A Meta-Analysis of ChatGPT's Influence on Learning Achievement

초록

This meta-analysis synthesized empirical findings on the influence of ChatGPT on learning achievement. An electronic database search using Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines was conducted with relevant keywords to identify eligible research studies published between November 2022 and December 2024. A total of 22 eligible publications that met our inclusion criteria were reviewed. ChatGPT had a moderate positive effect on learning achievement (g = 0.573), indicating its potential to enhance learning outcomes. Subgroup analysis revealed that ChatGPT had a larger effect on middle and high school students (g = 0.928) than on undergraduates (g = 0.538), although the difference was not statistically significant. This finding highlights the importance for instructors and educational practitioners to consider the applications of ChatGPT in middle and high school settings. No significant statistical differences were found among the cognitive, affective, and metacognitive domains. Given that nearly half the studies focused on the cognitive domain, it is important to diversify the application of generative AI across a variety of subjects in different learning domains. The most frequently used instructional approaches with ChatGPT applications were lectures (22.1%) and self-regulated learning (16.3%). The effect of ChatGPT was largest on self-regulated learning (g = 1.115), followed by case-based learning (g = 0.836), while its effect was smallest on game-based learning (g = 0.092). This study was conducted within two years of ChatGPT's emergence, limiting our ability to analyze a large number of publications. Nevertheless, this study offers meaningful implications for future research on the application of ChatGPT for educational purposes.

키워드