Recreational Football for Young-to-middle-aged Inactive Men

Randomization:

The participants were randomized 1:2 to the control or the intervention groups, respectively. Randomization was performed by a researcher not involved in data collection or outcome assessment. Blinding of participants and instructors, coaches and the investigation team leading the training sessions was not feasible in this type of exercise intervention. Outcome evaluators, who were not involved in intervention delivery or group allocation, were blinded to group assignment throughout the assessment time points (baseline and 12 weeks).

Statistical analysis:

Both intention-to-treat (ITT) and per-protocol (PP) approaches were applied. The ITT analysis included all randomized participants; missing post-intervention values were handled using multiple imputation. The PP analysis included only participants who attended ≥66% of the prescribed sessions. Descriptive statistics were calculated using observed data and are reported as mean ± SD. The distribution of continuous variables was inspected visually using histograms and Q-Q plots, and normality was further assessed using the Shapiro-Wilk test. Model assumptions were evaluated by visual inspection of residual plots and Q-Q plots of model residuals.

Outcome variables were analysed using linear mixed-effects models to account for repeated measurements within participants. For each outcome, Time (pre vs post), Group (intervention vs control), and the Time × Group interaction were specified as fixed effects, with a participant-specific random intercept to model within-subject clustering, according to the general form: outcome ~ Time + Group + Time × Group + (1 | participant). Fixed effects were evaluated using omnibus F-tests with Satterthwaite-approximated degrees of freedom, and model estimates are reported as regression coefficients (b) with standard errors and 95% confidence intervals. Where appropriate, post hoc pairwise comparisons of estimated marginal means were used to decompose significant effects, with multiplicity controlled using Bonferroni adjustment. Model fit was summarised using marginal and conditional R², and variance components were reported as random-intercept and residual variance. The intraclass correlation coefficient (ICC) was calculated to quantify the proportion of variance attributable to between-participant differences.

Effect sizes were expressed as partial eta squared (ηp²) for omnibus effects and Cohen's d for within- and between-group contrasts, using the standard deviation (SD), not the standard error, as the standardiser to avoid inflation (Lakens, 2013). Cohen's d was interpreted as trivial (<0.20), small (0.20-0.49), moderate (0.50-0.79), and large (≥0.80), following Cohen's conventional thresholds (Cohen, 1988). Partial eta squared was interpreted as small (ηp² ≈ 0.01), moderate (ηp² ≈ 0.06), and large (ηp² ≈ 0.14) (Richardson, 2011). These thresholds were used only as descriptive guidelines, and effects were interpreted alongside the magnitude and precision of the estimated mean differences. All tests were two-sided, with statistical significance set at α = 0.05. Statistical analyses were performed using jamovi (version 2.6.23; The jamovi project, Sydney, Australia) and R packages (R Core Team, R Foundation for Statistical Computing, Vienna, Austria) appropriate for each analysis.

To overcome the common limitation of low statistical power highlighted in previous recreational football interventions, a substantially larger sample size was deliberately recruited. This strategic design choice was undertaken not only to enhance the robustness of subgroup and ITT analyses but also to increase the precision of effect size estimates, thereby improving the generalizability of the findings.