ASA Case Study: Machine Learning to Support Success & Combat Inequity
ASA Research, LLC, has developed a model that predicts next-term retention at a large four-year public university approximately three to five times better than alternatives. The model leverages 80 variables from a single university’s student-level administrative data and 33 variables from public data, captured over 12 years, represented by the 2010 through 2021 student cohorts. Like all machine learning models, ours mines historical data for mathematical relationships between predictor variables and the outcome of interest, then “learns” to generalize these relationships so that it can predict future outcomes when only the predictor variables are known.
While developing the model, our team worked closely with a university’s undergraduate education division to prioritize transparency and equity. To ensure that the model combats existing inequities instead of reinforcing them, we included a range of voices when making key decisions, including students’ perspectives, and we carefully tested our model for algorithmic bias.
Once machine learning algorithms predicted which fall cohort students were at highest risk of not persisting into the winter term, our team worked with advisors in undergraduate education to provide on-time interventions to students identified in the model, with an emphasis on equity and not perpetuating bias. We provided 385 students in the intervention group with early advising from the advising team.
