B2001
Title: Small-sample test for comparing Cohen's d effect sizes between groups
Authors: Florin Vaida - University of California San Diego (United States) [presenting]
Anya Umlauf - University of California San Diego (United States)
Abstract: The fosuc is on comparing treatment effects between two or more different studies, or populations. Two instances when this question arises are meta-analysis, when determining whether treatment effects are homogeneous or heterogeneous over the studies; and neurocognitive testing, when comparing group effects (e.g., the effect of sex or race) between two different populations. We assume that for each study, the treatment effect is measured by Cohen's d, which is the difference between the means in the treatment and control groups relative to the common standard deviation. An approximately unbiased estimator of this effect is given by Hedges's g. When comparing two studies, a large-sample approximation to the distribution of the difference of Hedges's g, g1-g2, is currently used. We show that a finite-sample approximation for the distribution of this statistic, suitably modified, can be obtained. This is done using second-order expansions. The improvement of the proposed method is demonstrated via statistical simulations. The use is illustrated in an application from neurocognitive research in HIV.
