Cohen's d

Cohen’s d is usually computed as the difference between two means divided by a standard deviation, often the pooled within-group standard deviation. That standardization is what makes it unitless. In small samples, Cohen’s d is slightly upward-biased, which is why meta-analyses often prefer Hedges’ g, a closely related corrected version.

When the p-value says “yes” but your eyes should say “how much?”

A common research-paper trap happens right after the words statistically significant. Readers are trained to stop there, as if significance settles the story. It does not. A huge study can make a trivial difference look impressive, while a small study can miss a meaningful one. That gap is exactly why Cohen’s d effect size became so useful: it answers the question the p-value does not—how big is the difference?

Picture two choirs singing the same note. If their voices overlap almost completely, the groups are basically similar. If one choir’s sound sits clearly higher than the other, the difference is obvious even before you measure it. Cohen’s d turns that overlap into a number. Formally, it is the difference between two group means divided by their typical spread, usually a pooled standard deviation.

That is the surprise: Cohen’s d is not measuring raw units like pounds lifted, milliseconds, or blood levels. It rescales the difference into shared “group spread” units so you can judge magnitude across very different outcomes. A d of 0.5 means the two groups are separated by about half of one standard deviation. In plain English, the average person in one group is moderately shifted away from the average person in the other.

This is why Cohen’s d effect size interpretation is more portable than a raw mean difference. Ten milliseconds might be huge in sprinting and irrelevant in sleep duration. But a standardized gap lets you compare how separated the groups are relative to natural variation.

People often learn the rough guide: 0.2 small, 0.5 medium, 0.8 large. Those ranges are useful training wheels, not laws of nature. In a noisy field, 0.3 may matter. In another context, even 0.8 might not change a real decision. Still, the rule helps answer common questions: a Cohen’s d of 0.5 is usually described as a medium effect, and 1.2 is typically considered large—very large, in many practical settings. If Cohen’s d is greater than 1, the group means are more than one full typical spread apart, which usually signals a strong separation with less overlap.

You will see this idea appear in papers as Cohen’s d, standardized mean difference, or in meta-analyses as closely related versions such as Hedges’ g. If you are using a Cohen’s d calculator or running Cohen’s d in SPSS, the software is doing the same core move: mean difference divided by variability.

The one decision it helps you make

When two studies both say a supplement “worked,” do not first ask which p-value is smaller. Ask which study shows the larger, more believable effect size in a population like yours. That one move shifts you from Is there a difference? to Is the difference big enough to care about?

Cohen’s d will not tell you everything. It does not prove quality, remove bias, or replace judgment about outcomes. But it keeps you from mistaking a barely detectable nudge for a meaningful shift—and that is one of the most common reading errors in statistics.