When possible, we used mixed-effects models to independently test the effects of between-study moderators (i.e., gender composition). Within each level of the moderator, the mixed effect model operates under the same assumptions as does the random-effects model. In particular, at each level of the moderator, the model assumes there is random variation in the distribution of effect sizes beyond sampling error. In comparing effects across levels of the moderator, however, the mixed-effects model assumes that the moderator is associated with systematic differences (Borenstein et al., 2009). We additionally retained all previously described model specifications for moderator analyses, including averaging across multiple outcome measures or assessment points within samples to produce single effect sizes for each sample.
