Thursday, June 28, 2007

"Assessing the risk of a criminal reoffending gives poor results"

Read "Crime Prediction: The Jailers Dilemma" in The Economist, June 23-29, 2007.

The article states:
Risk assessments are routinely used to help decide who should be locked up, who should undergo therapy and who should go free. In America such tests are used when sentencing sex offenders and killers who may receive a death sentence. In Britain the tests help psychiatrists and psychologists determine whether someone should be held under the new laws that allow someone suspected of being dangerous to be detained indefinitely. Risk prediction is also set to be used to assess the threat posed by people ranging from terrorist suspects to potential delinquents.

Stephen Hart, of Simon Fraser University in British Columbia, Canada, and colleagues decided to determine how accurate these tests are when applied to individuals rather than groups. Typically the tests work by assigning a score to people depending on factors such as their age, the history of their relationships, their criminal past and the type of victims they have chosen. If someone's score places him in a group in which a known proportion has gone on to commit a crime on release from detention, then the risk that person will prove a recidivist is thought to be similar to the risk for the group as a whole.
Dr Hart and his colleagues focused on two popular tests that follow this logic.
They found that variations between members of the groups were very large. In one of the tests, for example, the standard estimate of the chances of members of the group sexually reoffending was put at 36% within 15 years. They calculated that the actual range was between 30% and 43% of the group, with a 95% confidence level. But calculating the average probability for a group is much easier than calculating the same probability for any individual. Thus, using standard methods to move from group inferences to individual ones, they calculated that the chance of any one person reoffending was in the range of 3% to 91%, similarly with a 95% confidence level. Clearly, the seemingly precise initial figure is misleading.
Comment: A great caution to use care in the interpretation of statistics. A study that may show a correlation between social and economic conditions and the propensity to commit crimes may be useful as a rationale for eliminating the unfortunate social and economic conditions, or even for mitigating the blame attached to people from those conditions who commit crimes, but it does not mean that all people from unfortunate social and economic backgrounds are criminals. JAD

No comments: