Calculate Cohen's d, Hedges' g, and Glass's Δ effect sizes with CLES, overlap metrics, confidence intervals, and benchmark interpretation tables.
Cohen's d is the most widely used measure of effect size for comparing two group means. It expresses the difference as a number of standard deviations, making it possible to compare effects across different studies, scales, and contexts. A d of 0.5 means the groups differ by half a standard deviation — regardless of whether you're measuring test scores, blood pressure, or reaction times.
This calculator computes four effect size variants: Cohen's d (pooled SD), Hedges' g (bias-corrected for small samples), Glass's Δ (uses control group SD only), and a control-SD variant. It also converts to point-biserial r and eta-squared, computes the Common Language Effect Size (CLES — the probability that a random person from Group 2 outscores a random person from Group 1), distribution overlap, and U₃ statistics.
Effect sizes are essential for meta-analyses, power analyses, and for determining whether a statistically significant result is practically meaningful. A p-value tells you whether an effect exists; Cohen's d tells you how big it is.
Statistical significance (p-values) doesn't tell you the size of an effect — it only tells you whether it's distinguishable from zero. Cohen's d fills this gap by quantifying how large the difference is in standardized units. This is critical for power analyses (planning study sample sizes), meta-analyses (combining results across studies), and practical decision-making.
This calculator provides everything you need: the d value with confidence interval, alternative variants for different assumptions, conversions to other effect size metrics, and intuitive interpretations like CLES and overlap. It's essential for researchers, students, and anyone conducting or reading quantitative research.
Cohen's d = (M₂ − M₁) / S_pooled S_pooled = √[((n₁−1)S₁² + (n₂−1)S₂²) / (n₁ + n₂ − 2)] Hedges' g = d × (1 − 3/(4(n₁+n₂−2) − 1)) Glass's Δ = (M₂ − M₁) / S₁ CLES = Φ(d / √2) Overlap = 2Φ(−|d|/2) r = d / √(d² + 4)
Result: d = 0.50 (Medium effect)
Groups differ by 5 points with pooled SD = 10, giving d = 5/10 = 0.50. This is a medium effect by Cohen's benchmarks, with CLES = 63.8% (a randomly chosen Group 2 member outscores a Group 1 member 64% of the time).
Jacob Cohen introduced his d statistic in 1962 and elaborated it in his landmark 1988 book "Statistical Power Analysis for the Behavioral Sciences." He proposed the small/medium/large benchmarks as rough guides when prior research was unavailable, emphasizing that discipline-specific norms should take precedence.
The distinction between statistical significance and practical significance became a major theme in the "replication crisis" of the 2010s. Many statistically significant findings had negligible effect sizes, calling into question their practical importance. Today, major journals require effect size reporting alongside p-values.
**Cohen's d** uses the pooled standard deviation and assumes equal variances. It's the default for independent-samples t-tests. **Hedges' g** applies a small-sample correction factor of approximately 1 − 3/(4df − 1), which matters for n₁ + n₂ < 50. **Glass's Δ** uses only the control group's SD, appropriate when the treatment changes variability (e.g., an educational intervention that affects not just the mean but the spread of scores).
Meta-analysis combines effect sizes from multiple studies using weighted averaging. Hedges' g is preferred because its correction removes the small-sample bias that would otherwise inflate the combined estimate. The confidence interval for d is also crucial — studies with wider CIs receive less weight in random-effects meta-analyses. This calculator's CI output is directly usable for meta-analytic input.
Use Hedges' g when either sample size is small (n < 20). Cohen's d has a slight upward bias for small samples, and the Hedges correction factor adjusts for this. For larger samples, d and g are virtually identical.
Cohen suggested d = 0.2 (small), 0.5 (medium), and 0.8 (large) as rough benchmarks. However, these are context-dependent — in education, d = 0.4 may be a meaningful intervention effect, while in pharmacology, d = 0.2 might be clinically important.
CLES translates d into the probability that a randomly selected person from Group 2 outperforms a randomly selected person from Group 1. For d = 0.5, CLES ≈ 63.8%. This is more intuitive for non-statisticians.
Overlap measures how much the two distributions intersect. With d = 0, overlap is 100% (identical distributions). With d = 0.8, overlap drops to about 69%. Lower overlap means the groups are more distinct.
Yes. The sign indicates direction: positive d means Group 2 is higher than Group 1. By convention, label the groups so that d is positive (treatment mean higher), but the magnitude |d| determines the effect size.
r = d / √(d² + 4) and η² = d² / (d² + 4). A medium d (0.5) corresponds to r ≈ 0.24 and η² ≈ 0.06. The conversions assume equal sample sizes; adjust for unequal n.