Testing the predictive utility of the STATIC-99: A Bayes analysis

Authors

  • Eric Beauregard,

    Corresponding author
    1. School of Criminology, Simon Fraser University, Burnaby, British Columbia, Canada
      Correspondence should be addressed to Dr Eric Beauregard, School of Criminology, Simon Fraser University, Burnaby, British Columbia, Canada V5A 1S6 (e-mail: ebeaureg@sfu.ca).
    Search for more papers by this author
  • Tom Mieczkowski

    1. Department of Criminology, University of South Florida, Tampa, Florida, USA
    Search for more papers by this author

Correspondence should be addressed to Dr Eric Beauregard, School of Criminology, Simon Fraser University, Burnaby, British Columbia, Canada V5A 1S6 (e-mail: ebeaureg@sfu.ca).

Abstract

Purpose. This study applies a Bayes analysis to the probability of a particular STATIC-99 score and its associated re-offence probability with recidivating for a sexual offence. We examine this probability over three time frames: within 5 years; within 10 years; and within 15 years.

Methods. This study was conducted using the same data from Hanson and Thornton (1999). This dataset is constituted from four different samples: Institut Philippe Pinel (Canada) sample; Millbrook Recidivism Study (Canada) sample; Oak Ridge Division of the Penetanguishene Mental Health Center (Canada) sample; and Her Majesty's Prison Service (UK) sample. The final sample for which sufficient information was available to score the STATIC-99 includes 1,086 sexual offenders. Bayes statistic has been used to analyse the data.

Results. Results are consistent with the STATIC-99 as a useful assessment tool. The Bayes-generated probabilities as well as odds ratios show a consistent increase in increased likelihood of re-offence as the score value increases.

Conclusions. The Bayesian analysis of the STATIC-99 shows that this method is very interesting in the context of risk assessment tools. This approach to risk assessment instruments may be more appropriate in the communication of analytic results as it can offer clinicians a combination of probabilities and likelihood ratios resulting a readily accessible profile of risk.

Ancillary