The findings indicate that applying weights to reflect the relative strength of risk factors for violence does not improve predictions, but rather results in statistical shrinkage effects. The more sophisticated the weighting algorithm, the greater the shrinkage effect. The authors advise, however, that even though the weighting techniques applied in this study showed poor results, this may not be the case should predictor variables be improved, exchanged, or complemented with other variables in future research. Another finding was that the area-under-the-curve (AUC) estimate varied significantly between the various combinations of randomly drawn subsets of the population. When reviewing the performance of the various weighting paradigms, aspects other than the predictive validity should also be considered. One such aspect is the transparency of the model, i.e., whether it is possible to go back and reconstruct what went wrong in cases where the model failed. The 404 study participants were diagnosed with either personality disorder or schizophrenia in Sweden from 1988 to 1993. The cohort was followed for an average of 8 years. The 10 risk factors used in the assessment were previous violence, young age at first violent incident, relationship instability, employment problems, substance use problems, major mental illness, psychopathy, early maladjustment, personality disorder, and prior supervision failure. 2 tables, 1 figure, 6 notes, and 64 references