<![CDATA[Accounting students often work on computationally intensive procedural tasks such as journal entries and depreciation schedules, where feedback is often delayed and generic, so it is unclear whether AI-based feedback interacts with learners' processing preferences. This study examines how three levels of AI-based feedback, rule-based direct feedback, retrospective ML personalization feedback, and prospective ML predictive feedback combine with learning styles (visual, kinesthetic, and convergent, classified through the Kolb Learning Style Inventory 4.0) to shape accounting learning outcomes. A 3×3 factorial experiment between subjects (N = 135) assigned students to one of nine combinations of force-based feedback. Because pre-test scores differed systematically across different pre-intervention feedback conditions, learning improvements (after the test minus pre-tests), rather than post-test scores, became the primary outcome, with naïve post-test and covariate analyses reported as convergence checks, followed by Tukey's HSD comparisons in which the omnibus effect was significantly relevant. Feedback conditions, learning styles, and interactions each contributed significantly to learning improvements, but the patterns were not uniform: some learning style groups gained relative gains regardless of feedback levels, while others showed feedback-dependent improvements that did not consistently support more technologically advanced conditions. These results suggest that AI-based feedback values are dependent on learner characteristics, not universal, and simpler feedback is not necessarily inferior after baseline differences are taken into account. Given the simple single-location sample, unresolved fundamental imbalances under various conditions, and reliance on composite outcome sizes, these findings should be treated as preliminary evidence justifying replication before force-based feedback configurations are adopted in practice.]]>
Copyrights © 2026
