This dissertation focuses on barriers to access in education for individuals from underrepresented or underserved populations.

Chapter 1 examines whether increased access to mental health services impacts adolescent mental health and behavior. I study school-based health centers, full-service clinics operating within K-12 schools. These centers aim to bridge the gap in healthcare access in low-income communities by offering physical, mental, and reproductive health services at low to no cost. Using a staggered difference-in-differences approach and a propensity-score matched control group, I find that the opening of a new school-based health center decreases suspension rates by around 1.1 percentage points (27% of the baseline rate). These effects are primarily driven by decreases in suspensions from "disruptive behavior", a known symptom of untreated behavioral issues in adolescents.

Chapter 2 studies whether low-touch information provision can overcome barriers to entry in the undergraduate Economics major that may disparately impact underrepresented minority (URM) students. I run two waves of a large-scale field experiment in an introductory undergraduate Economics course to test an email intervention containing information on research topics, careers, expected income, and diversity in Economics. For URM students, the intervention increases the probability of enrolling a subsequent Economics course by 9.9 percentage points and induces lower-performing students to enroll. Student survey responses suggest that the intervention has the strongest effects on URM students' beliefs about the breadth of research topics studied by Economics.

Chapter 3 aims to assess the relationship between GPA signals, gender, and persistence in STEM fields. Using nationally-representative panel data on college students we find that given the same GPA signals, women are more likely to switch majors and more likely to switch out of STEM entirely than men. This differential response is strongest at low average GPAs in STEM courses. Controlling for selection into initial fields, we find no quasi-experimental effect of low GPA for women in STEM; however, we continue to find differential responses to low GPA between men and women initially in STEM fields. This suggests that gendered responses to the same ability signals may explain part of the gender gap in STEM degrees.

This dissertation consists of three papers on the economics of education. All pertain to the notion of higher- or lower-achieving schools, in some way.

Chapter 1 estimates the magnitude by which the achievement level of a student’s high school affects college admissions once students make the decision to apply. Using administrative data from the University of California, I find a significant negative association between a school’s overall achievement level and the probability of admission at 5 out of 8 UC campuses between 2001–2006, including 5 of the 6 most selective. The average magnitude across all UC campuses is about -0.8 percentage points for a one-decile increase in school-average achievement.

Chapter 2 uses lottery data from San Diego Unified School District to study the long-term effects of school choice on students’ postsecondary outcomes. This paper contributes to the school choice literature by being among the first to focus on postsecondary enrollment and graduation, rather than nearer-term outcomes such as standardized test scores in the immediate years following the lottery.

Despite large effects on what secondary schools students attend, we find mostly insignificant effects on postsecondary enrollment and graduation outcomes from winning a school choice lottery.

Chapter 3 examines the sensitivity of school value-added estimates to the methodology used. Education scholars frequently use “value-added” models to assess schools’ effectiveness. Yet there are many variations of the model that scholars sometimes use, which may affect corresponding results. Using a menu of three model classes and three control specifications each—for a total of 9 specifications—we find a relatively modest sensitivity to the control specification, and a somewhat greater sensitivity to the class of value-added model. Precision appears sensitive to the inclusion of school-level controls and whether researchers use panel instrument techniques to address dynamic panel bias. We also observe a moderate positive correlation between value-added estimates and schools’ average levels of test scores.