Plan a study.
Select a test and what to calculate. Results update automatically. Enter the active inputs; calculated values are shaded.
Your study plan
How sample size changes power
Read the curve as a table
Check the assumptions behind the number.
What if the effect is smaller?
Explain your sample size.
You can edit this paragraph before copying or exporting it.
Changing a planning input regenerates the paragraph. Exporting preserves your edits.
See what power means.
Repeat the same study under a specified effect. Count how often it rejects the null.
Power is a repeated-study probability.
Run the experiment to compare the simulated rejection rate with calculated power.
Each mark is a complete study.
The first 100 simulated studies appear below. Filled marks reject the null; outlined marks do not.
Where the test makes its decision.
The curves show test statistics across repeated studies. Vertical lines mark the rejection boundaries.
Predict it before changing it.
If the true effect stays the same and you recruit more participants, what happens to power?
Three things power does not tell you
- It does not give the probability that a hypothesis is true.
- It does not guarantee a significant result in one study.
- It does not make a nonsignificant result evidence of equivalence.
Calculate and compare effect sizes.
Choose your comparison and enter summary statistics. Results update as you type.
Read the estimate in context.
These are point estimates, without confidence intervals. Positive mean effects describe B minus A, or after minus before. A positive change can represent improvement or decline, depending on the outcome. Ratios compare B with A.
Effect sizes use different denominators and are not interchangeable power inputs. Negative bias-corrected variance estimates are retained. Generalized eta squared here supports a single manipulated repeated-measures factor; mixed designs require their full sums-of-squares structure. Undefined ratios receive no automatic correction.
Mean differences and Hedges’ correction · Repeated-measures standardizers · ANOVA conversions · Rank effects
Know what the calculation assumes.
Power Studio 0.2.0 provides fourteen study designs, including three repeated-measures targets. It does not yet replace every G*Power analysis.
Available methods
Choose a planning question.
- Find sample size
- Specify the effect, alpha, target power, and design. Find the smallest feasible integer sample size in the supported range.
- Calculate power
- Specify the effect and design at an available sample size. Calculate the probability of rejecting the null under those assumptions.
- Find detectable effects
- Specify the sample size, alpha, and target power. Find the population effect that would achieve that power. Smaller effects can still produce significant results.
- Find a decision threshold
- Specify the sample size, effect, and target power. Solve for alpha. This is an advanced planning exercise; a higher alpha increases false-positive risk.
- Balance error probabilities
- Specify the sample size, effect, and beta/alpha ratio. Solve for alpha and beta. The chosen ratio is not a full cost-benefit analysis.
Coverage and limits
Welch power, rank-test power, logistic and Poisson models, multivariate methods, unequal-size or longitudinal cluster designs, reliability planning, equivalence, precision planning, and waveform/SPM power need additional engines and validation. This version does not silently approximate those designs with a different test.
The normal-theory methods use numerical integration and distribution calculations. Correlation and proportion calculations use the approximations identified above. The cluster calculation uses a two-group t test of equal-size cluster means. Repeated-measures calculations use the conventional univariate F planning method, with epsilon-adjusted degrees of freedom and noncentrality for within-person effects. ANCOVA assumes balanced covariates. Reference checks compare representative results with R; they do not establish validity for every possible input.
Teaching examples are illustrative assumptions. They are not published estimates or recommendations for a specific intervention.
References
- G*Power: official software and documentation
- Faul and colleagues, 2007: G*Power 3
- Faul and colleagues, 2009: correlation and regression power
- Lakens, 2022: Sample Size Justification
- CONSORT extension: cluster randomized trials
- Repeated-measures planning formulas and verification examples
- R documentation: t-test power and two-sided rejection