Predicting the Dangerousness of a Criminal
The accuracy of any predictions will be greatly influenced by the behaviour in question, both in terms of its likelihood of reoccurrence and in the factors used to make a determination. Statistical predictions are much more reliable if based solely on the individual's history of violence than are clinical predictions [Ward], even though the courts have shown a preference for clinical predictions [Bartol].
Research on the Canadian parole system has found that reasonable levels of accuracy can be achieved with re-offending in general, based on certain demographic and criminological factors. This set of factors did not work across the board however, providing poorer predictions with those offenders convicted of rape, homicide, assault or indecent assault [Montgomery]. For this population of offenders, one factor proved to be the best indicator: those with a history of three such [and similar] convictions had a repeat offending rate of 17.6%, and those with a history of five such offences had a repeat offending rate of 27.6% [Montgomery]. It is important to point out that while the accuracy will increase in line with the number of offences, even those with five or more offences will only repeat offend on release approximately one quarter of the time. The net result of this is that a prediction that the offender will re-offend could be wrong as often as 75% of the time.
It would seem that the amount of training one has in this issue is unrelated to the accuracy of any judgements made [Quinsey]. Furthermore, the amount of information supplied to the clinician is also unrelated to accuracy, though it does affect the degree of confidence in the predictions they make. The majority of the literature indicates that neither statistical nor clinical predictions have been very accurate, which presents a unique conflict for the professionals involved since these often-flawed predictions are all that is available.
It is often argued that it is better to be overcautious with predictions, stating more often that an individual will be violent when indeed they would not. The penalties for failing to correctly identify a dangerous individual have both social and professional implications. Consequently, most professionals will err on the side of caution and over predict dangerousness. Evidence suggests that even the most sophisticated methods yield a 60 to 70% rate of false positives.