Abstract
Based on a systematic review of some 23 (out of 300 originally located) studies, it is concluded that most studies show lower rates of re-offending following a non-custodial compared to a custodial sanction. However, this outcome may be biased because, in most quasi-experiments of this kind, subjects with the worst prospects of rehabilitation are likely to be sent to prison. In a meta-analysis limited to five randomised controlled trials and one natural experiment, it is concluded that custodial and noncustodial sanctions do not differ significantly in terms of re-offending. Lessons from this review include, in view of future evaluations, the need to increase randomised controlled trials, to use broader measures of re-offending and rehabilitation, to look at long-term effects, and to deal with possible Hawthorn effects. Equal rates of re-offending do not mean that "nothing works" or "nothing matters", but that criminal justice policies should not be based on the belief that short-term confinement will be damaging.
