The second chapter explores Truth in Sentencing (TIS) laws, which eliminate discretion in the release of inmates and reduce the incentive for rehabilitative effort. I use a regression discontinuity (RD) design to exploit a sharp cutoff in these incentives created by the introduction of TIS in Arizona. I show that judges reduced sentences by 20%, resulting in no change in time served, allowing parole incentives to be isolated as the mechanism of interest. I find that rule infractions increased by 50%, while education program enrolment fell by 20%. Further, I find offenders were 7 percentage points more likely to return to prison after release. To explore heterogeneity, I apply the generalized random forest algorithm to the RD framework. I find that the increase in violent infractions was largest for serious offenders, who saw an increase of 150% to 200% in infractions.The third chapter shows that the introduction of post-release supervision in North Carolina resulted in a 250% increase in returns to prison. This increase is due to 30% of those supervised receiving a release revocation. By 3 years post-release, supervised offenders are returned to prison 36% more often. This indicates that many who were returned to prison due to a revocation, may not have in the absence of supervision. I further show that supervised offenders were 4 percentage points less likely to have obtained a new conviction 2 years after release, and this gap grows to 6.5 percentage points 5 years after release. This indicates that incapacitation due to release revocations can not fully explain the reduction in convictions.In the final chapter, I explore a temporary homeless shelter program. I employ a difference-in-differences strategy to identify the causal effect of these shelters on property crime. I find that in the month a shelter opens there is no increase in reported crime near a shelter, but in later months crime increases by 120% on the block a shelter is located. There is little evidence these impacts extend beyond a shelter’s immediate block.
View record