Desistance Analysis

Strictly, you cannot from finite data — true termination is only confirmable by following someone for the rest of their life. Analysts therefore use operational proxies, such as a sustained offense-free period or a trajectory that has declined to near zero, and treat them as estimates rather than certainties. Survival models help by explicitly modeling censoring, and process framings reduce reliance on any single cutoff, but the inherent uncertainty about whether a lull is permanent is a defining challenge of the field.

Both framings are useful and increasingly combined. The event framing uses survival analysis to model the hazard and timing of a last offense, handling censoring well and answering 'when do careers end?' The process framing uses trajectory models to capture how the offending rate gradually declines, answering 'how do careers wind down, and for whom faster or slower?' Bushway and colleagues argued that desistance is fundamentally a process, so trajectory methods capture its gradual nature, while hazard methods remain valuable for timing and censored data.

They are closely related and often mirror images. Recidivism survival analysis models the hazard of reoffending after release, while desistance analysis models the hazard of ceasing — the absence of further offending. Both rely on time-to-event methods and must handle censoring, but desistance emphasizes long-run cessation and gradual decline over the life course, whereas recidivism analysis usually focuses on reoffending within a defined follow-up after a specific intervention or release.