Despite the policy relevance of college-in-prison, the existing research on these programs has important flaws, failing to address selection and self-selection bias. We address an important policy question: what are the effects of college-in-prison program? To do this, we provide the largest study published to-date of a single college-in-prison program. We analyze the effects of the Bard Prison Initiative (BPI) in New York, a liberal arts program that has offered college courses to incarcerated students since 2001. By leveraging the BPI admissions process, we employ a design-based approach to infer the causal effect of participation in BPI. We find a large and significant reduction in recidivism rates. This reduction is consistent across racial groupings. Moreover, people with higher levels of participation recidivate at even lower rates. In light of these findings, we provide policy recommendations that support college-in-prison programs.