A Competing Risk Analysis of Recidivism

In this paper we build and estimate an econometric model of recidivism, using a US data set of released prisoners from state prisons in 1983. The data set is unique in the sense that it contains information on recidivism for both federal and state crimes, in contrast of past studies restricted only to state samples with few or no information about federal crimes. The paper extends the literature on the application of econometrics to criminal behavior by treating the process of recidivism as a competing risks model where dependence arises from unobserved heterogeneity. Two models are estimated, an accellerated and a proportial hazard model, respectively. Both models incorporate the non-monotonic nature of the hazard of recidivism. The proportional hazard model appears to perform slightly better, although both models capture some key features of recidivism behavior reasonably well. In particular, the length of the sentence received for the last committed crime seems to reduce the risk of recidivism in both model, and for violent and non-violent crimes, which is an important conclusion for policy purposes. ¤This paper is based on the ...rst essay of Carvalho’s Ph.D. Dissertation at Pennsylvania State University. The ...nancial support of CAPES Foundation, Brazil, is gratefully acknowledged.