Download presentation
Presentation is loading. Please wait.
1
Group Risk Assessment Model Monitoring trends in re-offending among convicted offenders in adult and children’s court Fourth National Justice Modelling Workshop 10-11th July 2008 Dr Nadine Smith and Craig Jones
2
Background Justice agencies aim to reduce re-offending NSW target is to reduce re-offending by 10% by 2016 Re-offending is usually measured by reconviction in court
3
The problem Rates of reconviction are affected by policy to reduce re-offending characteristics of offenders in court Must account for offender characteristics when measuring changes in reconviction
4
The solution Build a statistical model that predicts re- offending based on offender characteristics Use the model to ‘adjust’ for changes in offender characteristics when assessing effectiveness of policy to reduce re-offending
5
The samples Convicted offenders given a non- custodial sanction in 2003/04 in a NSW ADULT COURT Local, District and Supreme CHILDREN’S COURT Children’s Court and Youth Justice Conferences
6
Data source The Re-Offending Database (ROD) from the NSW Bureau of Crime Statistics and Research (BOCSAR) ROD court data routinely collected linked over time demographics criminal history
7
The outcome RECONVCITION Counted from date of first conviction in 2003/04 (index offence) Reconviction occurring within 24 months of the index offence, and Proven in court within 27 months of the index offence
8
2003/04 reconviction rates Per cent reconvicted N 3265,562 ADULT COURT 563,706 CHILDREN’S COURT
9
Possible predictors of reconviction Indigenous status DEMOGRAPHICS Age Sex INDEX OFFENCE Jurisdiction Principal offence type Number of concurrent offences OFFENDER HISTORY Number of prior convictions in the past 8 years
10
The modelling process Logistic regression Identify which possible predictors yield the best model Examine the accuracy of predicting reconviction
11
Actual predictors of reconviction Area under the curve indicated acceptable fit for the adult court (c=0.728) and children’s court (c=0.768) models
12
ADULT COURT Adjusted odd ratios of reconviction 012345 4 or more vs no priors 10-21 yrs vs 40 or olderIndigenous vs non-Indigenous 2 or more vs no concurrentProperty vs drivingViolent vs drivingFemale vs maleUnknown vs non-Indigenous Odds ratios (95% confidence interval)
13
CHILDREN’S COURT Adjusted odd ratios of reconviction 012345 4 or more vs no priors Indigenous vs non-Indigenous10-14 yrs vs 18 or older 2 or more vs no concurrent Female vs maleUnknown vs non-Indigenous Odds ratios (95% confidence interval)
14
Change in reconviction rate ADJUSTED 2003/04 vs 2004/05 OBSERVED Calculate 2004/05 observed rate from data PREDICTED Apply 2003/04 coefficients to 2004/05 data to obtain predicted probability for each offender Calculate the mean predicted probability of reconviction for 2004/05 DIFFERENCE Predicted - Observed Using confidence intervals to assess differences
15
ADULT COURT Change in reconviction rate
16
CHILDREN’S COURT Change in reconviction rate
17
Progress towards target Percentage progress towards target = (Predicted-Observed)/Observed*100 ADULT COURT 1.3%* increase in reconviction rate CHILDREN’S COURT 2.3%* decrease in reconviction rate *No statistically significant progress towards target was identified
18
Caveats of the models Do not give perfect predictions, particularly for small sub-groups Should not be only source of information when making decisions about offenders Can identify individuals for consideration on programs to reduce re-offending
19
Summary Developed methodology to account for offender characteristics when measuring change in reconviction Fairly accurate models of reconviction can be built with routinely collected data Highlighted practical uses of these or similar models One limitation is that macro changes, such as the economy, may also impact reconviction rates and are not considered here
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.