Calculating effect size for continuous variables: D537 – a robust, quantile-based approach

The calculation of effect size is an important step in measuring the potential real-life significance of the effect of an intervention. In the case of continuous data, Cohen's d is frequently used. This scales the difference between the means of two groups, or the mean difference between pairs of measurements, by dividing by the standard deviation. However, outlying values, especially in small studies, can influence the size of d. This article presents D537, a robust formula for d that is based on rank statistics. The median is used as a measure of difference, while the scaling factor is the range between the 30th and 70th percentiles of the distribution; a range that is equal to one standard deviation when the data are normally distributed. When data are normally distributed, the value of D537 is equal to that of Cohen's d. As D537 is based on the 30th, 50th and 70th percentiles, it is robust to outliers.