University of Tasmania
Browse
Gordon_whole_thesis_ex_pub_mat.pdf (3.79 MB)

Prediction of recidivism in a Tasmanian population : evaluation and development of community based risk assessments

Download (3.79 MB)
thesis
posted on 2023-05-27, 11:59 authored by Gordon, HD
In the realm of criminal justice, there are few aims that are more integral to their practices than the assessment and prediction of risk. Many correctional agencies utilise formal risk assessments to provide a structured guide in order to accurately assess an offender's risk of recidivism. In doing so, it is important that the assessments chosen are psychometrically valid and reliable, as well as jurisdictionally appropriate. Therefore, it is important to validate an international risk assessment within the population it is intended to be used, particularly as there are concerns about the applicability of using an international risk assessment within an Australian offender population. This thesis is comprised of three papers. The first two papers evaluate the utility and predictive validity of the Level of Service/Case Management Inventory (LS/CMI) within one Australian offender jurisdiction. This information was then utilised in order to develop and pilot a revised risk assessment (Australian Risk/Need Inventory [ARNI]) tailored for the Tasmanian offender population. In study 1 of this thesis, the need profiles and validity of the LS/CMI was investigated for Tasmanian offenders serving community-based orders. The results of this study indicated that the LS/CMI had a weak discriminative ability (AUC = .664, 95% CI [.611, .717]) for non-Indigenous males (N = 569). However, it predicted recidivism in non-Indigenous female offenders (N = 113) at an accuracy level no greater than chance (AUC = .575, 95% CI [.433, .717]). For Indigenous male (N = 96) and female offenders (N = 29), the LS/CMI was not able to predict reoffending. The results for non-Indigenous females and Indigenous male and female offenders should be interpreted with caution due to the small sample sizes. These findings for non-Indigenous offenders are consistent with previous Australian and international research. Study 2 aimed to investigate the factor structure of the LS/CMI using Australian offenders who were completing community-based orders (N = 302). The results of study 2 indicated that the LS/CMI Total score achieved excellent internal reliability. However, there is some concern regarding the capacity for the subscales to function independently. Factor analysis determined a two-factor solution at a subscale level (criminal conduct and lifestyle considerations), whereas a more diverse factor solution was obtained at an item-level. The LS/CMI was determined to be predictive of recidivism, but this was a weak effect (AUC = .621, 95% CI [.546, .696]). This suggests that the LS/CMI as it is currently used in this population may not be the most appropriate assessment tool, requiring further research before an international risk assessment is adopted in an Australian jurisdiction. The third study presented in this thesis involved amalgamating the information obtained from the previous two studies in order to develop a risk assessment to be piloted within the Tasmanian Department of Justice. This instrument was piloted on offenders who were incarcerated or completing a community-based order (N = 301). The findings from this study indicated that from the original 78-item pool, 45 items added the most information in the development of the Australian Risk/Need Inventory (ARNI). The Cronbach's alpha for the total score indicated an excellent level of internal reliability (˜í¬± = .93). At a subscale level, the internal reliability ranged from excellent (˜í¬± = .92) to acceptable (˜í¬± = .62). In regard to the factor structure, a ten factor solution was identified. The ARNI total score and five of the ten subscales indicated a significant reasonable ability discriminate between offenders who did reoffend and those who did not reoffend within a six-month time frame. The preliminary results of the ARNI indicated that it is able to identify recidivists within a Tasmanian offender population and is internally consistent. It is suggested that extending the sample size (including increasing the heterogeneity of the offender sample) and increasing the follow-up reoffending period may increase the predictive utility and sensitivity of the ARNI total and subscale scores in discriminating between lower- and higher-risk offenders. However, the results of the studies presented indicates that in order to conduct a more in-depth risk assessment, specialised assessments (such as those addressing substance use and instrumental aggression) also need to be conducted alongside the general risk assessment. This will provide the most comprehensive risk assessment process and will allow criminal justice agencies to utilise their limited resources efficiently and effectively.

History

Publication status

  • Unpublished

Rights statement

Copyright 2016 the Author Chapter 2 appears to be the equivalent of a post-print version of an article published as: Gordon, H., Kelty, S.F., Julian, R. (2015). An evaluation of the level of service/case management Inventory in an Australian community corrections environment. Psychiatry, psychology and law, 22(5), 247 - 258. Chapter 3 appears to be the equivalent of a post-print version of an article published as: Gordon, H., Kelty, S.F., Julian, R. (2015). Psychometric evaluation of the level of service/case management inventory among Australian offenders completing community-based sentences. Criminal justice and behaviour, 42(11), 1089 - 1109 The final version of the Australian Risk/Need Inventory, (contained in appendices D & E), remains the joint intellectual property of the University of Tasmania and the Tasmanian Department of Justice, and access may be provided upon application to the author (hdgordon@utas.edu.au).

Repository Status

  • Open

Usage metrics

    Thesis collection

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC