What is effect size?

Magnitude, not significance

Sullivan and Feinn, writing in the Journal of Graduate Medical Education, put the case bluntly: a p value can tell a reader that an effect exists, but not how large it is, and with a big enough sample even a trivial effect becomes significant. Their example is the Physicians Health Study, where aspirin's effect on heart attack was highly significant in more than 22,000 men but tiny in size. Effect size fills that gap. It is the quantity a clinician, manager or policy maker actually uses, because decisions depend on how much benefit or harm to expect, not on whether a test crossed a line.

Raw or standardised

When the outcome has meaningful units, report the raw effect: minutes saved, points on a validated scale, kilograms, percentage points. Readers understand it directly. Standardised measures such as Cohen's d divide the difference by a standard deviation, which makes results comparable across studies that used different scales and is what meta analyses often pool. Lakens' practical primer in Frontiers in Psychology recommends reporting standardised effects for t tests and ANOVAs, and notes that Hedges' g corrects the small upward bias of d in small samples. On a poster, the raw effect usually goes first and the standardised effect beside it. If the scale is unfamiliar to the audience, one sentence translating the raw difference, such as about the gap between a mild and a moderate score, is worth more than either number alone.

Relative and absolute effects

For yes or no outcomes the choice between relative and absolute measures changes how a result feels. The Cochrane Handbook's chapter on effect measures covers risk ratios, odds ratios and risk differences and explains that relative effects tend to be more stable across populations while absolute effects convey the real world impact. A treatment that halves a risk from 2 percent to 1 percent has a risk ratio of 0.5 and a risk difference of one percentage point. Both are true, and a fair poster shows both, since a relative effect alone can make a small absolute benefit sound dramatic.

Small, medium and large are only a starting point

Cohen's conventions of 0.2, 0.5 and 0.8 for d are widely quoted, but Cohen offered them as a fallback when nothing better was known. In many fields the right benchmark is practical: the minimal clinically important difference on a pain scale, the change in test scores that shifts a grade, or the cost per outcome. A d of 0.2 applied to a whole population can matter a great deal, while a d of 0.8 in a laboratory task may mean little outside it. Say what the size means in context, not only which label it earns.

Common mistakes

Reporting an effect size without its confidence interval hides how precisely it was estimated. Calling an odds ratio a risk ratio overstates the effect when the outcome is common. Quoting only the relative effect makes modest benefits sound large. Using Cohen's labels mechanically ignores context. And comparing standardised effects across studies with very different populations can mislead, because the standard deviation in the denominator changes with how varied the sample was. A final one is choosing the effect measure after seeing which version looks most impressive; the measure should be named in the protocol or analysis plan before the data are examined.

Where it shows up in the poster builder

The key number block is the natural home for an effect size. It appears on 27 of the 40 layouts and holds a large value, a short label and a detail line meant for the interval, a p value or the sample size, so an effect such as 5.7 points lower sits above its interval. The generator writes key numbers, chart values and results from the brief, so the effect and its interval need to be stated there. The psychology example reports its anxiety difference with d = 0.52. Chart figures take categories and values without error bars, so a bar chart of group means shows the size of the difference while the interval stays in the detail line or the caption.

Questions people ask