This project has two main research goals that will lead to two related working papers and a grant application to a funding agency. The first is to document the adolescent skill (di)convergence by gender over the last four decades in terms of math, verbal, science, mechanical, and social skills and to use machine learning methods to identify the skills that are most predictive of one’s educational attainment, choice of college major and occupation. Focusing on gender segregation in majors and occupations within the context of the non-/pure-/applied-STEM fields, this project will identify the factors contributing to field segregation (e.g., early-life skills, academic preparation in high school, etc.) in terms of non-/pure-/applied-STEM majors and occupations. It will also perform a nonparametric decomposition to empirically assess the relative importance of different factors. A second goal of this project is to structurally estimate a model of high school course-taking, college attendance decisions, college major choices and occupational choices. Ultimately, the estimated model will be used to conduct policy experiments regarding high school graduation requirements, and financial incentives (such as college scholarships) to explore effective ways of increasing female representation in STEM fields.
This project analyzes the factors underlying underrepresentation of women in certain STEM fields. Although there has been substantial male-female wage convergence over time, there remains a persistent gap among college graduates that has been shown to be attributable, in part, to males and females choosing different college majors and occupations. Majors in applied STEM fields, such as computer science and engineering, are among the highest paid and are also those in which the representation of women is 20% or lower. Occupation and industry segregation was recently found by Cortes and Pan (2018) to constitute the largest portion of the explained component of the gender wage gap in 2010.
This proposal consists of two main lines of research that will lead to two related working papers and a grant application to a funding agency. The first project will use the NLSY79 and NLSY97 longitudinal datasets to document the adolescent skill (di)convergence by gender over the last four decades in terms of math, verbal, science, mechanical, and social skills and estimate machine learning models to identify the skills, high school course-taking and family background characteristics that are most predictive of one's educational attainment, choice of college major and occupation. We will distinguish among non-STEM, pure-STEM and applied-STEM majors, as the pattern of female entry into pure-STEM and applied-STEM categories has been quite different. We will perform a nonparametric decomposition to empirically assess the relative importance of different factors. A second related project will use the same data to structurally estimate a discrete choice dynamic programming (DCDP) model of high school course-taking, college attendance, college major choices and occupational choices, controlling for family background and gender. This model will inform how skill accumulation prior to college contributes to gender disparities in college attendance, college major and occupation choices. The estimated model will also be used to conduct policy experiments. We will analyze whether and to what extent closing early test score gaps, changing high school graduation requirements and providing financial incentives (such as scholarships to women to major in certain fields) are effective ways of increasing female representation in STEM fields.