In 1997 a social program called PROGRESA was introduced in Mexico using a design for a randomized experiment. We exploit a build in, but neglected, discontinuity in the eligibility rule and use the quasi-experimental Regression-Discontinuity design in order to estimate marginal average treatment...