Work

Solving the 2-Sigma Problem With ChatGPT: A Case Study on the Double-Entry Accounting System

Year: 2026
Type: article
Study design: Clinical Trial
Abstract: This study examined whether structured ChatGPT tutoring could approximate the learning gains historically associated with one-on-one human tutoring. Grounded in cognitive load theory and informed by Bloom’s two-sigma benchmark, the investigation used double-entry accounting as a procedurally dense testbed requiring coordinated schema construction and rule-based reasoning. A randomized proof-of-concept design assigned thirty first-year accounting students to either human tutoring or ChatGPT tutoring across an eight-week instructional interval. Learning outcomes were assessed at pretest, posttest, and delayed retention. Repeated-measures analyses revealed substantial within-subject learning gains across both instructional modalities. Performance improved significantly from baseline to post-instruction assessment, indicating acquisition of procedural schemas. A significant Group × Time interaction demonstrated differentiated learning trajectories, with students receiving ChatGPT tutoring exhibiting steeper acquisition gains and greater performance stability across the retention interval. Nonparametric robustness analyses using the ANOVA-Type Statistic reproduced the same inferential pattern. Survey findings indicated high engagement and instructional clarity across both conditions. Within the constraints of a proof-of-concept design, the magnitude of gains observed in the ChatGPT condition approached levels historically associated with individualized tutoring, suggesting that principled instructional structure may narrow the gap between personalized and scalable academic support.
Language: English
Open Access status: green
OpenAlex ID: W7204616583
FWCI:
Cites: 213
Cited by:
Related to: 10