Tuesday, September 20, 2022
2:00 - 4:00 PM (PDT) at Berkeley Way West 4500 and on Zoom
Early warning systems are algorithmic prediction tools that have recently become part of the de facto approach to improving high school graduation rates in US public schools. Despite widespread adoption and significant financial investment, the bottom-line impact of these systems on graduation outcomes is largely unknown. In this work, we draw on nearly a decade’s worth of data from an early warning system designed by the Wisconsin Department of Public Instruction to provide the first large-scale evaluation of these programs.
We find that while the risk assessments made by the prediction system are highly accurate, there is no evidence that these predictions have meaningfully impacted graduation outcomes. By examining what predicts future dropout in the first place, we find that these individual outcomes are robustly predictable largely due to the strength of environmental features describing a student's surrounding community. We conclude that dropout prevention efforts are not bottlenecked by the lack of early identification strategies, but rather by the difficulty of implementing effective educational interventions.
About the speaker:
Juan C. Perdomo is a graduate student in the EECS Department at UC Berkeley where he is advised by Professors Peter Bartlett and Moritz Hardt. His research centers on the empirical and theoretical foundations of machine learning within social systems. His work has been supported in the past by a National Science Foundation Graduate Research Fellowship.