Effect size
The smallest difference or relationship that would matter in the real world.
Free study-planning tool
Plan a study that can
answer the question.
Explore how sample size, effect size, and uncertainty work together—before you collect the data.
Private by default. Your inputs stay in this browser.
01 / Study design
Power planner
Solve for
Cohen's d
128 total
80–90% is common
02 / Your result
64per group
You need 128 participants total to have an 80% chance of detecting a medium-sized difference.
Well powered for this effect
Power curveHow recruitment changes your chance of detecting the effect
80%The honest part: This result is only as credible as your expected effect. Use prior evidence or the smallest effect worth detecting—not a convenient guess.
Common questions
Before you trust the number.
What power should I use?
80% is a common minimum; 90% may be more appropriate when missing a meaningful effect would be especially costly. The choice should be justified for the study, not selected automatically.
How should I choose an effect size?
Use prior evidence or the smallest effect that would matter in practice. Avoid choosing an effect merely because it produces a feasible sample size. Read the guide to choosing an effect size →
Can I calculate power after collecting the data?
Observed post-hoc power adds little beyond the p-value. After data collection, confidence intervals and the range of effects compatible with the data are usually more informative.
When do I need a custom analysis?
Use a design-specific analysis for clustering, repeated measures, unequal allocation, attrition, multiple primary outcomes, non-inferiority, or other structures not represented by this calculator.
Power, plainly
Four numbers.
Four numbers.
One design decision.
Power analysis connects the signal you care about to the evidence your study can realistically produce.
Sample size
How many independent observations your design needs—not simply how many are available.
Alpha
Your tolerated false-positive rate. Lower alpha demands stronger evidence.
Power
Your chance of detecting the effect if it truly exists. It equals 1 − β.
Built for defensible planning
A calculator should show its work.
StatPower computes t-test power with the exact noncentral t distribution, the method used by R and G*Power, and uses Fisher’s z for correlations and Cohen’s arcsine approximation for proportions. t-test and proportion results match R’s pwr package exactly. It is most useful before data collection, paired with a justified effect size and an analysis plan that matches your design.
- ✓ Inputs never leave your browser
- ✓ Reporting sentence included
- ✓ Clear assumptions and limitations
Complex design?
The right calculation starts with the right model.
Clusters, repeated measures, attrition, multiple outcomes, or uncertain effect sizes can change the answer substantially. Get a design review from DASS before the study is locked.
Preparing a grant? DASS can join the proposal team as a statistical methodologist for study design, power analysis, and analysis planning. When sponsor and institutional rules allow, our fees can be included in the proposal budget.
Discuss your grant