Compute statistical power, required sample size, and visualize power curves for two-sample t-tests.Understand the interplay between sample size, effect size, significance level (α), and power (1−β).
Statistical power is the probability that a hypothesis test correctly rejects the null hypothesis when a true effect exists. Formally, power = 1 − β, where β is the probability of a Type II error (failing to detect a real effect). A study with low power is a waste of resources and, more troubling, may lead to false negative conclusions that stall scientific progress.
Power = P( reject H₀ | H₁ is true )
where H₀ is the null hypothesis and H₁ is the alternative hypothesis.
The four key pillars of power analysis are: effect size (magnitude of the phenomenon), sample size (number of observations), significance level α (tolerance for false positives), and power 1−β (sensitivity to detect effects). Change any one, and the others adjust accordingly.
This tool computes power for the two-sample independent t-test, one-sample t-test, and paired t-test using the non-central t-distribution. Unlike normal approximations, the non-central t-distribution exactly models the behavior of the test statistic under the alternative hypothesis. The non-centrality parameter (NCP) is:
δ = d · √(n/2) (for two-sample t-test, equal n per group)
where d is Cohen's d and n is the sample size per group. Power is then computed as:
Power = P( Tν,δ > tcrit ) + P( Tν,δ < −tcrit )
with ν = 2n − 2 degrees of freedom and tcrit the critical value from the central t-distribution at level α. The calculator uses numerical integration (Gauss–Legendre quadrature) to evaluate the non-central t CDF, ensuring high accuracy across all sample sizes.
For sample size estimation, we perform a bisection search to find the smallest n that achieves the target power (default 0.80). The power curve is rendered on an interactive canvas, showing how power increases with sample size and marking your current design.
A power of 0.80 is conventionally considered adequate in many fields (Cohen, 1988). However, some domains (e.g., genomics, physics) demand higher power (0.90–0.99) due to multiple testing or safety requirements. Our badge system gives you a quick visual cue:
A researcher hypothesizes that a new teaching method improves test scores. Based on prior literature, the expected effect size is Cohen's d = 0.35 (small-to-moderate). With α = 0.05 and n = 40 per group, the calculator returns power ≈ 0.53. This means the study has only a 53% chance of detecting the effect—nearly a coin flip. The researcher increases n to 80 per group, boosting power to 0.83. The additional cost is justified by the much higher probability of obtaining meaningful results.
| Effect Size | Cohen's d | Typical Use Case | Power (n=30, α=0.05, two-tailed) |
|---|---|---|---|
| Very Small | 0.10 | Subtle behavioral differences | 0.08 |
| Small | 0.20 | Educational interventions, small effects | 0.17 |
| Medium | 0.50 | Typical psychological / social effects | 0.70 |
| Large | 0.80 | Strong medical / biological effects | 0.95 |
| Very Large | 1.20 | Obvious, easily detectable effects | 0.999 |
With the growing emphasis on replication in science, power analysis has become more critical than ever. Underpowered studies contribute to the “replication crisis” because they produce unreliable estimates and inconsistent results. By using this calculator, you can design studies that are adequately powered, increasing the likelihood that your findings are reproducible. We encourage researchers to report power analyses in their methods sections and to consider sensitivity power analyses when sample size is fixed.