Design thinking (DT), a problem-solving approach emphasizing creativity and innovation, is gaining popularity in education. Introduced by Rowe (1987) and first applied in education in 2005 (Çeviker-Çınar et al., 2017), DT is now used across various educational levels and contexts (Pande and Bharathi, 2020; Aris et al., 2022). While proponents suggest DT fosters creativity, problem-solving skills, and innovation (Beckman and Barry, 2007; Lor, 2017; Retna, 2016), its impact on student learning remains unclear. Previous studies offer conflicting results, highlighting the need for a comprehensive meta-analysis to assess DT's overall effectiveness and identify potential moderating factors. This study addresses this gap by conducting a meta-analysis to quantify DT's effect on student learning and explore the influence of various contextual factors.

Existing research on DT's impact on student learning is fragmented and inconclusive. Some studies report significant positive effects (Albay and Eisma, 2021; Bawaneh and Alnamshan, 2023; Chang and Tsai, 2021; Dawbin et al., 2021; Hsiao et al., 2017; Kuo et al., 2022; Ladachart et al., 2022; Lin et al., 2020a; Liu and Ko, 2021; Nazim and Mohammad, 2022; Padagas, 2021; Pratomo and Wardani, 2021; Simeon et al., 2022; Tsai, 2015; Ziadat and Sakarneh, 2021), while others show no significant impact (Khongprakob and Petsangsri, 2022; Kuo et al., 2022; Lin et al., 2020b; Yalçın and Erden, 2021) or even negative correlations (Chou and Shih, 2022; Lake et al., 2021). This inconsistency necessitates a systematic review to clarify DT's overall impact and identify factors that contribute to its success or failure. The lack of specific guidance on DT implementation and the challenges associated with its open-ended nature (Becker and Mentzer, 2015) further emphasize the need for a rigorous meta-analysis.

This meta-analysis followed the process outlined by Field and Gillett (2010). Data were collected from Web of Science, Scopus, and Google Scholar using keywords such as "Design Thinking" and "Learning Performance." The search period was January 2005 to June 2023, resulting in 1204 initial articles. After removing duplicates and screening titles and abstracts, 84 articles met preliminary inclusion criteria. A thorough full-text review yielded 25 eligible peer-reviewed articles. To minimize bias, multiple databases were searched (Stang, 2010), strict inclusion criteria were defined (study purpose, design, intervention, data availability, language), and literature quality was assessed using Downs and Black's (1998) criteria (all studies scored 18-21 out of 27). Eight potential moderators were coded: learning outcome (academic achievement, self-efficacy, motivation, problem-solving, creative thinking, engagement), treatment duration (<1, 1-3, >3 months), class size (1-30, 31-50, 51-100, >100), grade level (kindergarten, primary, junior high, high school, university), subject (STEM, non-STEM, multidisciplinary), DT model (9 models were identified), team size (1-4, 5-7, ≥8), and region (Asia, America, Austria, Europe, Africa). Pearson's correlation coefficient (r) was chosen as the effect size (Borenstein et al., 2005), and Fisher's Z-transformation was used for analysis (Lei et al., 2020). Publication bias was assessed using funnel plots, fail-safe N, and trim-and-fill methods. Heterogeneity was tested using Q and I² statistics (Higgins et al., 2003), and sensitivity analysis was performed using the one-study-removal method.

The meta-analysis revealed an overall upper-medium positive effect of DT on student learning (r = 0.436, 95% CI [0.342, 0.525], p < 0.001). Publication bias was deemed negligible. High heterogeneity (Q = 554.908, p < 0.001, I² = 92.611%) necessitated a random-effects model and moderator analysis. Moderator analysis indicated significant effects for learning outcome, treatment duration, grade level, DT model, and region. For learning outcomes, engagement (r = 0.740) and motivation (r = 0.608) showed the strongest effects. Class size did not significantly moderate the effect. Treatment duration showed a larger effect for durations of ≤1 month and ≥3 months compared to 1-3 months. High school and university students showed greater effects than primary and kindergarten students. Multidisciplinary subjects demonstrated a greater effect than STEM or non-STEM subjects. The OSIP (r = 0.766) and EDIPT (r = 0.522) models showed the largest effects. Team size did not significantly affect outcomes. Finally, Africa showed the strongest effect, followed by Asia.

The findings confirm the positive impact of DT on student learning, particularly in specific contexts. The significant moderating effects suggest that the effectiveness of DT is not universal but depends on various factors. The stronger effects on learning engagement and motivation highlight the importance of creating engaging and motivating learning experiences using DT. The duration effect suggests the optimal duration may require further investigation, as the 1-3 month duration produced the lowest effect size. The variations across grade levels suggest a need for tailored approaches based on students' cognitive development. The superiority of multidisciplinary contexts emphasizes the potential of DT to foster interdisciplinary collaboration and problem-solving. The variation in effects between DT models suggests that careful model selection is crucial for maximizing effectiveness. The significant regional differences suggest the importance of considering cultural contexts when implementing DT. This study contributes to a better understanding of DT's potential and limitations.

This meta-analysis provides robust evidence supporting the positive impact of design thinking on student learning. However, the effectiveness of DT is context-dependent, influenced by factors such as learning outcome, treatment duration, grade level, DT model, and region. Future research should focus on exploring these moderating effects further, particularly on under-represented regions, grade levels, and DT models.

This study has several limitations. First, regional, grade-level, and DT model representation was uneven. Second, the analysis was restricted to English-language publications. Third, considerable heterogeneity suggests potential moderators may have been overlooked. Fourth, the moderate sample size limits the generalizability of the findings. Fifth, a meta-analysis might not fully capture the complexity of DT in education. Future work could address these issues by expanding the inclusion criteria, exploring additional moderators, and using more robust methodologies.