Can AI Enhance University Admissions Decisions?

Admissions officers at universities typically use their judgements on information provided in applications to predict outcomes like academic performance, leadership potential, and long run career success of the applicants and use these judgments in admission decisions. However, there has been little empirical work assessing whether such judgements are consistent with actual outcomes and university admissions are indeed optimal, considering their stated objectives---and if not, how AI can improve admission decisions.

While data on academic outcomes at the university is easily accessible, the data on leadership and career outcomes are not typically available. In this project, we will use LinkedIn information on the applicants from past year to track their career outcomes. We will then use AI models to assess whether admissions officer judgments are consistent with actual outcomes; and if there are systematic deviations, understand the areas of deviations so that the model can serve as a tool/input to improve decision making. Further, the arrival of LLMs and other AI models that can deal with unstructured data also allows us to systematically code essays and video interviews----comparing the outcomes from these admissions inputs into evaluator judgements versus actual long-term outcomes. These comparisons can provide substantive insights that can improve fairness and transparency on admissions decisions.

The operational elements focus of this project includes:

• Scraping data from LinkedIn on career trajectories of applicants
• Leveraging LLMs/AI tools to convert unstructured data into substantively relevant variables that can be useful for predicting student outcomes.
• Developing effective ML methods to link student characteristics with mid-term (e.g., academic performance in core courses) and long-term (e.g., career advancement, leadership roles) outcomes.
• Gain insight into which applicant variables (e.g., test scores, qualitative assessments, extracurricular involvement) are most predictive of outcomes, using both structured and unstructured data---and where there are systemic gaps with respect to human judgements and admissions decisions
• Designing and testing AI-driven decision aids that help admissions officers weigh applicant attributes fairly and effectively.

Students working on this project will gain hands-on experience collecting and analyzing both structured and unstructured data. They will also engage with the theoretical literature on predictive modeling and human-AI collaboration in decision-making.