Table 2. Statistical Management of Data.

Analysis component	Approach or rationale
Standardized mean difference [33,34,49]	Post-intervention difference between treatment and control group divided by the pooled SD, for two-group comparisons. Difference between pre- and post-test scores divided by pre-intervention SD for single-group comparisons, under assumptions of no (ρ12 = 0.0) and high association (ρ12 = 0.80) between pre- and post-scores; calculated for all pre-post treatment and control groups with adequate data. Adjusted for small sample bias. Weighted by inverse of variance to address sample size differences. 95% confidence intervals for mean ES constructed from normal-theory stand errors. Studies with multiple treatment groups without a control group were treated as single-group studies.
Dependencies in data for few studies with one control group and two or three treatment groups	Each study's dependent ESs combined in a single independent ES by generalized least-squares approach and then treated as standard univariate random-effects analyses. [50]
Outlier determination	ESs examined graphically. Externally standardized residuals evaluated as removed each ES one at a time. Homogeneity analyses after omitting each case in turn.
Publication bias determination [51]	ES plotted against sampling variance. Examined for funnel-shaped plot indicating symmetrical distribution around estimated population mean ES.
Homogeneity	Q calculated from weighted sum of squares (chi-squared distribution).
Random-effects model	Assumes individual ESs vary due to both subject-level sampling error and other sources of study-level error such as variations among interventions. Consistent with heterogeneous study implementation. Weighted method of moments used to estimate between studies variance component.
Common Language Effect Size (CLES) [52]	Probability that a random treatment subject would score higher than a random control subject (an ES of d = 0.0 corresponds to a CLES of 0.50).
Conversion to original metric [47]	ES converted to original metric for variables with multiple studies reporting identical measures. Reported for minutes of PA per week and for steps per day.
Moderator analysis [53,54]	Mixed- and fixed-effects calculated, fixed-effects available from senior author. Continuous moderators: effects tested by unstandardized regression slope ( $\widehat{\beta }$) in meta-analytic analogue of regression. Dichotomous moderators: effects tested by between-group heterogeneity statistic (Qbetween) using meta-analytic analogue of ANOVA. Interpret findings cautiously when significant heterogeneity exists, large variance components decrease statistical power.
Multiple-df moderator analysis	Categorical moderator with more than two levels or two dichotomous moderators in factorial design. Multicategory moderator: Omnibus test of heterogeneity among levels (Qbetween) followed by focus contrasts among particular levels. Two dichotomous moderators: Omnibus test comparing mean ES among four cells (Qbetween), tests of each main effect and interaction, and contrasts comparing particular cells (e.g., simple main effects).