Topic Distributions in Latinx College Admissions Essays by Geography, Language, and Parental Education

Test-optional and test-blind college admissions have become increasingly popular in the US due in part to the strong correlation between SAT scores and income. Bias associated with any aspect of an admissions file is argued to degrade the process as unfair and non-meritocratic. As test-optional admissions become even more widely adopted due to the COVID-19 pandemic, the presumptive reweighting of other application materials raises the need to examine these other components with similar scrutiny. For Latinx applicants, who are vastly underrepresented in higher education, this imperative is even greater. The application essay, a student-produced component, is at once an opportunity for both bias and democratization in selective admissions. There is a burgeoning but limited literature that shows that the content of admissions essays are strongly related to gender among Latinx applicants and have a stronger relationship with household income than SAT scores do. This poster presents preliminary results from computational analyses of 79,330 admissions essays written by 39,665 Latinx applicants to the University of California in 2016. We extend related work by examining the relationship between geographic context, reported first language, and parental education levels of applicants to the topical and thematic content of their essays. To model the content of the essays, we use correlated topic modeling; to compare essays, we use traditional statistical tools such as t-tests and boxplots on the individual topic scores for each essay. We discuss how the patterning we find complicates and brings nuance to ongoing conversations about fairness, access, and equity in college admissions beyond the binary inclusion/removal of single components like test scores.