Although simple averaging implies the risk of overestimating the aggregated
effect by underestimating the variance among measures, and remedies for this
problem do exist (e.g., Gleser & Olkin, 1994; Rosenthal & Rubin, 1986), these
remedies require correlational information which may be neither reported nor
directly estimable for meta-analysis (excepting cases where raw data are
available). Also, while statistically meaningful, the difference of averaging
effects and computing a latent effect from multiple measures may be small
(see Appendix C for a worked example). For this study we judged such
averaging to permit a reasonable approximation of the true score effect,
capitalizing on the unit-free nature of the standardized mean difference
statistic (i.e., g ).