The Violence Risk Appraisal Guide (VRAG) was developed by Harris et al. in 1993.

Actuarial assessment instruments are typically generated using data from a training sample and are then applied to calibration samples. However, even slight differences in the composition of the samples can lead to different statistical models.

If an actuarial instrument is used in another country with a different legal system, culture and lan-guage, this problem is exacerbated. In the past, the predictive validity of the VRAG was confirmed for North America and for many European countries.

Our sample consisted of 107 violent and sex offenders released from a Swiss state penitentiary. Follow-up time was seven years.

The VRAG was predictive for general re-offending - independent of the inclusion of the PCL-R (Psychopathy Checklist-Revised; Hare, 1991) score. However, it failed to predict violent re-offending when the PCL-R score was excluded. Inclusion of the PCL-R score led to a significant - but low - result (AUC=0.62).

The calibration was acceptable for the model estimating any recidivism. In the case of violent recidivism, the calibration was poor.

It can be concluded that the VRAG can be used for risk assessment of offenders in Switzerland. However, the calibration of the risk categories should be adjusted to the base-rate of recidivism.