It seems like there's almost certainly a confounding variable here: the kids who are likely to engage in criminality are also the ones most likely to do poorly in school, skip classes, and be held back.
It's more correlation than causal - for the same reason that we couldn't just give every student straight A's and expect them to have similar outcomes as students who would have otherwise earned straight A's.
Working backwards like that is like trying to help someone lose weight by tweaking their scale to always show a healthy BMI.