Statistical Power and Sample Size: Why Most Studies Are Underpowered

An underpowered study wastes participant time, funding, and ethical goodwill on a result that cannot reliably distinguish a true effect from noise. Yet many manuscripts still reach peer review without a power calculation, according to Editage's research quality guide. Journals including JAMA Neurology, the British Journal of Surgery, and Molecular Genetics and Metabolism require power calculations to be clearly stated in the methods section. If you are planning a study, the power analysis determines your sample size before you collect a single data point. Use our Sample Size Calculator to run the numbers, then verify your study has adequate power with the Statistical Power Calculator.

What Is Statistical Power?

Statistical power is the probability that a study will detect a real effect if one exists in the population. A power of 0.80 means an 80% chance of finding a genuine effect and a 20% chance of missing it. That 20% miss rate is a Type II error, also called a false negative.

Power depends on four interlocking values. Change any one and the others shift:

• Sample size (n): the number of participants or observations
• Effect size: the magnitude of the difference or relationship you are trying to detect
• Alpha (significance level): the threshold for declaring a result significant, almost always 0.05
• Power (1 minus beta): the desired probability of detecting the effect, conventionally 0.80 or 0.90

If you know any three of these values, you can calculate the fourth. In practice, alpha is fixed at 0.05 and effect size is estimated from prior literature or pilot data. That leaves sample size as the value you solve for.

The Core Formulas by Study Design

Different study designs require different sample size formulas. The structure is similar across designs: sample size increases as power increases, as alpha becomes stricter, and as the expected effect size shrinks.

For a two-group comparison of means (independent t-test):

n (per group) = 2 x [(z(1-alpha/2) + z(1-beta)) / d] squared

Where d is Cohen's d, the standardized effect size. For alpha = 0.05 (two-tailed) and power = 0.80, z(1-alpha/2) = 1.96 and z(1-beta) = 0.84.

For a cross-sectional prevalence estimate:

n = [Z squared x p(1-p)] / e squared

Where Z = 1.96 for 95% confidence, p is the expected proportion, and e is the margin of error.

For a correlational study (Pearson's r):

n = [(z(1-alpha/2) + z(1-beta)) / C] squared + 3

Where C = 0.5 x ln[(1+r) / (1-r)], the Fisher r-to-z transformation.

Step-by-Step Example: Two-Group Comparison

A researcher plans a randomized controlled trial comparing a new intervention to a control group. Prior studies suggest a medium effect size (Cohen's d = 0.5). Alpha is set at 0.05, power at 0.80.

1. z(1-alpha/2) = 1.96 (for two-tailed alpha = 0.05)
2. z(1-beta) = 0.84 (for power = 0.80)
3. d = 0.5
4. n = 2 x [(1.96 + 0.84) / 0.5] squared = 2 x [5.6] squared = 2 x 31.36 = 62.72

Round up to 63 participants per group, 126 total. This is the minimum sample size to detect a medium effect with 80% power at a 0.05 significance level.

If the researcher wants 90% power instead, z(1-beta) becomes 1.28. The calculation: n = 2 x [(1.96 + 1.28) / 0.5] squared = 2 x [6.48] squared = 2 x 41.99 = 83.98, or 84 per group, 168 total. Increasing power from 80% to 90% requires 33% more participants.

If the expected effect size is small (d = 0.2) instead of medium: n = 2 x [(1.96 + 0.84) / 0.2] squared = 2 x [14] squared = 2 x 196 = 392 per group, 784 total. Halving the effect size roughly quadruples the required sample size.

Power by Study Design: Comparison

The cross-sectional formula does not use power in the same way because it estimates a proportion rather than testing a difference. The sample size of 350 for a prevalence estimate at 35% with 5% margin of error comes from n = (1.96 squared x 0.35 x 0.65) / 0.05 squared = 349.6, rounded up. If the survey uses multistage cluster sampling, multiply by a design effect of 1.5, raising the target to 525 respondents.

Why Underpowered Studies Are an Ethical Problem

Exposing participants to interventions when the study cannot detect a meaningful effect is ethically problematic. This is not a marginal concern. A 2026 study published in BMC Medical Research Methodology found that robust power and sample size calculations remain difficult to carry out for many common study designs, and that traditional approaches often rely on restrictive distributional assumptions.

Underpowered studies carry several consequences:

• False negatives: a real effect exists but the study fails to detect it
• Inflated effect size estimates: only chance-inflated large effects reach significance in low-powered studies, systematically overstating true effects
• Wasted resources: time, funding, and participant burden are spent on studies that cannot yield reliable conclusions

The CONSORT reporting guidelines require authors to justify sample size in the methods section. The American Psychological Association strongly recommends reporting a power analysis. Grant reviewers expect to see one in the proposal.

Common Mistakes in Power Analysis

Using Cohen's conventional benchmarks without domain context. Cohen's small (d = 0.2), medium (d = 0.5), and large (d = 0.8) benchmarks are rough guides. A clinically meaningful effect in a specific field may be much smaller than Cohen's "small." Use prior literature or pilot data to estimate the minimum practically important difference.

Optimistic effect size estimates. If the expected effect size is too high, the calculated sample size will be too low. This is the most common cause of underpowered studies. Use the lower bound of the 95% confidence interval from prior studies as your effect size estimate.

Ignoring attrition in longitudinal designs. Longitudinal studies measure the same participants repeatedly. The correlation between repeated measurements reduces the required n, but anticipated attrition increases it. If you expect 20% dropout, inflate your sample size by 25% (divide by 0.80).

Not accounting for cluster designs. Cluster randomized trials require multiplying the individual-level sample size by a design effect. A typical design effect of 1.5 to 2.0 can double the required sample size.

Running a post-hoc power analysis. Power is a pre-study concept. Calculating power after seeing the data produces a circular argument. Report the a priori power calculation that justified your sample size, not a post-hoc calculation based on observed results.

Related Tools on ProfessionCalculators.com

• Sample Size Calculator to determine the minimum n for your study
• Statistical Power Calculator to verify your study has adequate power
• Confidence Interval Calculator to calculate precision around your estimates

FAQ

What is the minimum acceptable statistical power? The conventional minimum is 0.80 (80%). Some fields, particularly clinical trials and high-stakes research, aim for 0.90 or higher. Studies below 0.50 are generally considered inadequately designed.

Does a larger sample size always increase power? Yes, but with diminishing returns. Doubling sample size does not double power. The relationship is non-linear. At some point, increasing n further detects trivially small effects that have no practical importance.

What effect size should I use for my power analysis? Use the minimum effect size that would be practically meaningful in your field. If prior studies report effect sizes, use the lower bound of their confidence intervals. Cohen's conventional benchmarks (small, medium, large) are fallbacks, not primary references.

Can I use G*Power instead of a calculator? G*Power is a free, widely used tool that covers t-tests, F-tests, chi-square tests, correlations, and regressions. It produces the same results as any properly implemented power calculator. The tool you use matters less than whether your inputs (effect size, alpha, power) are well-justified.

What happens if my study is underpowered? An underpowered study has a high probability of missing a true effect (Type II error). If it does find significance, the effect size estimate is likely inflated. Peer reviewers and editors increasingly reject underpowered studies, and journals following CONSORT guidelines require sample size justification.

Conclusion

Statistical power determines whether your study can answer the question it sets out to address. The conventional threshold of 80% power means a 20% chance of missing a real effect, and many studies fall below even that bar. The required sample size depends on your study design, expected effect size, alpha level, and desired power. Smaller effects, stricter significance thresholds, and higher power targets all require more participants. Estimate your effect size conservatively from prior literature, account for attrition and design effects, and run the calculation before you start collecting data.