A comprehensive, three-year analysis of the re-imprisonment patterns of released prisoners has clearly demonstrated that Corrections has mastered the ability to both reliably and accurately, predict the likelihood of any individual offender returning to prison.
Supporting the Department's prediction method is the mathematically-based Risk of Reconviction (multiplied by) Risk of Imprisonment (RoC*RoI) assessment tool developed by analysing the offence histories of more than 133,000 people convicted of imprisonable offences in the 1980s/1990s - the largest investigation of its type in the world.
Developed by Corrections psychologists working for Community Probation and Psychological Services, and Canterbury University's mathematics department the RoC*RoI programme has implications for sentencing, the degree of restriction required for the person's imprisonment, the intensity of the rehabilitation programme required, and how closely they should be monitored and supervised when on parole or supervision.
"Risk impacts on almost every decision made about an offender passing through the justice system. Our ability to accurately predict an offender’s level of risk is critical," says Director of Psychological Services David Riley.
Research on developing a method for measuring risk of offending went back as far as 1928 when American investigator Ernest Burgess developed a 'scientific' risk assessment scale. His study involved more than 3,000 men paroled from an Illinois penitentiary.
After reviewing the criminal records, Burgess noted 21 different 'facts' based on the type of offence committed and the length of sentence given. He then evaluated to see if they were associated with any form of re-offending.
"Not surprisingly, a number of factors were shown to have a statistical relationship with re-offending, and it was by combining these items into a simple checklist that the first risk measure became available," says David.
"Small-scale studies of this type followed, but it wasn't until the 1970s that larger scale investigations focused on the prediction of criminal recidivism began to be published.
"This is probably a reflection of the fact that computers were now being used to store criminal records and that a lot more information was easily retrieved and able to be analysed statistically."
David says most modern jurisdictions use statistical approaches. But the New Zealand RoC*RoI method goes further by providing a precise statistical estimate of the probability that an offender will be both reconvicted and imprisoned within a five-year follow-up period. (Please note that predicted rates of re-imprisonment are now being attained within three years.)
Advances in computer storage capacity and computer power that enabled large amounts of data to be statistically manipulated made it possible for the RoC*RoI model to be developed.
New Zealand was also unique in having a single national justice system, and a long-standing, comprehensive electronic database of criminal offending from the mid-1970s stored on the Wanganui-based government computing facility.
David says initially, the criminal histories of all offenders who committed an offence punishable by imprisonment in 1989 were downloaded and processed in order to produce close to 90 distinct pieces of information called variables. The relationship of these variables to the re-offending and future imprisonment of the entire sample was then statistically examined over a five year follow-up period, with care taken to ensure that only those variables which contributed uniquely to the outcomes were retained for inclusion in the final prediction equation.
The first models were tested on all offenders who committed offences punishable by imprisonment in 1983, 1988 and in 1993 - a move that reassured those involved in the project that the prediction measure was stable over an extended time period.
This prediction measure uses information such as number of previous offences, number and length of prison sentences, time an offender is at liberty between prison sentences, seriousness of offending, and personal information such as age and gender.
The final prediction 'equation' that’s installed on the Department's offender management system automatically generates probabilities in relation to any given offender's likelihood of being re-convicted and imprisoned within a five-year period.
"It's important to note that this model is a statistical and very factual approach to risk assessment which rests upon historical events rather than factors that are subject to change," David says.
"As such, the RoC*RoI model is insensitive to the offender’s personal circumstances and of course, any attempts they may have made to rehabilitate themselves.
"Other factors, such as an offender's anti-social attitudes, beliefs or associates, their employment situation, drug and alcohol addiction and the like are not picked up by this measure, and may need to be considered when interpreting the RoC*RoI risk estimate. However, it should come as no surprise that more deeply entrenched anti-social attitudes, and other criminogenic factors, tend to be associated with higher risk levels."
The measure, and the weightings given to the various variables, are kept under constant review in order that changes in legislation and sentencing patterns can be reflected in the equations run on the Department's management system.
The Corrections Policy, Strategy & Research Group responsible for carrying out the most recent testing of the RoC*RoI assessment process, produced a graph that illustrates the predictive accuracy of the model.
The graph below uses a technique known as "survival analysis" which was developed in medicine to test the effectiveness of various approaches to treatment. In the medical field, "survival" meant literally that - staying alive, and in survival analysis the graph plots over time the percentage of cases which "survive".
In the graph produced by Corrections, "survival" means staying out of prison with the proportion of each risk category (RoC*RoI band) who do so, plotted over a three-year period. The graph shows those who are in the lower risk categories 0.1, 0.2, 0.3, are re-imprisoned over time at a much lower rate, and that at the end of the three year period the majority of these people have not been re-imprisoned.
Conversely, those in the higher risk groups 0.7, 0.8 and 0.9 re-offend at a much higher rate, with far fewer remaining out of prison as time progresses. The other point to note from the graph, is that the results have been aggregated into RoC*RoI bands, each band representing 10 per cent increments in risk of re-offending.
Those in the 0.2 to 0.3 risk band for example, have a 20 to 30 per cent chance of being re-imprisoned whereas those in the 0.8 to 0.9 band have an 80 to 90 per cent chance of being re-imprisoned. Significantly, what the graph also shows is that offenders in each risk band have a differential rate of re-imprisonment - the rate corresponding to the level of risk estimated by the measure. As above, predicted five-year rates of re-imprisonment are currently being attained within three years.
The large number of offence histories used in the development and testing of the RoC*RoI measure makes it accurate for both male and female offenders, that it predicts equally well for all age groups, and that it is accurate for all types of offending. It should be noted that while RoC*RoI is equally accurate for sex offenders as a group, among that sub-group, behavioural factors have been identified which allow more precise identification of those individuals who are highly likely to re-offend.
While there are few areas of consensus among social scientists specialising in criminal justice the two areas of agreement are:
Click here to view a larger file of the graph (GIF: 42KB)
Got a story for Corrections News or want to request the print edition?
Email commdesk@corrections.govt.nz or phone (04) 460 3365.
ISSN 1178-8453