Question 1

What is effect size in plain terms?

Accepted Answer

It's a number that tells you how big a difference or association is, regardless of sample size. P‑values answer 'is there an effect?'; effect size answers 'how large is it?'

Question 2

What is Cohen's d and how is it computed?

Accepted Answer

Cohen's d is the standardized difference between two means: d = (mean₂ − mean₁) / sₚ, where sₚ is the pooled standard deviation sₚ = sqrt(((n₁−1)s₁² + (n₂−1)s₂²) / (n₁ + n₂ − 2)).

Question 3

How are odds ratios and their confidence intervals computed?

Accepted Answer

We compute OR = (a×d)/(b×c). The 95% CI uses the log‑scale Wald method: ln(OR) ± 1.96 × sqrt(1/a + 1/b + 1/c + 1/d), then we exponentiate the limits.

Question 4

Does Cohen's d assume equal variances?

Accepted Answer

Yes. Using the pooled SD assumes equal population variances across groups. If variances differ markedly, consider alternatives like Glass's Δ (uses the control SD) or Welch‑based approaches.

Question 5

What is Hedges' g and when should I use it?

Accepted Answer

Hedges' g applies a small‑sample correction to Cohen's d: g = J·d with J ≈ 1 − 3/(4(n₁ + n₂ − 2) − 1). For small samples, g is slightly less biased than d.

Question 6

How do I interpret Cohen's d values?

Accepted Answer

Common guidelines: |d| ≈ 0.2 (small/subtle), 0.5 (medium/noticeable), 0.8 (large/substantial). These are rough conventions—what counts as meaningful depends on your field and context.

Question 7

How do I interpret odds ratio values?

Accepted Answer

OR = 1 means no difference. OR = 2 means twice the odds in one group. OR = 0.5 means half the odds. Values further from 1 indicate stronger associations.

Question 8

Are odds ratios the same as risk ratios?

Accepted Answer

No. They are similar when outcomes are rare, but can diverge for common outcomes. Risk ratio compares probabilities; odds ratio compares odds.

Question 9

When should I use Cohen's d vs. Odds Ratio?

Accepted Answer

Use Cohen's d for continuous outcomes comparing two means (e.g., time, revenue). Use Odds Ratio for binary outcomes in 2×2 tables (e.g., success/failure, case/control).

Question 10

Any best practices for reporting?

Accepted Answer

Always report effect sizes with confidence intervals and context. A tiny but significant effect may be unimportant; a moderate effect with wide CI may be uncertain.

Effect Size Calculator

Group 1

Group 2

How to interpret d

Common questions