Case 1: Arthur Age 45: Convicted of indecent assault x4 against niece Background Oldest child in family - 2 younger sisters Unhappy childhood: Physically abused, few friends, bullied Poor academically at school; no behavioural problems Intermittent employment in unskilled jobs Married for 15 years, marriage broke up when convicted No substance misuse 3 convictions for theft in teens, no history of violence

Case 1: Arthur (cont.) Niece 13 at the time of the offences, which took place over a year. Describes offences as occurring in the context of a relationship. 1 previous sex offence conviction : 12 year daughter of a family friend, 10 years ago Admits to offending against two other teenage girls in similar circumstances.

Case 2: Kevin Age 27: Convicted of indecent exposure x4 Community Rehabilitation Order Background Only child Unhappy and isolated as a child, few friends, bullied Poor academically at school, truancy Regular periods of unskilled employment A number of short term relationships, cohab x3, longest one year Alcohol abuse No history of violence, no offending apart from indecent exposure

Case 2: Kevin (cont.) indecent exposing since late teens; 7 previous convictions targets schoolgirls never approaches or follows; if confronted runs off past treatment from psychiatrist (psychotherapy) also “counselling” while on probation current offence while on probation order

The Meaning of Risk Likelihood of offending Immediacy of offending Frequency of offending Consequences of offending Escalation of offending

The Meaning of “Level of Risk” What does low medium high very high mean?

The meaning of “likelihood” “a 60% chance of reoffending” a measure of subjective belief given the same circumstances, this man will reoffend 6 out of 10 times (like rolling a dice) 6 out of 10 individuals like this man will reoffend

RMA Air the importance of benchmarking, or relative risk

Prediction is tricky Lawrence ‘Yogi’ Berra especially about the future.

Predictors of sexual recidivism? Hanson & Morton (2004) child phys, emotional, or sexual abuse psychopathy negative relationship with mother or father low self esteem child molester attitudes emotional identification with children sexual preoccupation deviant sexual interests denial, minimisation, lack of motivation for Rx non-compliance with supervision impulsivity lack of victim empathy self regulation problems poor problem solving loneliness social skills deficits any substance abuse conflicts in intimate relationships none or little small to medium moderate to large

Predictors of sexual recidivism Hanson & Morton (2004) Medium to large: On their own

Predictors of sexual recidivism Hanson & Morton (2004) - child physical, emotional, or sexual abuse - negative relationship with mother or father - low self esteem - social skills deficits - loneliness - denial, minimisation, lack of motivation for Rx - lack of victim empathy - any substance abuse - child molester attitudes On their own none or little: ? ? ? ? ?

Static + Stable + Acute = Current Risk genetic vulner. blood pres. cholesterol smoking diet long term risk treatment targets monitoring supervision Risk and myocardial infarction

Static + Stable + Acute = Current Risk historical long term characteristics immediate behaviours long term risk treatment targets monitoring supervision Risk and sexual reoffending actuarial clinical other techniques

Actuarial risk assessment rules objective data unbiased quantifiable: defines low, medium, high deductive

How well they work “Clinical” r = 0.10 just better than chance Sex precons r = x better than “clinical” Clinical + Research r = Scales r = (ROC areas ~ 0.75)

sex violence

Risk Matrix Sex STEP ONE 1. Age 2. Sexual sentencing appearances 3. Criminal sentencing appearances AGGRAVATING FACTORS - male victim- non-contact sex offence - stranger victim- never ‘married’

Risk Matrix Violence age violence sentencing appearances burglary

RM2000 (sex) Per cent reconviction for each risk group 15 year follow up or? 7% 19% 36% 59%

RM 2000 (sex) - results 20% 39% 28% 13% per cent in each category

RM2000 (violence) Per cent reconviction for each risk group 15 year follow up

RM 2000 (violence) - results

Risk Matrix validation England prison releases 1979: year follow-up; n= : 4 year follow-up; n=311 ‘early 1990’s’: 2 year follow-up; n=647 No information on: - missing cases (representative sample) - confidence intervals (distinct categories?) - comparison of risk categories between studies - peer review (one limited report)

Predictive Accuracy Risk Matrix Sex AUC = 0.77 (2 year follow up) AUC = 0.75 (long term follow up) Risk Matrix Violence AUC = 0.78 Long term follow up AUC = 0.80 Long term follow up Moderate predictive accuracy

Risk Matrix Validation small study populations selective samples lack of published data distinct categories? meaning of categories? similarity of performance in different populations? % reconviction? (base rate problems)

Other studies Craig et al, (2007): n=85 Craissati and Beech (2004): n=235, but 9 reconvictions Knight and Thornton (2007): high risk sample,

Risk Matrix Scotland all sex offenders released between minimum follow-up 5 years Risk Matrix Sex 771 of 1029 (75%) subjects scored + follow-up Risk Matrix Violence 974 of 1029 (95%) subjects scored + follow-up

Risk Matrix (S) Scotland results % in each category

Risk Matrix (V) Scotland results % in each category

Risk Matrix (S) Scotland results Reconviction (11% at 5 years) % reconviction (5 year) in each risk category (overlap in confidence intervals between high and vh only) Likelihood ratio Low0.24 Medium0.91 High2.25 V. High3.58

Risk Matrix (S) Scotland results % reconviction (5 year) in each risk category Odds ratios Med - Low 3.7 High - Med 2.9 VH - High 1.6 VH and H apparent by 1 year

Reconvictions: RM (S) Scotland v RM (S) England Scotland 5 year reconviction = 11% England 5 year reconviction = 20% * Odds rations similar LR’s differ for high risk nearly differ for med.

Predictive Accuracy for the S-Scale England & Wales 2 year: AUC = 0.77 England & Wales 15 year: AUC = 0.75 Scotland 5 year: AUC = 0.73 Moderate predictive accuracy

Risk Matrix (S) Seriousness of reconviction Low36% Medium30% High13% V. High13% Low + Medium significantly higher proportion than High + Very High 3+ year custodial sentence:

Risk Matrix (V) Scotland results Reconviction (12% at 5 years) % reconviction (5 year) in each risk category (no overlap in confidence intervals) Likelihood ratio Low0.23 Medium0.81 High2.23 V. High4.43

Risk Matrix (V) Scotland results % reconviction (5 year) in each risk category Odds ratios Med - Low 3.6 High - Med 2.7 VH - High 2.0 H and M apparent by 1 year

Reconvictions: RM (V) Scotland v RM (V) England Scotland 5 year reconviction = 11% England 10 year reconviction = 19% Odds rations differ (high-med) LR’s differ for medium

Predictive Accuracy for the V Scale England & Wales 10 year: AUC = 0.78 England & Wales 15 year: AUC = 0.80 (.76?) Scotland 5 year: AUC = 0.76 Moderate predictive accuracy

Risk Matrix (V) Seriousness of reconviction 15 received sentences of > 1 year 5 Life sentences: 4 in medium risk offenders 1 in a low risk offender

Static + Stable + Acute = Current Risk historical long term characteristics immediate behaviours long term risk treatment targets monitoring supervision Risk and sexual reoffending actuarial clinical other techniques

Risk Matrix and Static 99

Risk Matrix Summary for Scotland Reconviction Risk RM (S)RM (V) Low < 5% 5% Med10% (4x low) 10% (3.5x low) High 20-25% (2.5x med) 25% (3x med) VH 33% (1.5x high) 40% (2x high)

Recommendations training quality assurance routine collection of relevant information monitor base rate of reconviction