500 likes | 1.02k Views
Indicators of violent reoffending: the new OASys Violence Predictor. Philip Howard, Senior Research Officer Jackie Seaton, Senior Probation Office Offender Assessment and Management Unit NOMS, Ministry of Justice for England and Wales. Overview of topics covered in this presentation.
E N D
Indicators of violent reoffending: the new OASys Violence Predictor Philip Howard, Senior Research Officer Jackie Seaton, Senior Probation Office Offender Assessment and Management Unit NOMS, Ministry of Justice for England and Wales
Overview of topics covered in this presentation • About OASys • What is it? • When/how is it used? • Why do we need a new predictor of violent reoffending? • Details of the OASys Violence Predictor (OVP) • How was it produced (and what does it predict)? • What items are included? • Does it work?
Overview of topics covered in this presentation • How does OVP affect other NOMS business processes? • Risk of Serious Harm assessment • Sentence Planning • Tiering in the Offender Management framework • Parole • User consultation, piloting and implementation
What is OASys? (1) • OASys is a structured clinical tool used to assess and manage over 250,000 offenders each year in England and Wales • Piloted 1999-2001, rolled out 2001-2005 • Used for offenders aged 18+ on custodial and community sentences • Three main elements • Offending-related factors (“OASys Two”) • Risk of Serious Harm • Sentence Plan
What is OASys? (2): offending-related factors • Sections 1-12 • 1: offending information [criminal history] • 2: [current] offence analysis • 3: accommodation • 4: education, training and employability • 5: financial management and income • 6: relationships • 7: lifestyle and associates • 8: drug misuse • 9: alcohol misuse
What is OASys? (3): offending-related factors • Sections 1-12 (continued) • 10: emotional well-being • 11: thinking and behaviour • 12: attitudes • Items in each section are scored 0/2 or 0/1/2, where 2 = significant problems • Section scores are presented on a Summary Sheet, and thresholds are used to create binary “criminogenic need” indicators (e.g. there are 5 questions on Alcohol Misuse, so the maximum score is 10; a score of 4+ is a criminogenic need) • A weighted total score indicates the likelihood of any reconviction, but imprecisely (the scale is 0-168) • OASys also computes OGRS, an actuarial predictor based on age, sex and previous convictions (Section 1 data). OGRS 3 has just been implemented.
When/how is OASys used? (1) • OASys is used • to inform Pre-Sentence Reports, but only for Standard Delivery (full length) reports • at the start of Community Orders at Offender Management Tier 2, 3 or 4 and release on licence from custody, then reviewed every 16 weeks and at the end of order/licence • at the start of custodial sentences of at least 12 months, then reviewed every year • to inform Parole Board hearings • other, less frequent purposes (e.g. Hostels, bail)
When/how is OASys used? (2) • Case ID and criminal history info is completed by support staff • The main body of the assessment is usually completed by probation officers. There is a countersigning process. • Training is received in the standard probation or prison officer training package • Completion takes a few hours, and workload pressures mean that there is little scope for a net increase in the length of the process
Why do we need a new predictor of violent reoffending? • The current OASys score is a mediocre predictor of any reoffending, and has no predictor of future violence. • Since OASys was introduced, priorities have shifted towards preventing serious reoffending • Risk of harm ratings vary greatly between probation areas, and past efforts to achieve greater uniformity have failed. An objective tool to underpin ratings should help. The next slide shows the extent of variation in 2004/05 and mid-2007.
How was OVP produced? • Offences were split into two groups – “violent-type” and non-violent – on the basis of patterns of criminogenic need and reoffending. After changes due to user consultation, violent-type offences now include • homicide (non-motoring) and assaults • threats and harassment • possession of offensive weapons • public order offences • criminal damage • robbery
How was OVP produced? (2) • OASys assessments at the start of community sentences or discharge from custody were filtered for data completeness, then matched with the Police National Computer (PNC). • Criminal history details from the PNC, and item and section scores from sections 2-12 of OASys, were used as predictors of violent-type reoffending (within 2 years) in a logistic regression model • The original model • gave 71% of the maximum possible score to static factors and 29% to dynamic factors • was not readily understandable by OASys users, as probation officers tend not to be trained in logistic regression
How was OVP produced? (3) • The user-friendly version of OVP • Gives 40% rather than 29% weighting to dynamic factors • This does not damage predictive validity very much • Future research will examine whether changes in OASys scores upon review improve predictive validity. If so, then giving high weights to dynamic factors should enhance this ability to reflect change. • Allocates each item a weight out of 100. Offenders get an overall score out of 100, which is translated into predictions of proven reoffending within 1 and 2 years of order/discharge. Longer followups are possible. • We’ve also produced OGP, the OASys General Reoffending Predictor, to cover all other offences
A quick note on measuring predictive validity • Measure predictiveness using the “Receiver – Operator Characteristic” (ROC) method • This generates a statistic called Area Under Curve (AUC), which ranges from 0.5 to 1. A score of 0.8 is extremely good in the context of a reconviction study • The AUC score can be thought of as follows • Put all the reconvicted offenders in one list, and all the non-reconvicted offenders in the other list • Pick one offender from each list and compare their scores on your predictor • The AUC is the probability that the reconvicted offender has the higher of the two scores (so a high AUC is good) • With large samples (which we had), split it into construction and validation halves. Build the model on ‘construction’ cases, report its AUC on ‘validation’ cases
Applying OGP and OVP: a case study • Joseph Keaton, aged 32, is convicted of stealing a train. He has a short criminal record, starting age 22: one common assault, one public order offence and one motoring offence. • He has no accommodation problems, scores 1 point on employment, has no drugs problems but a very serious alcohol problem • He has some problems with temper control, but overall his thinking and behaviour is quite good. His attitudes are not too good, and he does not recognise the impact of his offending. He spends a lot of time performing potentially criminal activities.
Keaton’s likelihood of violent-type reoffending (weighted OVP score 48/100, exact=15%/28%)
Keaton’s likelihood of non-violent reoffending (weighted OGP score 35/100, exact=16%/27%)
Using survival analysis to estimate likelihood over time • Survival analysis techniques allow us to estimate the likelihood of proven reoffending at any time period • month-by-month rates are easily displayed • ‘censoring’ allows offenders who receive pseudoreimprisoments or imprisonments for other offences to be included until that imprisonment occurs, rather than either wholly rejecting them (which could bias the sample) or pretending they’re always at risk of reoffending (which certainly biases the results) • The ‘hazard’ shows the likelihood in a specific time period, while the survival function is cumulative
Using the new predictor in practice – integrating with other business processes
Risk of Serious Harm assessment • In OASys Risk of Serious Harm assessment, a Screening is completed for all offenders. A Full Analysis is completed where the Screening indicates it is necessary. • Risk ratings are • Low (automatic if no Full Analysis) • Medium • High • Very High • Ratings are produced for risk in community to children, known adults, public and staff: we usually quote the highest of these four for each offender • A lot of clinical judgement is involved, and as shown earlier there is a great deal of variation between probation areas
Risk of Serious Harm Screening and Full Analysis by OVP and sexual offending history – possible guidance
Tiering • OGP will replace the old OASys score as an indicator of the risk of general reoffending and level of criminogenic need • An OGP score of 75 – 100 will be indicative of high likelihood of non-violent reoffending (90%+ within 2 years) and push offenders towards Tier 4 • OVP will not have a direct impact, but will affect Tiering through its impact on RoSH • Detailed guidance will be developed in co-operation with the Offender Management team
Targeting to offending behaviour programmes • ART/ CALM etc. should generally only be considered for offenders with ‘above average’, ‘high’ or ‘very high’ OVP scores • GOBPs and standard-intensity substance misuse OBPs should only be considered for offenders with ‘Medium’ or ‘High’ OGP scores • Lower intensity alcohol misuse and DID OBPs should not be considered for offenders with ‘high’ OGP scores • Sex offender and domestic violence programmes are not affected by OGP/OVP as targeting should be based upon RM2000 & SARA
Use of OGP/OVP in PSRs • Similar to use of OASys score now • Usually only available when SDR prepared • Courts could be informed of scores via description within categories: e.g. OVP score 36/100 plus OGP score 64/100 = “ low to average likelihood of violent reoffending and a high likelihood of non-violent reoffending”
Release Decisions • OVP scores, as part of risk of serious harm considerations, are likely to be important in parole and other supervised licence decisions • Detailed tables will be supplied to the Parole Board translating scores out of 100 into likelihoods of violent-type reoffending across a range of followup periods (as in the survival analysis graphs earlier)
User Consultation Initial design workshop Nov 2006 Ongoing consultation with IT developers throughout 2007 Introducing the predictors to key stakeholders Establishment of National Reference Group, Dec 2007
PILOTING THE PREDICTORS • Pre-pilot: Nottinghamshire, June 07 • ‘Live’ pilot involving 3 probation areas, for 1 month each, Feb-April 08 • Focus group feedback • Report & sign off
IMPLEMENTATION OF OGRS 3/OGP/OVP • OGRS 3 already implemented in e-OASys April 08 • OGP/OVP planned to go into OASys version 4.3 if it can be managed • Otherwise will be introduced with OASys-R