Cohen's d Calculator

What Is Cohen's d?

Cohen's d is a standardized effect size that measures the difference between two means in standard deviation units:

$d = \frac{M_1 - M_2}{SD_{pooled}}$

If d = 0.5, the means differ by half a standard deviation—a "medium" effect by conventional standards.

Unlike raw mean differences, Cohen's d enables comparison across studies using different scales. A 10-point difference on a 100-point scale means something different than on a 1000-point scale. Cohen's d removes this dependency.

Cohen's Benchmarks

Jacob Cohen (1988) proposed interpretive benchmarks:

These are guidelines, not rigid cutoffs. Context matters—a "small" effect can be hugely important if the intervention is cheap or the outcome is crucial.

The Formula

Cohen's d for two independent groups:

$d = \frac{\bar{x}_1 - \bar{x}_2}{S_p}$

Where $S_p$ is the pooled standard deviation.

For equal sample sizes:

$S_p = \sqrt{\frac{s_1^2 + s_2^2}{2}}$

For unequal sample sizes:

$S_p = \sqrt{\frac{(n_1-1)s_1^2 + (n_2-1)s_2^2}{n_1 + n_2 - 2}}$

Interpretation

Visualizing d

Cohen's d relates to overlap between distributions:

With d = 0.8, the average person in the treatment group scores higher than 79% of the control group.

Practical Examples

d = 0.2 (small): Difference between heights of 15-year-olds and 16-year-olds

d = 0.5 (medium): Height difference between 14-year-olds and 18-year-olds

d = 0.8 (large): Height difference between 13-year-olds and 18-year-olds

Applications

Meta-Analysis

Combining results across studies requires a common metric. Cohen's d enables:

• Comparing effect sizes across studies
• Calculating weighted averages
• Assessing heterogeneity

Power Analysis

Sample size calculations use expected effect size:

• Small effects need larger samples
• Large effects are detectable with smaller samples

For 80% power to detect d = 0.5 at α = 0.05, you need about 64 per group.

Research Reporting

Journals increasingly require effect sizes:

"The treatment group outperformed controls, d = 0.62, 95% CI [0.38, 0.86]"

This tells readers both the magnitude and precision.

Educational Interventions

Comparing teaching methods:

"Method A showed d = 0.4 improvement over traditional instruction"

Clinical Psychology

Treatment effectiveness:

"Therapy produced d = 0.8 reduction in depression scores"

Why Not Just Report Mean Differences?

Problem 1: Scale dependence

Study A: 5-point improvement on a 10-point scale (50% of scale) Study B: 5-point improvement on a 100-point scale (5% of scale)

Same raw difference, vastly different meanings. Cohen's d standardizes.

Problem 2: Variability matters

5-point difference where SD = 2: d = 2.5 (huge effect) 5-point difference where SD = 20: d = 0.25 (small effect)

High variability means the difference is less distinguishable.

Variants of Cohen's d

Glass's Δ (delta)

Uses only the control group SD:

$\Delta = \frac{\bar{x}_{treatment} - \bar{x}_{control}}{s_{control}}$

Appropriate when treatment might change variability.

Hedges' g

Corrects for small-sample bias:

$g = d \times \left(1 - \frac{3}{4(n_1+n_2)-9}\right)$

Nearly identical to d for large samples, but slightly smaller for small n.

Cohen's d for Paired Data

For pre-post or matched designs:

d = M_difference / SD_difference

Or: d = M_difference / SD_pooled (using both time points' SDs)

Confidence Intervals for d

Effect size estimates have uncertainty. The approximate SE:

$SE_d \approx \sqrt{\frac{n_1+n_2}{n_1 \times n_2} + \frac{d^2}{2(n_1+n_2)}}$

$95\% \text{ CI} = d \pm 1.96 \times SE_d$

This CI tells you the range of plausible population effect sizes.

Historical Context

Jacob Cohen (1923-1998) introduced the d statistic in his 1962 paper and expanded it in his 1969 and 1988 books "Statistical Power Analysis for the Behavioral Sciences."

Cohen was motivated by the over-reliance on p-values, which conflate effect size with sample size. His work established that statistical significance doesn't equal practical importance—a crucial distinction.

His benchmarks (0.2, 0.5, 0.8) were based on surveys of psychological research literature, representing what he observed as typical small, medium, and large effects.

Limitations

Benchmark Dependence on Field

Cohen's benchmarks came from psychology. In other fields:

• Medical interventions might consider d = 0.2 clinically significant
• Educational research might find d = 0.4 practically important
• Physics might require d > 3 for significance

Always interpret in context.

Sensitivity to Outliers

The pooled SD is affected by outliers, potentially distorting d. Consider:

• Trimmed statistics
• Robust effect sizes
• Examining data quality

Assumption of Equal SDs

Cohen's d assumes similar variability in both groups. When variances differ substantially, Glass's Δ may be more appropriate.

Related Calculations

• Pooled Standard Deviation Calculator — Denominator of d
• CI for Difference in Means — Raw difference with CI
• Mean Calculator — Group means
• Standard Deviation Calculator — Group SDs
• Odds Ratio Calculator — Effect size for categorical outcomes

Summary

Cohen's d standardizes mean differences into a common scale—standard deviation units—enabling comparison across studies and measurement instruments. The conventional benchmarks of 0.2 (small), 0.5 (medium), and 0.8 (large) provide interpretive guidance, though context always matters. Essential for meta-analysis, power analysis, and complete research reporting, Cohen's d answers a question p-values cannot: "How big is the effect?" Report effect sizes alongside significance tests to give readers a full picture of your findings.