Meta-Analysis of Generative AI Feedback Effects on Learning in Higher Education

52 independent studies provide coded variables and effect sizes on how Generative AI feedback shapes human-AI interactions in higher education. Data was systematically extracted from academic databases following PRISMA guidelines, covering variables like intervention duration, course type, and interaction mode. Shengwu Pan published this dataset on figshare in April 2026.

Use Cases

• Analyze the relationship between GenAI feedback types and learning outcomes based on the coded functional roles (Diagnostic/Corrective, Coaching/Scaffolding, etc.).
• Compare the effectiveness of different interaction modes (Student-Machine vs. Teacher-Student-Machine) on constructs like Behavioral Intention and Cognitive Development.
• Investigate how participant proficiency levels (Novice, Intermediate, Proficient) moderate the impact of GenAI interventions.
• Examine the influence of instructional models (Personalized Learning, Problem-Based Learning) on the measured effect sizes.
• Study the role of interface design (Structured Prompts, Free Dialogue) and communication channels (Text-Visual, Multimodal) in learning interactions.

Strengths

• Data is derived from a systematic review of 52 independent studies, providing a structured meta-analytic foundation.
• Variables are explicitly categorized (e.g., 12 studies with interventions <1 week, 22 in STEM courses), allowing for granular analysis.
• Includes calculated effect sizes (Hedges' g) for four primary learning constructs: Behavioral Intention, Cognitive Development, Emotional and Attitudinal outcomes.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain analytical methods.
• The dataset's 1.8 MB size suggests it contains summary or coded data rather than raw participant-level records.

Provenance

Source

figshare, author Shengwu Pan.

Collection Method

Data extracted through systematic literature searches across WOS, Scopus, and CNKI, screened per PRISMA guidelines.

Freshness

Last updated 2026-04-27 05:04:31; freshness should be verified.