Risk Assessment Flashcards
who are called upon to include statement about offender risk?
forensic psych
- treatment summaries
- section 34 assessments
- probation referrals
- school board q
what is the first gen of risk prediction instruments?
expert opinion
- clinical impression, subjective assessment, professional judgment
- collection of info to assess risk
- INTERVIEW and FILE REVIEW
- nature and interpretation subject to personal discretion
- final estimate is decided by them
pros to first gen
- readily available
- incorporates clinician’s experience
- clinician has freedom to consider any data
- short term validity (48-72 hrs)
cons to first gen
- questionable accountability and fairness
- difficult to explicate decision riles as they are variable
- poor reliability and validity for med and long term predictions
second generation
addition of objectivity
- VRAG, PCL-R
- identify variables empirically associated with risk and assign values, or weight to each
- dynamic/clinical variables but often do not
- actuarial approach
- predict particular outcome -> insurance companies, likelihood of violence expressed as relative probability
pros to second gen
- objective
- scientific
- lend themselves well to appropriate levels of scientific and legal scrutiny
- consistency and accurate
- less dependent on rater’s clinical experience
cons to second gen
- can be insensitive to changes - dynamic variable change than legitimately lowers risk then instruments dont give credit to that
- optimization can restrict generalization
-> supervised learning
**VRAG
violence risk appraisal guide
- assigns patient to one of 9 categories or bins, each bin associated with specific probability of recidivism
- forward stepwise regression
- neg correlation
- area of overlap = correlational which is directly proportional
- the unique contribution is what is left
graph for VRAG
highest p of recidivism is 4
- highest number is in 9 but that is rare
reporting risk estimates
- risk cohort
- type of recid
- time frame
underlying risk model
stable over time - equally likely to commit offense
- individuals can be higher or lower than reference pop
what is third gen
risk-needs assessment
- SPJ
- arose with risk management orientation
- certina needs are linked to criminal behavior and so alter needs, change likelihood of behavior
- criminogenic needs = dynamic risk predictors
what are the 3 categories in third gen?
historical
clinical
risk (RF)
what can intervention do?
affect clinical dynamic factors
what can static factors do?
affect the amount of risk
= above threshold = likely to reoffend
what is not included in second gen tool?
dynamic factors
- perfect sin wave then predicted value - 0 or close to then the highs and low cancel out
risk = ?
static risk + dynamic risk
- threshold exceeded then offending becomes highly likely
- change static = focus on dynamic
what is the best for long term prediction and stratifying offenders by risk level
2 gen
what is better at identifying circumstances under which further offending will take place
SPJ
pros for 3 gen
- formal structure on eval but felxibility around decision-making
- sensitive to changes - dynamic factors
- better tahn unaided clinical judgment
cons for 3rd gen
less evidence on reliability - not good long term
- less validity
- no empirical optimization of item selection or weighting
- accuracy unclear when scores adjusted by clinicians
- more rf = higher risk = arbritariness taken out of account
MASORR
multifactorial assessment of sex offender risk for recidivism
- AUC - measure of predictive accuracy higher numbers mean better predictions
serious vs sexual pre and post treatment
sexual > serious for both
- pre and post for serious not that much different
what are correlation coefficients
r scores
r^2 = percentage of variance accounted for by the correlation
base rate problem
serious problem evaluating instruments on the basis of hit rate - correct classification
base rate
proportion of a reference population that exhibits has/undergone a target attribute or event
- 50% going to reoffend
LOOK AT BASE RATE CHARTS
as BR decreases
false positive increases
BR increases
fal negatives increase
sensitivity =
TP / (TP +FN)
- # of predicted recidivists / total # of recidivists = probability of correctly classifying future recidivist
specificity
TN / (TN + FP)
= # of predicted non-recidivists/total # of non-recidivists
= probability of correctly classifying future non-recidivist
what does a good risk assessment tool include?
larger mean score difference for 2 groups
- produce reasonably small variance to minimize variance
AUC
area under curve of receiver operator characteristics
- insensitive to base rates
- measure of comparison
what does AUC reflect?
proportion of non-recidivists that have a lower test score than a randomly chosen recidivist
does aging offenders pose lower risk over time for physical reasons?
decrease in recidivism in older age
- accuracy of VRAG when controlled for aging and time at risk
what outperforms clinical judgement
actuarial assessments
what has short term utility
clinical judgement
what provides little specific info and insensitive to change
actuarials
what uses dyanmic variable?
3rd gen tool
- modest controbution to predicitve accuracy
what did HCR-20 v3 add?
relevance: RF may be more or less influential based in indidivual
case formulation: movement away from using RA instruemnts as means of estimaitng risk
pitfalls to avoid
- failing to stick to publishers scoring criteria
- inappropriate instrument selection - SAVRY
- vague reporting
**threat vs risk
- risk is ultimately reflecting one’s tendency to offend
- threat includes target availabity, vulnerability, offender risk
*threat =
risk x target avail x vulnerabilities
vulnerabilities
lapses in security measures and/or capacity to response
