Dalrymple writes at the Library of Law and Liberty on a criminological study that attempts to develop a method for predicting the rate of recidivism. Unfortunately, the study begins with some faulty philosophical premises…
The authors also tell us that the prediction of reoffending may help authorities to decide on dates of release of prisoners. This can mean only that prisoners deemed at low risk of reoffending are to be released sooner than those deemed at high risk, even if their crimes that were proved beyond reasonable doubt were similar or the same. In other words, prisoners are to be punished (or relatively rewarded) not for what they did do, but for what they might do in the future.
This would not be so arbitrary and contrary to the rule of law if the predictions were 100 per cent accurate, or very nearly so: but of course they are not.
…and then proceeds to make statistical mistakes as well:
Whether the prisoners reoffended or failed to reoffend was counted in binary fashion: yes or no. One, two or a hundred offences counted as one. Swedish criminals may be less productive of crimes than British; but even so, it is likely that the true recidivism rate was more than 100 per cent, as measured by crimes committed and not by convictions. Of this the authors of the paper showed no awareness whatever.