Job market paper

Abstract

I study the effects of a cognitive-behavioral intervention on incarcerated offenders in Canadian prisons using novel micro data. To address inmates' self-selection into the program, I exploit the inmates' random assignment to evaluators with varying propensities to recommend the program. I find large and significant reductions in recidivism: within six months upon release, the program reduces recidivism by up to 16 percentage points. Moreover, estimated reductions in incarceration time through enrollment in the program indicate an average of $3,800 saved per participant in incarceration costs. I explore heterogeneity in treatment effects, finding that first-time offenders are especially responsive to therapy. The program's group composition and the timing of participation relative to the release date play a role in both the magnitude and persistence of the treatment effects.

Cognitive-behavioral therapy: effective?

A large body of evidence has found cognitive-behavioral therapy (CBT) to be effective in treating a range of issues, from substance abuse to mental disorders, and even involvement in criminal activities with at-risk youths. Whether CBT is effective with incarcerated adults is still an empirical question. Why?

  • Data are difficult to obtain! Most prison-based programs are managed within facilities, and getting participation data that can be merged with future outcomes is quite challenging.

  • CBT programs work on a voluntary basis, and one must address the selection issue when assessing the effectiveness of such programs.

For my job marker paper, I collected unique data on CBT participation in prisons in Quebec, Canada. At the onset of their sentences, inmates are matched with an evaluator who can recommend the program. I exploit variations in these recommendations to assess the impact of CBT on disciplinary infractions, the likelihood of being granted parole, and recidivism.

What do I find?

I find significant reductions in recidivism in the short-term: within a year, participants see their likelihood of reoffending shrink by half! I further find suggestive evidence that the effects persist beyond this horizon, especially for first-time offenders. First-time offenders, just like drug offenders, are also more likely to engage in therapy if they are recommended to do so. In fact, 32% of inmates comply with the recommendations!

Using machine learning

The literature on machine learning has exploded recently, and I use the new advances to predict individual treatment effects. Going beyond an average treatment effect is useful: for each individual, one can get an idea of how effective CBT is for a given set of characteristics. In my sample, I estimate that most inmates would benefit from CBT: at least, there are no adverse effects from participation.

Timing of participation matters

Imagine if a participant finishes the program while having one more year to serve behind bars... Will CBT have any effect at the time of release? If CBT is administered during the early stages of incarceration, the newly acquired skills could depreciate over time. This is what I find: inmates who happen to receive therapy closer to their release date experience larger and more persistent treatment effects.

Interested?

I am always happy to talk about my research (or yours!). Feel free to get in touch.