680 words; 3 minutes to read

Part of the response of the criminal justice system to COVID 19 has been to release people from prison since jails and prisons are hotbeds for disease given crowded conditions.  Some provincial governments in Canada have released some people  but the federal government has so far done little to release people – even those who are very old or infirm, or near the end of their sentences.

The debate around releasing prisoners often turns into a discussion of risk.  The argument is that we should not release people who are a serious risk to commit new crimes.  But this raises the problem of how accurately we can predict who might commit a new crime; how accurately we can predict risk.

Research on risk assessment

There is an increasing amount of research on this question, and it has produced some answers that don’t fit at all with what many people believe.  It turns out that we are not all that good at predicting who is likely to commit another crime if released into the community.  Several earlier blogs and two new sources provide useful perspectives on this problem.

One of our earliest posts in the fall of 2017 reviewed four studies on prediction.  All of them (and one was itself a summary of 68 studies) found that various prediction instruments had unacceptably low success rates, and could be wrong 50% of the time.  Even the best instruments are only accurate from 2/3 to ¾ of the time, which is a huge error rate when one is dealing with whether or not people should be imprisoned.  Would any of us be satisfied with a medical diagnosis that was wrong one time in 3 or 4?

More supervision can make things worse

A related post was based on work by Jennifer Doleac  who found that risk does not just relate to the person, either.  Several studies she reviewed found that stricter supervision of parolees actually led to more crimes.  So how we support people after release also affects the risk they present.

And a post last summer based on work by Canadian researcher Karl Hanson showed that the chances of re-offending decline steadily both with age and with time out of jail.  “Rather than trying to change “offenders” into something that they are not, effective rehabilitation can be seen as supporting desistance processes that are naturally occurring and already present to some degree. Whether they are aware of it or not, desistance is their future.” Strict conditions on release may work against these processes.

Expert assessments may also carry bias

Now two new sources strengthen these points.  An article by Jenn Karmorowski and colleagues points out that risk assessment instruments, while providing an impression of objectivity, can actually still carry biases based on ethnicity or type of crime or other factors.  Those doing the assessments still have, like any human, biases that make their way if not into the mathematical results of the assessment, into the recommendations made as a result.  “some experts may be influenced more by the moral dimensions of certain risk factors than by those factors for which there is more substantive scientific support.” Not surprisingly, it turns out that assessments done for prosecutors are often more negative than those conducted for the defense.

A recent piece in Scientific American by Sophie Bushwick makes the same point.  Human predictions about what people will do in the future are  very often wrong, while predictions done by instruments and algorithms, while having the appearance of being scientific, are often not a lot better.  And accuracy varies depending on the circumstances in which the assessment is done, as Kamorowski and team found.

Assessing risk matters less than providing supports

The increasing body of evidence tells us to be very cautious when we are assessing the future risk of people in the criminal justice system.  The odds of making a wrong choice, with devastating consequences (either more crimes or people being unnecessarily imprisoned for long periods of time) remain quite high.  Our best choice is to release most people early and then provide the supports they need to desist from crime, a strategy we do know is effective.




Comments are closed here.