Association for Public Policy Analysis & Management

The idea motivating half a century of youth employment programs has been that offering training and search assistance to disconnected youth can improve future labor market outcomes, and perhaps as a result, increase the opportunity cost of crime. This paper uses two field experiments, each targeting different populations of disadvantaged youth, to argue for a rethinking of what these programs do and for whom. We report the effects of the program on crime, schooling, and employment outcomes over the 2-3 years after random assignment. We also use a new machine learning method that predicts treatment heterogeneity from observables to describe who benefits, as well as to explore potential mechanisms. We conclude that these programs can generate substantively important behavioral change, but for different outcomes and youth than those most studied in the literature.