MerQuR+Statistics Glossary

Statistics Glossary → Glass's Δ

Glass's Δconcept

A standardized effect size that divides the difference between two group means by the standard deviation of the control group only.

Cohen's d divides the difference by the pooled standard deviation of both groups and assumes the two groups are equally spread. If an intervention changes the spread as well as the mean, the pooled standard deviation represents neither the control nor the treatment group. Glass's Δ then uses only the control group's standard deviation as the standardizer: Δ = (mean_treatment - mean_control) / SD_control. The effect is thus expressed in units of the 'natural' spread that the intervention has not touched.

Glass's Δ is read much like Cohen's d (for example, Δ = 0.5 means the treatment mean lies half a standard deviation above the centre of the control distribution). Its drawbacks are that it is less precise, because it relies on one group's standard deviation, and that it depends on which group is labelled 'control'; swapping the groups changes its value. When there is no clear control group, d_av, which averages the two variances, or Welch-type approaches may be more suitable.

An everyday example

A school runs a new reading programme. In the control class, reading speed averages 100 words per minute with a standard deviation of 15; in the programme class the average is 115, but because some pupils speed up a great deal, the standard deviation rises to 30. Glass's Δ = (115 - 100) / 15 = 1.0: the programme class's average sits one standard deviation above the centre of a typical class.

In MerQur

In the Independent t-Test, the 'Glass Δ' checkbox divides the difference by the SD of the second group (taken as the control). In the Effect Size Calculator, the 'Standardizer (extra)' option gives two Glass Δ values for Cohen's d, taking each group in turn as the control.

Analyses where this concept appears

Independent-Samples t-TestEffect Size Calculator

Related concepts

Cohen's d (effect size)Hedges' g