Abstract Physician supply in medically underserved areas has long been an interest of health and labor economists. I employ a rich database containing the location of physicians at 5-year intervals to compare the types of locations chosen by alumni and nonalumni of a United States program charged with increasing physician supply. Using a multinomial logit model with discrete unobserved heterogeneity to account for endogeneity of enrollment in the program, I find that eliminating the program would decrease the supply of physicians in medically underserved communities by roughly 10%.