Inverse variance weighted average effect sizes (M r) were estimated post test and at follow-up using random-effects models. Pearson’s r values were converted to z scores for meta-analysis. Effect sizes were interpreted using Cohen’s criteria for small (r=0.10), medium (r=0.30) and large (r=0.50) effects.
Heterogeneity investigations were conducted using inverse variance weighted random-effects models with results reported as Q statistics (not further defined). Univariate analyses were conducted and correlations between moderators were tested for. Continuous variables were standardised in a z score format; linear and quadratic effects were tested for to decrease the risk of model misspecification. Average length of follow-up was included in models for follow-up effect sizes when this factor produced a significant effect.
Fifteen potential moderating factors were examined relating to participant features (risk status selective or universal, gender, ethnicity and mean age), intervention features (intervention content, duration and inclusion of homework), provider features (professional interventionist versus endogenous provider such as a nurse or teacher) and design features (use of random assignment, outcomes assessed by interview versus self-reported, peer review publication, same unit of analysis and unit of randomisation and length of follow-up).
Sensitivity analyses were conducted including only one effect size from trials that reported multiple effect sizes (where more than one depression prevention programme was evaluated per trial).