When sufficient information about pre and posttest scores were not provided, effect sizes were calculated based on F or t values. Formulas for all analyses are available from the first author. Studies with larger samples are likely to produce more accurate results than those with smaller samples. Thus, individual effect size statistics needed to be refined by giving more weight to larger studies [Hedges and Olkin, 1985]. Further, as discussed later, studies with better design features and reporting of key variables are also likely to be more trustworthy. Based on a procedure recommended by Shadish and Haddock [1994], a weighted effect size was calculated that took into account both sample size and quality. In order to account for sample size, each d was multiplied by a weight derived from the reciprocal of the variance. The logic of weighting by the inverse of the variance is that studies with larger sample sizes are assumed to more closely approximate the true effect size for the population of all similar studies.