Static, Dynamic, or Adaptive? Comparing Interface Scaffolding Methods in an Intelligent Tutoring System
Learning@Scale: The Seventh Annual Workshop on A/B Testing and Platform-Enabled Learning Engineering
2026
Abstract
Intelligent Tutoring Systems (ITS) support enhanced student learning outcomes, but sustained engagement can be a challenge for adult learners. In this work, we present an A/B experiment comparing three scaffolding interfaces in the Apprentice Tutors ITS, deployed as supplemental support across nine sections of College Algebra at a technical college: Full Scaffolding, which presents all fine-grained problem steps; Adaptive Scaffolding, which fades steps based on estimated mastery; and On-Demand Scaffolding, which presents only a final answer field with optional steps. We find that both dynamic conditions improve course outcomes relative to Full Scaffolding, with students assigned to On-Demand earning significantly higher overall grades (b = 9.82, p = .018) and completing substantially more problems (b = 239.04, p = .001). Causal mediation analysis indicates that approximately 28% of the On-Demand grade advantage can be attributed to increased tutor engagement and more problems solved. These results suggest that for supplemental educational technology in higher education, improving learner retention can be a gateway to improving learning outcomes.
BibTeX
@inproceedings{reddig-pele-2026,
title = {Static, Dynamic, or Adaptive? Comparing Interface Scaffolding Methods in an Intelligent Tutoring System},
author = {Reddig, Jennifer and Gupta, Adit and Siddiqui, Momin and Izmirilian, Elias and MacLellan, Christopher J.},
booktitle = {Learning@Scale: The Seventh Annual Workshop on A/B Testing and Platform-Enabled Learning Engineering},
year = {2026},
}