What it does
A p-value tells you whether a difference is likely to be real; an effect size tells you how big it is, in a way you can compare across studies. Journals in education and psychology ask for both. This calculator covers the effect sizes teachers and education researchers use most:
- Two groups (for example, two classes taught differently): Cohen’s d, Hedges’ g (d corrected for small samples) and Glass’s delta (the difference divided by the control group’s SD).
- Before and after on the same people: dz (the mean change divided by the SD of the changes) and dav (the mean change divided by the average of the two SDs), plus their small-sample corrected versions.
- Learning gain: Hake’s normalised gain, both the class-average gain ⟨g⟩ and the average of each student’s own gain.
Confidence intervals for d, g and dz are exact, found from the noncentral t distribution; they match Python’s scipy to many decimal places. Nothing you type is sent anywhere.
It works well with the sample size and power calculator, the reliability and agreement and the uncertainty propagation calculator.
How to use it
- Choose Two groups, Before and after or Learning gain.
- Type the means, standard deviations and sizes from your report or spreadsheet, or switch to Raw scores and paste two columns.
- For before-and-after summaries, give the correlation between the two measurements, or the SD of each person’s change (it is then used instead).
- For learning gain, set the maximum score of the test.
- Pick a 90, 95 or 99 % confidence level, and download the results as CSV, image, PDF or Word.
Worked example
Two groups (made up): a new method gives a mean of 78.4 (SD 9.6, n 24), the usual one 72.1 (SD 10.8, n 26). Cohen’s d is 0.615, 95 % interval 0.044 to 1.180, and Hedges’ g is 0.605 (0.043 to 1.162): a medium effect, but the interval is wide. The t test gives t(48) = 2.173, p = 0.035. Glass’s delta is 0.583.
Before and after: 20 students go from 61.2 (SD 12.4) to 68.9 (SD 11.7), with r = 0.72 between the tests. dz is 0.852 (0.329 to 1.358) and dav is 0.639.
Learning gain: 10 students on a 25-mark test have a pre-test mean of 13.7 and a post-test mean of 16.7, so ⟨g⟩ = (16.7 − 13.7) / (25 − 13.7) = 0.265, a low gain. The average of individual gains is 0.279.
Which one should I report?
- For two groups, report Hedges’ g with its interval. Use Glass’s delta when the treatment might change the spread of scores.
- For before-and-after, dav is comparable with two-group studies; dz goes with the paired t test and is usually larger when the two measurements are strongly correlated.
- Hake’s gain is common in physics and science education with concept inventories. Report which of the two averages you used.
- Cohen’s 0.2, 0.5 and 0.8 are rough labels. In education, an effect of 0.2 can matter a lot.
Frequently asked questions
Why are some intervals marked approximate? Glass’s delta and dav have no exact interval; they use standard normal approximations.
What about very large studies? When the noncentrality is very large (above 36), the calculator uses a normal approximation to the noncentral t, which is accurate to about 0.01 in probability.
Is my data sent anywhere? No. Everything runs in your browser and the page keeps working offline.
Where the numbers come from
- Formulas: Cohen (1988); Hedges (1981) small-sample correction and Hedges and Olkin (1985); Glass (1976); Lakens (2013) dz and dav; Borenstein et al. (2009) repeated-measures variance; noncentral t intervals (Steiger and Fouladi 1997) with Lenth's (1989) algorithm; Hake (1998) normalised gain.
- Example numbers: made up for this page.
Standard scientific values, such as constants, atomic weights and melting points, are facts found in every textbook; we name where ours were checked. If you are an author or publisher and want a value, a credit or a table changed or removed, tell us and we will act quickly.
Last updated 2026-11-09.