In the era of 21st-century digital transformation, computational thinking (CT) has emerged as a core competency, and in Korea, software (SW) education was formally institutionalized through the 2015 revised national curriculum. This study represents a...
In the era of 21st-century digital transformation, computational thinking (CT) has emerged as a core competency, and in Korea, software (SW) education was formally institutionalized through the 2015 revised national curriculum. This study represents a second-generation meta-analysis that extends prior research limited to the early stage of SW education implementation, by systematically examining the effects of SW education on CT development in domestic K–12 contexts published between 2015 and 2025. A total of 356 effect sizes were analyzed using a random-effects model, with moderator variables including educational level, instructional approaches, and AI-assisted implementation, and publication bias was assessed.
The overall effect size was Δ ̂ = .689, indicating a medium-to-large effect, and publication bias analyses confirmed the stability and reliability of the results (p = .222). Subdomain analyses revealed stronger effects in cognitive CT components such as algorithmic thinking and problem decomposition, whereas relatively lower effects were observed in affective components such as self-efficacy. Moderator analyses showed statistically significant effects for educational level (Qb = 647.90, p < .001) and AI-assisted implementation (Qb = 7.17, p = .007). The largest effect was found among upper elementary students (Δ ̂ = .889), followed by high school (Δ ̂ = .653), lower elementary (Δ ̂ = .505), and middle school students (Δ ̂= .503). AI-assisted learning environments (Δ ̂ = .543) demonstrated lower effects than non-AI environments (Δ ̂ = .740); however, this finding should not be interpreted as evidence of ineffectiveness of AI use, but rather as reflecting the early stage of AI-assisted instructional models in Korean educational settings and the limited accumulation of empirical research.
Comparative analysis with the first-generation meta-analysis indicated that the effects of SW education were maintained or strengthened in the second generation, with particularly stable increases observed among general student populations, suggesting that SW education has evolved into a form of broadly effective instruction beyond specific learner groups.
The findings of this study offer several implications. First, the higher effects observed in upper elementary and high school levels suggest that the effectiveness of SW education may be optimized at particular developmental stages. Second, there is a need to move beyond cognitively focused approaches toward assessment systems that integrate affective dimensions such as learning motivation and self-concept. Third, to realize the educational potential of AI tools, systematic integration within instructional models and enhanced teacher capacity are required, rather than mere technological adoption. Overall, this study provides empirical foundations for restructuring CT education within the implementation of the 2022 revised curriculum and AI–SW convergent learning environments.