Parole outcomes vary predictably by measurable background factors

Evidence

Burgess notes, for instance, that parole violation among those with a strong work record, high intelligence, and a farming background who had served one year or less was half the state average. On the other hand, parole violation among those living “in the criminal underworld,” whose prosecutors or judges had argued against leniency, and who had served five or more years was twice the state average. — Brian Christian, The Alignment Problem, 244

Categorization

• corroborated: 2 independent passages support this claim — Brian Christian, The Alignment Problem, 244
• XREF: connects to: Prior biased policing leaks into race-neutral variables like conviction history — Brian Christian, The Alignment Problem, 296; Statistical prediction of parole success outperforms subjective judicial decisions; Parole prediction is possible only as aggregate statistics, not certainty — Brian Christian, The Alignment Problem, 245 (+2 more)

Related

• Science only recently began predicting parolee behavior after release (similarity 0.87)
• Parole prediction is possible only as aggregate statistics, not certainty (similarity 0.83)
• Prior biased policing leaks into race-neutral variables like conviction history (similarity 0.80)

Broader Connections

• Parole prediction is possible only as aggregate statistics, not certainty #qualifies (0.75) — Brian Christian, The Alignment Problem, 245
• Statistical prediction of parole success outperforms subjective judicial decisions #expands (0.70)
• ProPublica tested COMPAS recidivism predictions against actual reoffense outcomes for bias #exemplifies (0.65)
• Science only recently began predicting parolee behavior after release #corroborates (0.60)