How to Calculate Effect Size

Effect size measures how large a difference or relationship actually is — distinct from statistical significance, which only tells you whether an effect is likely to be real. A tiny, practically meaningless effect can still be "statistically significant" with a large enough sample.

Continuous outcomes

Mean difference (MD) — the raw difference between group means, useful when all studies use the same measurement scale.

Standardized mean difference (SMD) / Cohen's d — the mean difference divided by the pooled standard deviation, used when studies measure the same construct with different scales. Rough interpretation: 0.2 = small, 0.5 = medium, 0.8 = large — though this varies by field and should not be applied rigidly.

Hedges' g — a small-sample-corrected version of Cohen's d, generally preferable when individual study sample sizes are small.

Binary outcomes

Risk ratio (RR) — the ratio of event probability between two groups; intuitive to interpret (RR of 1.5 = 50% higher risk).

Odds ratio (OR) — the ratio of event odds between groups; common in case-control studies and logistic regression output, but less intuitive than RR and easily over-interpreted as if it were a risk ratio when event rates are high.

Hazard ratio (HR) — used in survival/time-to-event analysis, representing the instantaneous risk ratio over the follow-up period.

A common mistake

Mixing effect size types within the same meta-analysis (e.g., pooling odds ratios and risk ratios as if equivalent) without converting them to a common metric first. Most meta-analysis software includes conversion formulas — use them rather than approximating by eye.