Abstract
We study the choice of acquiring STEM college education using variation induced by the proximity to universities offering different types of programs. We adopt the methodology by Heckman and Pinto (2018) allowing the identification of the distribution of response types and treatment effects with multiple unordered choices. We combine survey data for Italy with historical information about the founding dates of all universities and faculties. We find that most compliers are women at the margin of choosing STEM education versus not going to college. Expanding the supply of STEM education could reduce the gender gap in STEM by 20%.