I examine the extent to which the returns to college majors are influenced by selective migration and occupational choice across locations in the US. To quantify the role of selection, I develop and estimate an extended Roy model of migration, occupational choice, and earnings where, upon completing their education, individuals choose a location in which to live and an occupation in which to work. In order to estimate this high-dimensional choice model, I make use of machine learning methods that allow for model selection and estimation simultaneously in a non-parametric setting. I find that OLS estimates of the returns to business and STEM majors relative to education majors are biased upward by 15% on average. Using estimates of the model, I characterize the migration behavior of different college majors and find that migration flows are twice as sensitive to occupational concentration as they are towage returns.