Intersource agreement on the prediction of recidivism

  • Richard John Parker

Student thesis: Master's Thesis


In a wide range of counselling situations, including those involving offenders, researchers have bemoaned the lack of consensus about outcome. Some researchers have argued that a lack of consensus is due to the fact that the different sources retain unique points of view, which can never be amalgamated into a common outcome. The current paper argues that, while sources will have their own unique perspective, it is possible to develop a meaningful consensus, if it is done very carefully. The factors which need to be taken into consideration are: measuring different outcomes; the different interpretation of the question by each source; and bias. An important outcome in correctional settings is an offender's current level of criminality, or likelihood of reoffending. It was hypothesised that the 12 Probation and Parole Officers' predictions about the likelihood of recidivism of 368 offenders would correlate with fresh charges recorded within 12 months by the Magistrates Court. This hypothesis was supported. It was also hypothesised that the Level of Service Inventory - Revised (LSI-R) would also correlate with fresh charges and that the former correlation would be different to the latter. While the LSI-R did correlate significantly with fresh charges, the predictions of the Probation and Parole Officers were not significantly different. The use of correlation coefficients to assess predictive validity has been criticised as they are affected by base rates of offending and selection ratios of offenders to high and low risk categories. However when a more appropriate statistical tool, the area under the Receiver Operating Characteristic curve (ROC) was employed, the results were not substantially different. The predictive abilities of offenders was also tested against the same criterion and Intersource Agreement on the Prediction Recidivism were found sorely wanting, failing to correlate with fresh charges or any of the other predictive measures. It is argued that the offenders failed to predict accurately because of bias and/or poor ability to predict their own behaviour. The ability of the Probation and Parole Officers to predict reoffending to a much greater level than usually recorded in the criminological literature was ascribed to the higher level of information about their own predictive abilities, through a natural feedback mechanism which does not usually apply to professionals making clinical predictions about recidivism, and to training in predictors of reoffending which they received when they were trained in the administration of the LSI-R. It was also found that the ability of the LSI-R to predict recidivism was improved by incorporating offender age through linear regression. Suggestions were made for improving consensus among sources in counselling outcome studies. In particular, it was noted that sources do not necessarily interpret the question in the manner the researcher desires and it may be necessary to test the subjects' ability to distinguish between similar questions when this distinction is important.
Date of Award2002
Original languageEnglish
Awarding Institution
  • University of Canberra
SupervisorSandi Plummer (Supervisor)

Cite this