Work and Crime: A Causal Treatment Effects Survival Model

• In Event: New Applications and Controversies in Survival and Trajectory Models

Abstract

Policy makers, politicians, and criminologists often assume that jobs will reduce crime. The strongest evidence on this assumption comes from the National Supported Work Demonstration project, an $82.4 million study that randomly assigned 5,005 chronically unemployed addicts, drop-outs, and offenders to Supported Work or a control group. The original evaluation failed to find an overall treatment effect. Some evidence was found for treatment heterogeneity: Supported Work appeared to reduce offending among early referrals to the program, drug addicts, and older participants. During the course of the experiment, there were substantial departures from randomization as some experimentals quit their jobs, while some controls got jobs on their own. We explore using an instrumental variable approach to modeling departures from randomization within survival models of desistance, which raises challenging statistical and methodological issues. Using the new model, we reevaluate the substantive results of the experiment on offending.

Authors

• Jun Hwang, University of Washington

• Ross L. Matsueda, University of Washington

• Elena A. Erosheva, University of Washington

• Christopher Uggen, University of Minnesota