Estimating the Standardized Mean Difference in Intervention Studies

Publisher Website
Google Scholar
Abstract
Methods for approximating confidence intervals for the population standardized mean difference, δ , were evaluated analytically in small samples. A procedure based on the large sample approximation of the distribution of the sample standardized mean difference d was quite accurate over most of the range .25 ≤ δ ≤1.5 and 6 ≤ N ≤ 40 examined. Use of Hedges’s adjustment to d yielded confidence intervals that were slightly conservative, whereas the unadjusted d produced intervals that were somewhat liberal. An empirically determined simple linear adjustment to d demonstrated the most consistent precision when used to construct confidence intervals for δ , having a maximum error of less than 2% of the nominal level of confidence over the effect size range of most interest to behavioral scientists.