Quantitative analysis of demographic gaps in STEM persistence

It is well known that, at the undergraduate level and beyond, many STEM fields, particularly ``hard sciences'' and engineering, are far less diverse than the US population more broadly. This is usually modeled by the ``leaky pipeline,'' which posits that we lose STEM talent, particularly from historically marginalized populations, at many points throughout K-16 education -- including after students arrive in college. I will present data showing that, in fact, historically marginalized students generally intend to major in STEM at \textit{higher }rates than overrepresented students but graduate with a STEM degree at lower rates -- indicating much of the leaky pipeline is within the university itself. I will present a detailed quantitative analysis exploring the factors correlated with receiving a STEM degree for different historically marginalized populations. I find that the data tell different stories for women, students of color, first generation students, and students from lower income families. This analysis is limited to a single public research institution in the United States, but I hope to encourage other institutions to conduct similar analyses to see how consistent these patterns are.