Impact Evaluation of the prisonbased core Sex Offender

Impact Evaluation of the prison-based core Sex Offender Treatment Programme (SOTP): Aidan Mews, Laura Di Bella, Mark Purver

Background The prison-based core Sex Offender Treatment Programme (SOTP): • SOTP was a cognitive-behavioural psychological intervention intended to reduce sexual re-offending amongst sex offenders • Delivered by NOMS (HMPPS) to imprisoned sex offenders • Accredited by CSAAP since 1992 • Available in approximately one-quarter of male prisons in England Wales • Available for those sentenced to over 12 months - with a current (index) sex offence or a history of sexual offending; min 18 years old The study aimed to evaluate the ‘core’ prison-based SOTP only, and not other course variants

Previous research Evaluation of the pre-2000 SOTP (Friendship et al. 2003): • No impact on sexual reoffending • Reduced reconviction rates for treated offenders for combined sexual & violent reoffending The international research on the effectiveness of treatment for sex offenders provides a mixed picture. • Some quantitative reviews of available research have found that treatment reduces reoffending in sex offenders (Hanson et al. , 2002; Hanson et al. , 2009; Lösel and Schmucker, 2005; Schmucker and Lösel, 2008; Schmucker and Lösel, 2015) • A review of randomised control trials only found no beneficial effect of psychological interventions (Dennis et al. , 2012).

Common challenges in the field Widespread agreement that there is a need for high quality evaluations of treatment programmes for sex offenders ‘Reviewers are forced to consider which of the less than ideal studies are “good enough. ” Often they disagree’ (Hanson et al. , 2009, p. 866). Main challenges in the field: • • Low reoffending rates for sex offenders Heterogeneity of sex offenders Small samples/short follow-up periods Poorly matched counterfactuals (only a handful of RCTs)

Creating the database Reoffending Cohort Database Master Treatment Database 11, 582 identified prisoners receiving any treatment (1996 -2013) Derived from several NOMS datasets Police National Computer (PNC) Comprehensive records on individual’s criminal/court history Yearly prison release data (2002 -2012) ANALYSIS DATASET 2, 5621 Treatment 13, 219 Comparison Offence types Re-categorisation of 320 sex offences into 30 categories of victim characteristics & seriousness OASys Offender Assessment System database (20042012) Criminal/ personal history, attitudes & motivation 1 Treated people who met inclusion criteria

Study design: Selection of candidates 2000 2012 2002 2015 Treatment Comparison Treatment Period Release Period Re-offending Period Index offence (sex offence excluding breach/soliciting Index Sentence Release Sexual Offence (SO) Prison Sentence Treatment (SOTP) Non-Sexual Offence (NSO) • Index sex offence • Released during the release period • Over 18 yr old at release First release that fit these criteria • Sentence 12 months + Comparison Group: Didn’t received core SOTP or other sex offender treatment during index prison sentence Treatment Group: As above + treated on core SOTP within the prison spell 6

Propensity Score Matching • Established technique to evaluate the effect of treatments in the absence of a randomised controlled trial • Compare cohort of treated individuals with a counterfactual/comparison group – matched in terms of their propensity score (likelihood) to undergo the treatment • Stata 10, psmatch (Leuven & Sianesi, 2003) • Chosen PSM model: kernel matching on the logit of the propensity score with a normal distribution, a bandwidth of 0. 03 and a requirement of common support. 7

Matching factors A wide range of matching factors were included: • Either deemed theoretically important, • And/or were empirically related to both selection onto SOTP and one of the main five outcome measures (overall, sexual, adult serious, child contact, and child image binary reoffending) at the 20% significance level. Excellent matching: No statistically significant differences between the matched groups across any of the 87 matching variables (all standardised mean differences were < 5%). 11 offenders in the treatment group could not be matched. • They had high propensity scores. • • They received index life sentences (69 matched treatment group offenders also received life sentences). Two of them reoffended.

Distribution of the logit of the propensity scores across the treatment and comparison groups, before and after matching

Time from offence to reconviction Around 80% of all non-sex offences are convicted within 6 months of taking place Around 20% of sexual re-offences among the SOTP treatment and control groups are convicted within 6 months of taking place (excluding breach and soliciting) Main SOTP analysis does not limit time from offence to conviction

RESULTS: Reoffending rate over follow-up period (average 8. 2 years) Treated rate Compariso n rate Difference T-value Total 39. 4% 38. 9% 0. 5%pts 0. 35 Sexual 10. 0% 8. 0% 2. 0%pts 2. 45* Non-sexual 24. 0% 23. 7% 0. 3%pts 0. 27 Adult serious 2. 7% 2. 4% 0. 3%pts 0. 59 Adult other 2. 1% 1. 4% 0. 7%pts 1. 86 Child contact 2. 2% 2. 1% 0. 1%pts 0. 18 Child image 4. 4% 2. 9% 1. 6%pts 2. 96* Child other 0. 5% 1. 0% -0. 4%pts -1. 86 Breach 21. 9% 22. 4% -0. 5%pts -0. 41 Soliciting 0. 3% 0. 1% 0. 2%pts 1. 36

Sexual reoffending survival curves

Child image reoffending survival curves

Sensitivity analyses • Were the results stable across groups of offenders with different characteristics? • In each case, the matching process was re-run with the specific offenders under analysis

RESULTS: Reoffending rates for sensitivity analyses Analysis Treated rate Comparison Difference T-value rate Overall 10. 0% 8. 0% 2. 0%pts 2. 45* With OASys (OASys variables included) 5. 1% 6. 2% -1. 1%pts -0. 93 With OASys (OASys variables excluded) 5. 3% 6. 2% -0. 9%pts -0. 71 Extended SOTP 17. 5% 10. 0% 7. 5%pts 2. 71*

Sexual reoffending survival curves, with comparison observations censored at the point of Core SOTP treatment in follow-up period

Quantifying bias: sensitivity of the results to unobserved factors • Sensitivity analysis to quantify the magnitude of bias from unmeasured factors that would need to be present to change the key results of the study. • This indicates the amount of bias required for the sexual reoffending impact to become not statistically significant (i. e. the p-value to rise above 0. 05) when applying a two-sided hypothesis test.

Quantifying bias: sensitivity of the results to unobserved factors The level of bias needed, after controlling for factors included in the model, for the sexual reoffending impact to become statistically insignificant is equivalent to: • 48% increase in the odds of treatment and a 48% rise in the odds of reoffending. It may be for example that we are missing one important binary factor from the PSM model, e. g. overall measure of sexual deviancy. If one third of the treatment group are highly deviant, then these odds would require their probability of commencing SOTP to be 20. 3% (compared to 14. 7% for those not highly deviant) and the sexual reoffending rate to be 9. 6% (compared to 6. 7% for those not highly deviant). • A one quarter increase in the odds of treatment and a doubling of the odds of reoffending or vice versa

Survival analysis – going one step further In addition to the survival curves we also did some survival analysis: • Provides a combined view of binary reoffending and time to reoffence. • Makes better use of the data than a binary reoffending measure. • Enables more powerful hypothesis testing. • Addition to sensitivity analysis Methodology • Kaplan-Meier (nonparametric) vrs Cox proportional hazards model (parametric but allows controlling for confounders) • Log-rank test (equally sensitive to events over time) vrs Peto test (more sensitive to earlier events). Overall the findings reaffirmed the conclusions of the binary reoffending outcomes, with a statistically significant increase in sexual reoffending of around 3%pts after 13 years and 11 months.

Limitations • PSM is not as robust as a prospectively-matched evaluation or a RCT; can only match on observed factors. • Core SOTP is intended for individuals who are not in denial of their offending and are willing to engage in treatment; comparison group included individuals regardless of whether they were willing to engage in treatment or in denial of their offending. • It is possible that a large number of comparison individuals did other cognitive behavioural programme (e. g. Enhanced Thinking Skills). • Dataset was derived from administrative IT systems which, as with any large scale recording system, are subject to possible errors with data entry and processing. • Some comparison individuals attended Core SOTP in the follow-up period (<1%). • Sex offenders are a heterogeneous group, so there are challenges in drawing conclusions on the effectiveness of treatment for the sex offender population. However, matching offenders on the type of sexual index offence and previous sexual offences should minimise this. • Only proven reoffending was measured. • Potential for geographical bias. • On occasions the OASys assessment may have been completed following the onset of Core SOTP, and so have been influenced by the treatment. This could reduce the estimated impact of the treatment.

Conclusions • Results suggest that while Core SOTP is generally associated with little or no changes in reoffending, there were some statistically significant differences. • In particular, there were small increases in the sexual and child image reoffending rates. • The effect was stable across most of the sensitivity analyses performed. • The small changes in the sexual reoffending rate suggest that either Core SOTP does not reduce sexual reoffending as it intends to do, or that the true impact of the Programme was not detected. • This study does not reveal the extent to which Core SOTP reoffending outcomes are due to treatment design or poor implementation.

Media responses

Further research • Further research into the effectiveness of SOTP using matching techniques should focus on measuring potentially relevant confounding factors. • General data issues would also be addressed, including ensuring that sex offender treatment names and dates are accurately recorded in the HMPPS treatment database. • More research could be undertaken to explore in further detail the characteristics of certain subsamples of the treatment group. • In addition, a process study could be undertaken to understand any systematic bias between being selected onto SOTP and not being selected. • To undertake as robust as possible an evaluation of SOTP, a randomised control trial should be considered.

Extra slides 24

Index Sentence Release Date Non-Sexual Offence (NSO) Sexual Offence (SO) Illustrative examples of candidates Comparison 2000 2002 Treatment Period Release Period Re-offending Period >18 <18 Treatment incomplete treatment Excluded 25 Any treatment commencing before 2000 >18 2012 Prison Sentence Treatment (SOTP) 2015

Kernel Matching Retains all in untreated pool within the bandwidth Weights scores closer to the ‘treated’ score more (so will have a bigger weight in effect size calculation) weighted 0. 07 weighted 0. 6 26 weighted 0. 2 weighted 0. 3

Offence categorisation Offence Category Sub-categories Adult Violent/Serious Adult - Violent/Serious (Adult) Images Adult – Other Historical Description Rape, sex assault, etc. Possession and distribution of banned material Offences involving incest among adults/historical Includes offences against vulnerable people and non contact offences (voyeurism) Any child sex offence where direct sexual contact is Child - Contact involved Teachers etc. victim under 18 (excludes those abuse of Abuse of position of trust offences where contact is made explicit) Other Adult Offences Child - Contact Child – Other Facilitating (child abuse) Trafficking, exploitation, prostitution of children Meeting child after grooming online or offline Grooming Child Images Other Child Non-Contact Offences typically relating to sex in presence of child Possession and distribution of child images Child Images Breach Prostitution/ Soliciting Prostitution/Soliciting Non-Sex Violent Non Sex – violent Non-Sex Non-violent Non Sex - non-violent Breach (non-compliance) of conviction/release conditions Prostitution or soliciting related offences such as running or promoting brothels etc. All violent non-sexual offences - robbery, assault etc. All other non-sexual and non-violent offences

Offence categorisation: Victim labels Child Victim can be identified as a child Child victim under 13 Victim can be identified as a child < 13 Male victim is identified Female victim is identified Vulnerable victim Victim is identified as vulnerable (e. g. mentally disordered patients) Family Member Victim is identified as family member

OFFICIAL SENSITIVE – NOT FOR WIDER CIRCULATION Main static and dynamic observed variables PNC variables ·Index offence details: age, type of offence, year. ·Index prison sentence: age at release, length of sentence, type of prison. ·Previous offences/convictions details: number, type ·First offence/conviction details: age, type ·Previous custodial sentences: number, length ·Copas rate: Actuarial risk band ·OGRS 4 (Offender Group Reconviction Scale): predictor of reoffending based on static risk ·RM 2000/S Static actuarial tool designed for adult male sex offenders ·Ethnicity OASys variables ·Whether offender recognises the impact of their offending on their victim/whether offender accepts responsibility for the current offence ·Motivation to address offending/offenders own understanding of motivation to offend ·Whether pornography was a disinhibitor at the time of the current offence ·Whether the victim of the current offence was a stranger ·Current or previous psychiatric problems ·Relationships: Offender’s living arrangements, domestic violence, and relationships ·History of behavioural problems, head injuries, childhood experiences (including abuse), learning difficulties ·Impulsivity/aggressiveness/temper control ·Manipulative, predatory lifestyle score ·Drug/alcohol misuse ·OVP (Offender Violent Predictor) 29