Using AI to predict grade point average from college application essays

Jonah Berger and Olivier Toubia used natural language processing to understand what drives academic success. The authors analyzed over 20,000 college application essays from a large public university that attracts students from a range of racial, cultural, and economic backgrounds and found that the semantic volume of the writing, or how much ground an application essay covered predicted college performance, as measured by grade point average.

They published their findings in PNAS Nexus.

Essays that covered more semantic ground predicted higher grades. Similarly, essays with smaller conceptual jumps between successive parts of its discourse predicted higher grades.

These trends held even when researchers controlled for factors including SAT score, parents’ education, gender, ethnicity, college major, essay topics, and essay length. Some of these factors, such as parents’ education and the student’s SAT scores, encode information about family background, suggesting that the linguistic features of semantic volume and speed are not determined solely by socioeconomic status.

According to the authors, the results demonstrate that the topography of thought, or the way people express and organize their ideas, can provide insight into their likely future success.