This report summarizes the factors associated with recidivism, prior research on the role of these factors in predicting recidivism, and the research design the author used to develop recidivism predictions as part of the National Institute of Justice’s (NIJ’s) Recidivism Forecasting Challenge.
The author submitted an entry with predictions for only the third period of the Challenge; the entry placed fourth in the Male Parolees category for that period. The author entered the Challenge in his capacity as the Executive Director of Duddon Evidence Top Policy Research, a law and policy research and consulting business he operates as a sole proprietorship. In developing his models and subsequent predictions, he consulted with an experienced quantitative psychologist and psychometrician, as well as a Georgia attorney who had previously served as a public defender for DeKalb County, Georgia. Time and financial constraints limited his entry to predicting recidivism for the third Challenge period only. The descriptions of the variables and models he used are discussed. Among the variables discussed are dispositions in the criminal justice system, age when incarcerated, and location variables after release. He explains why the “lasso” was used as a model as an estimator of coefficients in a regression model that includes a penalty term to address over-fitting. The lasso is particularly useful in selecting relatively few out of many possible variables that affect an outcome.
- Tooth morphology and population affinity: Testing rASUDAS2 on modern African and European-derived samples
- Disparities in Segregation for Prison Control: Comparing Long Term Solitary Confinement to Short Term Disciplinary Restrictive Housing
- MPrESS: An R-Package for Accurately Predicting Power for Comparisons of 16S rRNA Microbiome Taxa Distributions including Simulation by Dirichlet Mixture Modeling