MerQuR+Statistics Glossary

Statistics Glossary → Common-language effect size (CLES)

Common-language effect size (CLES)concept

An effect size expressed as the probability that a randomly chosen person from one group has a higher value than a randomly chosen person from the other group.

It was proposed by McGraw and Wong (1992) to make standard-deviation-based measures such as Cohen's d understandable to readers without statistical training. If both groups are normal with equal variances, CLES = Φ(d/√2), where Φ is the standard normal cumulative distribution function. When d = 0, CLES = 0.50 (which group comes out ahead is a coin toss); when d = 0.8, it is about 0.71.

It can also be computed without distributional assumptions by comparing all pairs directly: P(X > Y) + ½·P(X = Y). This version equals the Mann-Whitney U statistic divided by n₁n₂ and is also known as the probability of superiority or the AUC; it is related to the rank-biserial correlation by r = 2·CLES − 1. CLES tells how 'often' one group comes out ahead, not how 'large' the difference is.

An everyday example

People who joined a diet programme are compared with people who did not in terms of weight lost over six months, and CLES = 0.68. This means that if one participant and one non-participant are picked at random, there is a 68% chance that the participant lost more weight.

In MerQur

In the independent-samples t-test, the 'CLES' checkbox gives Φ(d/√2), assuming normality and equal variances. In the Mann-Whitney U test, the 'Effect sizes: CLES' option computes the same measure without distributional assumptions as P(X > Y) + ½P(X = Y), where X is the first group.

Analyses where this concept appears

Independent-Samples t-TestMann-Whitney U

Related concepts

Cohen's d (effect size)Rank-biserial correlation