But the decision between probation or a custodial sentence is based partly on the attitude, risk profile and situation of the offender. Surely that's going to skew the statistics?
It's hard to have a meaningful discussion of these issues in the abstract, but the short answer is:
1) it depends
2) bias correction is not a trivial matter
Bias correction is useful and often legitimate, but its use should always be interpreted as weakening the epistemological rigor of the analysis. In lay terms: bias correction should be consumed with a grain of salt.
General entry point into the subject matter: https://stats.stackexchange.com/q/31253/17054