Bias correction for Cohen's d

Abstract

Cohen's d - a common effect size - contains a positive bias. The traditional bias correction, based on strict distribution assumption, does not always work for a small study with limited data. The non-parametric bootstrapping is not limited by distribution assumption and can be used to remove the bias in Cohen's d. A real example is included to illustrate the implementation of bootstrap bias estimation and the removal of sizable bias in Cohen's d.

Keywords: Bias; Cohen’s d; Effect size; bootstrap.