An effect size calculator finds Cohen's d, the standardized difference between the means of two groups. Enter the mean and standard deviation of each group, and optionally the group sizes, and you get d, the pooled standard deviation, a description of the size of the effect and Hedges' g, a small-sample correction of d. It shows how big a difference is in a way that does not depend on the units or on the sample size. It runs in your browser.
How to use the effect size calculator
- Enter the mean and standard deviation of group 1 and of group 2.
- Enter the group sizes if you know them.
- Press Calculate.
Cohen's d
The Wikipedia article on effect size defines Cohen's d as the difference between the two means divided by a standard deviation: d = (x̄₁ − x̄₂) ÷ s. For two independent samples, Cohen's definition of the pooled standard deviation is s = √(((n₁ − 1) s₁² + (n₂ − 1) s₂²) ÷ (n₁ + n₂ − 2)).[1] The page adds that a version with n₁ + n₂ in the denominator is known as the maximum-likelihood form, which this calculator does not use. With the means 10 and 8, the standard deviations 2 and 3 and 30 people in each group, the pooled value is √6.5 = 2.55, and d = 2 ÷ 2.55 = 0.78. If you leave the sizes empty, the calculator assumes equal groups, for which the pooled value is the square root of the average of the two variances, and it gives the same result here.
How big is big
The article has a table of guidelines for d. Jacob Cohen suggested 0.2 for a small effect, 0.5 for a medium effect and 0.8 for a large one. Sawilowsky extended the list with 0.01 as very small, 1.2 as very large and 2.0 as huge.[1] Funder and Ozer proposed other values based on benchmarks from research, such as 0.41 for medium and 0.63 for large.[1] The calculator uses the first two sets to name the size, so that a d of 0.78 is medium and 1.3 is very large. A negative d means that group 2 has the higher mean, and the size is judged by its absolute value.
Cautions about the guidelines
The page quotes Cohen as saying that small, medium and large are relative, not only to each other but to the area of behavioural science in which they are used. A report from the US Department of Education warns that using the guidelines without thought can be inappropriate and misleading, and it suggests using norms from comparable studies. Funder and Ozer advise judging an effect by benchmarks and by its consequences.[1] A small effect can matter a great deal if it applies to millions of people, or if it concerns a matter of life and death. A large one may be trivial in practice.
Why use an effect size
A p-value tells you whether a difference is likely to be due to chance, but it grows with the sample size, so a tiny difference in a huge study can be highly significant. The page notes that d, unlike the t statistic, is not affected by the sample size.[1] It therefore describes how large a difference is. Reporting both is good practice, and many journals ask for an effect size with each test.
Hedges' g
Cohen's d overstates the effect a little in small samples. Hedges' g reduces it by a correction factor, which is approximately 1 − 3 ÷ (4(n₁ + n₂) − 9). The calculator uses this common approximation. The part of the page that deals with Hedges' g was not available to us in full, so the correction is stated as an approximation, and the exact form uses gamma functions. For the example above, g is 0.77, a little below d, and the two get closer as the groups grow. With 100 people in each group the correction is only about 0.4 percent.
Planning a study
Before a study, researchers choose the smallest effect that would matter, and use it to find how many people they need. A larger effect can be found with fewer people. If you expect a medium effect of 0.5, you need far more participants than for a large effect of 0.8, to have the same power. The calculator does not do a power analysis, but it shows the effect size that you can use for one.
Reporting the result
When you write up a study, give the means, the standard deviations and the sizes of the groups, then the value of d with a word for its size and the test result. For example: the treatment group (M = 10, SD = 2, n = 30) scored higher than the control group (M = 8, SD = 3, n = 30), with a medium effect, d = 0.78. A confidence interval for d is even better, but it needs more than this calculator gives. Say which guideline you used for the words.
Common mistakes
- Using the standard error. Enter the standard deviation of each group.
- Using it for paired data. Paired designs use the standard deviation of the differences.
- Treating the labels as rules. They depend on the field.
- Forgetting the sign. A negative d only means that group 2 is higher.
What next
Use this often? Add Effect Size to your Home screen.
Frequently asked questions
What is an effect size calculator?
A tool that finds the standardized difference between two group means, such as Cohen's d, and describes how large it is.
What is the formula for Cohen's d?
d = (mean 1 − mean 2) ÷ the pooled standard deviation, where the pooled value is √(((n₁ − 1)s₁² + (n₂ − 1)s₂²) ÷ (n₁ + n₂ − 2)).
What counts as a small, medium and large effect?
Cohen suggested 0.2, 0.5 and 0.8. Sawilowsky added 0.01 as very small, 1.2 as very large and 2.0 as huge.
What is Hedges' g?
A version of Cohen's d with a correction for small samples, approximately d × (1 − 3 ÷ (4(n₁ + n₂) − 9)).
Does the sample size change Cohen's d?
Not systematically. Unlike the t statistic, d is not affected by the sample size.
Sources
Was this tool helpful?
What went wrong?
We only receive the reason you pick, never what you typed into the tool.
Thanks for your feedback.