This article explores the main determinants of the hazard of recidivism among ex-prisoners. We use a nationally-representative sample of prisoners released in 1996-1997 in France, drawn from a 5-year follow-up survey run by the French correctional administration. We estimate semiparametric duration models which deal with violations of the proportional hazards hypothesis. Our results confirm the importance of gender, age, nationality, access to employment and prior convictions on recidivism within five years after release from prison. We also find significant differences in hazards of recidivism by type of initial offense, penal status at entry, and type of release (early release under parole, etc.), while controlling for prison fixed effects. Finally, our study casts doubt on the influence of certain variables (marital status at entry, education, homelessness) and on the effectiveness of semi-liberté as a way to prevent recidivism.
