IFS Seminar presented by Jonathan Guryan (Northwestern)

Abstract: We would like to know which social programs work at large scale before we scale them. But it would seem the only way to tell whether a program works at scale is to test it at scale. We propose a way out of this Catch-22 for one important scale-up challenge: inelastically supplied inputs. We show it is possible to operate a program at modest scale but learn about treatment effects at a much larger scale by randomly sampling the inputs the provider would have hired if they operated at the larger scale. We present two empirical applications, one related to police hiring, and the other related to an in-school intensive math tutoring program.