Overview

Statistical power is the chance a study will detect an effect that really exists. This estimates power from your expected effect size and sample size, and the sample needed for 80% power.

Enter the effect size as Cohen's d, the planned sample, alpha and the number of tails. As a rough guide, 0.2 is small, 0.5 medium and 0.8 large.

Worked examples

InputsResult
Expected Effect Size (d)0.45Planned Sample Size (n)120Alpha (%)5TailsTwo-tailedApprox statistical power: 99.85%Effect size: 0.45Required n for 80% power: 39
Show 2 more examples
Expected Effect Size (d)0.2Planned Sample Size (n)100Alpha (%)5TailsTwo-tailedApprox statistical power: 51.6%Effect size: 0.2Required n for 80% power: 197
Expected Effect Size (d)0.5Planned Sample Size (n)30Alpha (%)5TailsTwo-tailedApprox statistical power: 78.19%Effect size: 0.5Required n for 80% power: 32

Approx statistical power by expected effect size

With the calculator's starting values, the approx statistical power is 99.85%. The table shows how it changes with the expected effect size.

Expected Effect Size (d)Approx statistical power
0.390.76%
0.499.23%
0.45starting value99.85%
0.599.98%
0.6100%

Other inputs: Planned Sample Size: 120 n · Alpha: 5%. Change them in the calculator above for your own numbers.

How it works

It uses a normal approximation for a one-sample or paired design: the effect is scaled by √n and compared with the critical z value for your alpha. The required sample for 80% power comes from the same relationship. For two separate groups, each group needs roughly twice the figure shown. Use dedicated software such as G*Power for final planning.

The formula

Power ≈ 1 − Φ(z_α − d × √n)
dEffect size in standard deviations
z_αCritical z for your alpha and tails

Practical tips

  • An effect of 0.45 with 120 participants gives about 99.85% power, and 39 would be enough for 80%.
  • A small effect of 0.2 needs about 197 for 80% power at 5% two-tailed.
  • Base the effect size on earlier studies or the smallest effect that would matter, not on a hopeful guess.

Frequently asked questions

Why is 80% power the usual target?

It's a common convention that balances the cost of a bigger study against the risk of missing a real effect.

What happens if my study is underpowered?

You may miss real effects, and any significant results you do find tend to overstate the effect.

Does this work for comparing two groups?

Roughly. Double the required n for each group in an independent two-group design.