Latent Stories, Latent Space: Unsupervised Discovery Methods for Educational Counternarratives

• In Event: Critical Methodological Innovations in Quantitative Research

Abstract

This study introduces a novel framework for discovering potential counternarratives in education data using unsupervised machine learning. We trained an autoencoder—a type of neural network—on survey responses from 1,134 first-year college students and analyzed the latent space to identify educational trajectories that subvert expectations rooted in dominant narratives. Rather than specifying which stories mattered most, we defined both normative and counternarrative reference cases to validate model behavior. High reconstruction error and distinct cluster profiles flagged students with unusual experiences, including groups that challenged prevailing educational narratives. This approach supports “unforgetting” by surfacing meaningful patterns without imposing interpretation. Our method offers researchers a discovery tool grounded in algorithmic humility and creates space for community-led inquiry into overlooked educational trajectories.

Authors

• Christopher Irwin, Florida International University

• Remy Dou, Florida International University

• Sue Sunbury, Harvard University

• Philip M. Sadler, Harvard University

• Gerhard Sonnert, Harvard University