Abstract
In many biological applications, the primary objective of study is to quantity the magnitude of treatment effect between two groups. Cohens'd or strictly standardized mean difference (SSMD) can be used to measure effect size however, it is sensitive to violation of assumption of normality. Here, we propose an alternative metric of standardized effect size measure to improve robustness and interpretability, based on the overlap between two sample distributions. The proposed method is a non-parametric generalized variant of SSMD (Strictly Standardized Mean Difference). We characterized proposed measure in various simulation settings to illustrate its behavior. We also investigated finite sample properties on the estimation of effect size and draw some guidelines. As a case study, we applied our measure for hit selection problem in an RNAi experiment and showed superiority of proposed method.