Problem: This project aims to understand how machine learning predictions improve college choice decisions in centralized admissions systems.

Sources: The project rests on the use of administrative and survey data.

Methods: A RCT design is used to evaluate the relative importance of each key element in the typical machine learning practice loop including (1) human expertise or domain knowledge, (2) data, (3) algorithms, and (4) computation.

Challenges: This project meets challenges in collecting administrative data on historical admissions outcomes for each college-major in each state in China, building an “optimal’’ prediction model, and recruiting survey participants.

Findings: I expect to identify key components that enable machine learning predictions improve decision-making and thus propose scalable interventions to improve  students’ college choice decisions and admissions outcomes.