Determinants of Mass Incarceration: Using Machine Learning to Assess Differences Across Time and Place

• In Event: Rethinking Place and Spatial Dynamics

Abstract

Many theories and hypotheses have been put forward to explain the period of rapid growth in the incarceration rate of the United States known as mass incarceration (occurring roughly from the mid-1970’s to the mid-2000’s). Macro-level theories and hypotheses point to few factors as being the main determinants of mass incarceration: rising crime rates, structural racism and racial segregation, increased political competition among USA political parties, the increasing political salience of crime, and the hyper-local nature of the administration of criminal justice in the USA. To better understand the relative predictive power of each theory, we propose training and testing a series of machine learning models (e.g, boosted decision trees, support vector machines, LASSO regression) operationalizing each theory to predict the incarceration rate of counties from 1982 - 2016 using data from the Vera Institute Incarceration Trends. Additionally, we aim to test if: 1) certain theories are more or less predictive during different time periods, and 2) if certain theories are more or less predictive in certain geographic regions of the United States.

Authors

• Joe Risi, Penn State University

• Corina Graif, Penn State University