Effective implementation of the Sport Education Model in physical education: A meta-analysis of participant and intervention characteristics

PLOS One | https://doi.org/10.1371/journal.pone.0331228 October 16, 2025 5 / 17 Statistical analyses If three or more studies had sufficient data to compute the effect size (ES), we performed a meta-analysis [29,30]. Hedg- es’g (the effect size) is calculated as the means and standard deviation (SD) from pre- and post-measurement data of learning efficiency for the SEM group and control group. Data are adjusted for the post-intervention pooled standard deviation and a random-effects model is utilized so that underlying heterogeneity between studies could influence the SEM effect [29,31]. The ES values were accompanied by 95% confidence intervals (CI), and the ES values were inter- preted according to the following scales: (a) values greater than 0.8 are considered “large,” (b) scores between 0.5 and 0.8 are classified as “moderate,” and (c) values between 0 and 0.5 are described as “small” [28]. For studies involving multiple different intervention groups, the sample size of the control group is divided proportionally to ensure that all group members are comparable [32]. If there is not enough data stated (e.g., missing in graphs or absent) we try to contact with corresponding author. If there is no response from the authors or they cannot provide the data, we exclude the study. The I2 statistic is used to assess the heterogeneity of the studies, with values below 25% indicating low heterogeneity, 25%– 75% representing moderate heterogeneity, and above 75% indicating high heterogeneity [33]. An extended Egger test is

conducted to test the risk of publication bias [34]. The significance level used in the meta-analysis is p < 0.05 to establish statistical significance. All analyses are performed using Comprehensive Meta-Analysis software (Version 3.0; Biostat, Englewood, NJ, USA). Results Study selection As shown in Fig 1, a total of 769 articles were retrieved from the databases, and an additional 4 articles were obtained through references and Google Scholar. After manually removing duplicate articles (343 articles), 430 articles remained. Following a preliminary assessment of the titles and abstracts of these records, strictly adhering to the inclusion and exclusion criteria, 60 articles were identified as eligible for full-text review. After a thorough evaluation of the full texts of these 60 articles, 45 were excluded, resulting in a final selection of 15 studies that met all the criteria for this meta-analysis (See S3 Appendix. Description of 430 records). Risk of bias in individual studies and certainty of evidence Three cluster randomized controlled trials [20,35,36] were assessed using the ROBINS-2 tool, while 12 non-randomized controlled trials [7,8,21,37–45] were evaluated using the ROBINS-I tool. Among these trials, 2 showed an overall high (or serious) risk of bias [35,42], 11 demonstrated a moderate risk of bias [7,8,20,21,36,38–41,43,45], and only 2 exhibited a low risk of bias [37,44], as illustrated in Figs 2 and 3. The visualization of the ROB2 assessment results is presented in Fig 2.

In the three cluster randomized controlled trials [20,35,36], none reported the methods for random sequence generation, merely stating adherence to randomization methods, resulting in a judgment of moderate risk of bias during the randomiza- tion process. One study [35] was identified as having a high risk of bias due to missing data, which was related to a dropout rate exceeding 10%. The visualization of the ROBINS-I assessment results is shown in Fig 3. One study [42] was deemed to have a high risk of bias due to re-grouping during the intervention process; the primary reasons for the moderate risk of bias in other studies were the lack of detailed reporting on missing participants or the absence of blinding of outcome assessors. Table 1 presents the results of the GRADE analysis. The GRADE analysis revealed that the certainty of the evidence supporting the findings was very low. Study characteristics The characteristics of the 15 included studies are presented in Table 2. Among them, 10 studies were published between 2020 and 2025, involving a total of 2,890 participants, with 1,318 in the experimental group and 1,572 in the control group.