When Scale and Replication Work: Learning from Summer Youth Employment Experiments

Because successful human capital interventions often fail to scale or replicate, public investment decisions require understanding how program size, context, and implementation shape program effects. This paper uses two new randomized controlled trials of summer youth employment programs in Chicago and Philadelphia to demonstrate how multiple experiments can help explain replicability and inform the expansion of promising approaches. Even when these programs grow or change models across contexts, participation consistently reduces criminal justice involvement. It may also decrease the need for child protective services and behavioral health treatment. Experimental variation in program model and local provider generates no detectable heterogeneity, suggesting that effects replicate partly because variability in implementation does not matter. There is, however, individual-level heterogeneity that explains differences in effect magnitudes across populations and informs optimal targeting; youth at higher risk of socially costly outcomes experience larger benefits. Identifying more interventions that combine this pattern of treatment heterogeneity with robust replicability could aid efforts to reduce social inequality efficiently. Sara Heller University of Michigan Department of Economics 611 Tappan Street, Lorch Hall Room 238 Ann Arbor, MI 48109 and NBER sbheller@umich.edu An online appendix is available at http://www.nber.org/data-appendix/w28705