The within-group variances were assumed known. Observations were weighted by the inverse of the sampling variance [51].

An intercept-only model was created, estimating the weighted mean ES across all studies and treatment groups. Second, a basic model was created which only included the class of the group (treatment or control) as a predictor. A full model was then created with the following predictors: the class of the group (treatment or control), whether or not the groups were protein matched, training status (experienced or novice), blinding (DNA Damage inhibitor double, single, or none), gender (male, female, or mixed), age (young or old), body mass in kg, and the duration of the study in weeks. The

full model was then reduced by removing one VS-4718 purchase predictor at a time, starting with the most insignificant predictor [54]. The final model represented the reduced model with the lowest Akaike’s Information Corrected buy AUY-922 Criterion (AICC) [55] and that was not significantly different (P > 0.05) from the full model when compared using a likelihood ratio test (LRT). Model parameters were estimated by the method of restricted maximum likelihood (REML) [56]; an exception was during the model reduction process, in which parameters were estimated by the method of maximum likelihood (ML), as LRTs cannot be used to compare nested models with REML estimates. Denominator df for statistical tests and CIs were calculated according to Berkey et al. [57]. The treatment/control classification variable was not removed during the model reduction process. Separate analyses were performed for strength and hypertrophy. ESs for both changes in cross-sectional area (CSA) and FFM were pooled in the hypertrophy analysis. However, because resistance exercise is associated

with the accretion of non-muscle tissue, separate sub-analyses on CSA and FFM were performed. Because the effect of protein timing might interact with whether the treatment and control groups were matched for total protein intake, an additional model was created that included an interaction term between the treatment/control classification variable and the protein match variable. Also, because the effect of protein timing Phosphoglycerate kinase might vary by training experience, a model was created that included an interaction term between the treatment/control classification variable and the training status variable. Adjustment for post hoc multiple comparisons was performed using a simulation-based procedure [58]. All analyses were performed using SAS Enterprise Guide Version 4.2 (Cary, NC). Effects were considered significant at P ≤ 0.05. Data are reported as means (±SEs) and 95% CIs. Results Study characteristics The strength analysis comprised 478 subjects and 96 ESs, nested within 41 treatment or control groups and 20 studies.