the type, severity and motivation for violence.