Environmental heterogeneity in models predicting criminal recidivism: Isolating the contribution of individual-level characteristics
In this dissertation, the links between characteristics of an offender, characteristics of the environment the offender is released to and the likelihood of criminal recidivism are analyzed in detail. The county is used to proxy the environment of release, and differences among the 40 counties included are measured in terms of the social, economic and judicial conditions present therein. To mitigate some of the problems associated with small and ill-conditioned samples, the Generalized Maximum Entropy method is used to assess these links. Additionally, the GME framework easily extends to allow for group-wise heteroskedastic error to account for unbalanced county-level samples. Information parsing is performed to assess the relative importance of knowledge about the 'individual' and the 'environment' in reducing uncertainty about the risk of recidivism. The analysis suggests that in addition to an individual's youthfulness and criminal history, the social, economic and judicial climates in the environment of release do influence the likelihood of recidivism. However, these influences are often indirect and are only seen when a comprehensive analysis of sub-groups is conducted.