Power Analysis Calculator

Pick a test, enter the effect size, alpha and power, and get the sample size you need.

Loading calculator…

This free power analysis calculator finds the required sample size for your study, or the power you already have. It is a statistical power calculator and a sample size calculator in one free, browser-based tool, an online power analysis for the most common designs. Searches such as power calculator sample size, power analysis sample size calculator or power and sample size all lead to the same job: power analysis links the effect size, the significance level, the target power and the sample size, so entering any three solves for the fourth. It covers t-tests, two proportions, correlation, one-way ANOVA and chi-square tests, and it shows each step, so you can calculate power, check it in R, Python or Excel and copy a ready methods sentence. For a medium effect (d = 0.5) at alpha 0.05 and 80% power, the answer is 64 participants per group, 128 in total.

How to use this power analysis calculator

  1. Choose what to find. Pick Sample size to calculate the number of participants, or Power to calculate power for a sample you already have.
  2. Pick the test. Choose the independent samples t-test, the one-sample t-test or paired differences, the two-sample proportion test, correlation, one-way ANOVA or chi-square.
  3. Enter the effect size. Type your own value, or press Small, Medium or Large to fill Cohen’s benchmark. For two proportions, type both percentages and the calculator works out the effect size.
  4. Set alpha, power and tails. The defaults are 0.05, 80% and two-sided. Open More options for an unequal allocation ratio and expected dropout.
  5. Press Calculate. You get the sample size per group and in total, the achieved power, the critical value, the noncentrality and a numbered working.

Each result comes with two tables, the sample size for small, medium and large effects and the sample size at other power levels. In Power mode you also see the smallest effect your sample could detect. The inputs are not saved anywhere, so reload the page to start fresh.

What is statistical power? Meaning and definition

Statistical power is the probability that a hypothesis test detects a real effect of a given size. It is written as 1 − β, where β is the type II error rate, the chance of missing an effect that is really there. A study with 80% power has a 20% chance of missing the effect it was designed to find. The power of a test, the power of a statistical test and statistical power all mean the same thing, and people also ask what is power in statistics or what is power in stats. Power is a planning target, not a measurement of a finished study. The statistical power definition, in one line: the chance of finding an effect that exists. Power statistics is the part of statistics that studies it.

Eighty percent is the common target. High-stakes studies, such as clinical trials, often plan for 90% or 95%. Raising power always costs participants, so the target is a trade-off between the risk of a miss and the cost of a bigger study.

Two t curves: no effect and a real effect of d = 0.5, with alpha, beta of 19.85% and power of 80.15% shaded

The picture shows why. The left curve is the test statistic when there is no effect, and alpha is the area beyond the critical value. The right curve is the test statistic when the effect is real. The area of that curve past the critical value is the power, and the area before it is β. Power rises when the effect is larger, the sample is bigger, alpha is looser or the data are less noisy.

What is a power analysis and why do it?

A power analysis, in research, is a calculation that links four quantities: the effect size, the significance level, the power and the sample size. Fix any three and the fourth is determined, so a power analysis statistics workflow can solve for whichever one you do not know. Researchers use it to decide how many participants to recruit, to defend a sample size in a grant or ethics application, and to explain why a study had too little power to find an effect.

The reason to do it before the study is practical. A study that is too small may miss a real effect, and a study that is too large wastes time and, in medical research, exposes more people to a treatment than needed. Power analysis helps you find the sample size that gives a fair chance of success without waste. It also forces you to state, in advance, how big an effect would matter.

Calculation modes: a priori, post hoc and sensitivity

The calculator solves for sample size or for power. Power analysis comes in three kinds, depending on what you know and when you ask:

  1. A priori (before data collection). You choose the expected effect size, alpha and target power, and solve for the required sample size. This is the main use and the default mode here. Many guides call it how to do a power analysis, and the effect size selection is the hardest part, and it is the answer to how to determine sample size from power.
  2. Sensitivity analysis. The sample is fixed, for example 30 per group, and you ask for the smallest effect it could detect with 80% power. For 30 per group that is d = 0.7356. It is the honest option when the sample is already set.
  3. Post hoc power. Power is computed after the study from the observed data. It adds no information beyond the p-value, because a non-significant result always has low observed power. Report a confidence interval instead, and use a sensitivity line to describe what the study could have found.

Power achieved depends on the existing sample size, so in Power mode the calculator shows the power at your sample size and the sensitivity line. For d = 0.5 with 30 per group, power is 47.8%, which is well below 80%. For d = 0.8 with 20 per group, power is 69.3%.

Supported tests and effect sizes

Each test has its own effect size, and all of the calculator’s tests use Cohen’s conventions. Cohen’s d is the standardized mean difference for the independent t-test, and the others have their own letters:

Tests in the power analysis calculator and their effect sizes
TestEffect sizeSmallMediumLargeMethod
Two independent groups (t-test)d0.20.50.8Exact noncentral t
One sample or paired (t-test)dz0.20.50.8Exact noncentral t
Two proportionsh0.20.50.8Normal approximation
Correlationr0.10.30.5Fisher z approximation
One-way ANOVAf0.10.250.4Noncentral F
Chi-square testw0.10.30.5Noncentral chi-square

The t-test comes in independent, paired and one-sample forms, and the other tests have one form each. The effect sizes are standardized, so the benchmarks work for any unit. A one-sample or paired t-test uses dz, the mean difference divided by the SD of the differences. If you only have means and standard deviations, use the helper to get d, which is the mean difference divided by the pooled SD.

Inputs and parameters

  1. Effect size. The input with the largest influence. Halving the effect roughly quadruples the required sample: at 80% power, d = 0.5 needs 64 per group and d = 0.25 needs 253. Use the smallest effect that would matter in practice, or an estimate from prior work, and if you are unsure take a cautious estimate, which is the usual fallback when prior work is thin. Given an effect size you trust, the rest follows.
  2. Significance level (alpha). The default is 0.05. Lower values such as 0.01 are stricter and need more participants: at d = 0.5 and 80% power, alpha 0.01 needs 96 per group.
  3. Desired power. The default is 80%, with buttons for 90% and 95%. The options in the other-power table are 70%, 80%, 85%, 90% and 95%.
  4. Tails. Two-sided is the default. A one-sided test needs 51 per group instead of 64 at d = 0.5, but it only detects an effect in a prespecified direction, chosen before the study.
  5. Available sample size. In Power mode, type the participants you have and the calculator solves for power, so power is the output and the sample is the input.
  6. Allocation ratio and dropout. Both are optional and sit under More options.

Mean difference, SD and sample size are tied together through d, so a power analysis calculator with effect size input is also a way to see how noise affects a plan.

Worked example: power analysis for an independent t-test

Suppose you plan to compare two groups, expect a medium effect of d = 0.5, and want 80% power at α = 0.05, two-sided. The steps the calculator shows are:

  1. The critical t value for a two-sided test with df = 126 is 1.979.
  2. For 64 per group, the noncentrality is 0.5 × √(64 ÷ 2) = 2.828.
  3. Power is the probability that a noncentral t with that noncentrality passes the critical value: 80.15%. The type II error rate β is 19.85%.
  4. With 63 per group, power is 79.52%, just under the target, so the answer is 64 per group, 128 in total. The calculator rounds up and never down.
Line chart: power of a two-group t-test rises with sample size and reaches 80.15% at 64 per group for d = 0.5
Worked example: two groups, d = 0.5, alpha = 0.05, two-sided
ItemValue
Participants per group64
Total participants128
Achieved power80.15%
Type II error (β)19.85%
Degrees of freedom126
Critical t1.979
Noncentrality2.828

The curve shows the same story: power climbs quickly at first and then flattens, so the last 10% of power costs far more participants than the first 50%.

How to calculate statistical power: formula and steps

There is no single formula for every test, but the steps are the same, and they are how to calculate power and how to compute power by hand:

  1. Find the critical value of the test from alpha, the tails and the degrees of freedom.
  2. Compute the noncentrality, which measures the true effect on the test statistic’s scale. For two equal groups, it is d × √(n ÷ 2).
  3. Power is the probability that a noncentral t with that noncentrality passes the critical value.

Power = P(test statistic beyond the critical value | real effect)

That is the power calculation behind the worked example, and how to calculate statistical power in general. The calculator solves the power analysis formula in reverse when it searches for the smallest n that reaches your target. A table of power calculations, like the ones below, repeats it for several n. The statistical power formula for the normal approximation gives n per group = 2 × ((z for alpha ÷ 2 + z for power) ÷ d)², which is only a guide for the t-test.

How to increase statistical power

  1. Increase the sample size. This is the most direct way, but power flattens as n grows.
  2. Target a larger effect. A bigger real effect, such as a stronger treatment, is easier to detect.
  3. Reduce noise. Better measurement or a paired design shrinks the standard deviation, which raises d.
  4. Loosen alpha or use one tail. Alpha 0.10 or a one-sided test raises power, at the cost of more false positives or of missing the other direction.

Powering a study well means deciding these things before the data come in. Done this way, power analysis in research is a planning step, not an afterthought. If your goal is to find the smallest whole-number sample that reaches your target, the calculator rounds up for you.

Sample size for small, medium and large effects

The same settings give very different answers across effect sizes. The tables use two-sided independent t-tests at alpha 0.05 and show why the effect size is the decision that matters most.

Sample size per group by effect size (80% power)
EffectdPer groupTotal
Small0.2394788
Medium0.564128
Large0.82652
Sample size per group by power level (d = 0.5)
PowerPer groupTotal
70%51102
80%64128
85%73146
90%86172
95%105210

Going from 80% to 95% power at d = 0.5 needs 41 more participants per group, which is 64% more. The calculator gives both tables for your own settings.

Adjusting for dropout

Some participants leave before the end of a study. To keep the analysed sample at the planned size, inflate the sample by dividing by (1 − dropout rate) and rounding up. With 64 per group and 10% expected dropout, 64 ÷ 0.9 = 71.1, so enrol 72 per group, which is 144 in total. The calculator shows the analysed sample and the number to enrol side by side.

Other designs: paired, proportions, correlation, ANOVA and chi-square

The same logic works for the other tests. At alpha 0.05, 80% power and two-sided tests, the calculator gives:

Sample size for the other tests at 80% power and alpha 0.05
DesignEffect sizeResult
Paired t-test, mediumdz = 0.534 pairs
Paired t-test, smallerdz = 0.390 pairs
Two proportions, 50% vs 65%h = 0.3047170 per group, 340 in total
Correlationr = 0.385 participants
One-way ANOVA, 3 groupsf = 0.2553 per group, 159 in total
Chi-square, df = 1w = 0.388 participants

The calculator has no one-sample proportion test and no Poisson rate test. A paired design needs far fewer people because each person acts as their own control: 34 pairs instead of 128 participants for the same medium effect. The two-proportion comparison uses the effect size h, which turns percentages into a standardized difference. For any test, the sample size follows once you give an effect size and the other inputs, and a given effect size at a different alpha changes the answer.

Methods, assumptions and limitations

For the t-tests the calculator uses the exact noncentral t distribution. The power is computed by numerical integration, and the answers match R and Python to many decimals. The other tests use the standard formulas:

  1. Two proportions: a normal approximation using Cohen’s h, without continuity correction. It is accurate for moderate samples, and unreliable for rare events or very small counts.
  2. Correlation: the Fisher-z normal approximation, n = ((z for alpha ÷ 2 + z for power) ÷ atanh(r))² + 3, rounded up. For r = 0.3 that gives 84.93, so 85.
  3. One-way ANOVA: the noncentral F distribution with fixed effects and equal group sizes, where the noncentrality is f² times the total sample.
  4. Chi-square: the noncentral chi-square distribution, with the noncentrality w² times the sample.

The assumptions are the same as the tests themselves: independent observations, normality or a moderately large sample, and equal variances (equal population variances) for the independent t-test. The ANOVA covers fixed-effects designs only. Small samples make the approximations less accurate, so treat proportion and correlation results as planning figures. The calculator does not include a design effect for clustered designs, and it does not cover regression, repeated measures, survival or Poisson rate designs. Those limitations mean a quick estimate here is a starting point, not a substitute for a statistician on a complex trial.

Reporting a power analysis

Journals and ethics boards ask you to report the inputs, so a complete statement names the test, sidedness, alpha, target power, effect size and its source, and the resulting sample size, plus any dropout allowance. A methods sentence for the worked example reads:

An a priori power analysis for a two-sided independent-samples t-test (d = 0.5, alpha = 0.05, power = 80%) indicated a required sample size of 64 per group (128 in total).

State where the effect size came from, such as a pilot study, prior work or the smallest effect of practical interest. Press Copy results to put the power analysis statement, with your own numbers, on your clipboard.

Check it yourself in R, Python and Excel

You can check the calculator’s answer for the worked example in free software:

  1. R: power.t.test(delta = 0.5, sd = 1, sig.level = 0.05, power = 0.8) gives n = 63.77 per group, which rounds up to 64. The pwr package gives the same answer with pwr.t.test(d = 0.5, power = 0.8, sig.level = 0.05).
  2. Python: TTestIndPower().solve_power(effect_size=0.5, alpha=0.05, power=0.8) from statsmodels gives 63.77, and 64 after rounding up.
  3. Excel: Excel has no noncentral t function, so it can only approximate. The formula =2*((NORM.S.INV(1-0.05/2)+NORM.S.INV(0.8))/0.5)^2 gives 62.79, which rounds up to 63. This is approximate and usually one or two below the exact answer of 64.

Our z table lists the normal values used in the approximations. For the spread of your own data, use the standard deviation calculator, and for how precisely a sample estimates a mean, see the standard error calculator.

Free online alternative to G*Power and desktop tools

G*Power (also typed gpower) is a long-standing desktop program for power analysis, and many search for a power analysis calculator in G*Power style, such as g power power analysis. This page is a free online alternative for the common cases: no download, no account and each step shown. G*Power and other desktop tools cover many more designs, such as regression and repeated measures. If your design is on the list above, the calculator is quicker, and if it is not, use a desktop tool.

Related calculators

To describe the data you already have, open the mean calculator and the standard deviation calculator. See the methodology page for how every calculator is checked.

Power analysis calculator FAQs

What is power analysis?

Power analysis is the step in planning a study that links four numbers: the effect size you want to detect, the significance level (alpha), the power you want and the sample size. Fix any three and the fourth follows. Most researchers use it before collecting data, to find how many participants a study needs.

What is statistical power?

Statistical power is the probability that a test detects a real effect of a given size. It equals 1 − β, where β is the type II error rate. A power of 80% means that if the effect is real and as large as you planned for, the test will reach significance in about 8 of 10 studies.

How do you calculate statistical power?

Compute the critical value from alpha and the degrees of freedom. Then find the chance that the test statistic passes it when the effect is real. For a t-test that chance comes from the noncentral t distribution, with a noncentrality of d × √(n ÷ 2) for two equal groups. The calculator does this and shows each step.

How many participants do I need for 80% power?

It depends on the effect. A two-sided independent t-test at alpha 0.05 needs 394 per group for a small effect (d = 0.2), 64 for a medium effect (d = 0.5) and 26 for a large effect (d = 0.8). Halving the effect roughly quadruples the sample.

What is a good power level?

Most fields treat 80% as the minimum. Studies where a miss is costly, such as clinical trials, often plan for 90% or 95%. At d = 0.5 the per-group sample is 64 for 80%, 86 for 90% and 105 for 95%.

What is the difference between a priori, post hoc and sensitivity power analysis?

A priori power analysis happens before data collection and solves for the sample size. Post hoc analysis computes power after a study from the effect found in the data, which adds nothing beyond the p-value. Sensitivity analysis asks what effect a fixed sample could detect, and it is the honest option when the sample is already set.

Why is post hoc power not useful?

Observed power is a direct function of the p-value. A non-significant result always comes with low observed power, so the number tells you nothing new. If a study is finished, report a confidence interval and use the sensitivity line to describe what effect the sample could have detected.

How do I choose an effect size?

Use the smallest effect that would matter in practice, or an estimate from prior work. If neither exists, take a cautious estimate, because published effects tend to be larger than later replications. Cohen’s benchmarks (small, medium, large) are reference points, not facts about your field.

Should I use a one-tailed or two-tailed test?

Use two-sided unless the direction is prespecified and an effect in the other direction would not matter. A one-sided test needs fewer participants, 51 per group instead of 64 at d = 0.5, but it cannot detect an effect in the opposite direction.

How do I adjust the sample size for dropout?

Divide the analysed sample by (1 − dropout rate) and round up. With 64 per group and 10% expected dropout, 64 ÷ 0.9 = 71.1, so enrol 72 per group, 144 in total.

Does the calculator work for unequal group sizes?

Yes. Open More options and set the allocation ratio n2 ÷ n1 for the independent t-test or two proportions. With a ratio of 2 at d = 0.5, you need 48 in the first group and 96 in the second, 144 in total.

Is this a G*Power alternative?

It is a free browser-based alternative for the common cases: t-tests, two proportions, correlation, one-way ANOVA and chi-square. G*Power is a desktop program with many more designs, such as regression and repeated measures. If your design is not listed here, use a desktop tool or ask a statistician.

Can I use it for regression, survey or Poisson designs?

Not yet. The calculator covers six tests. Multiple regression, repeated measures, survival, cluster-randomized and Poisson rate designs need other tools. A survey sample size for a margin of error is a different question and is not what this page answers.

Is the answer exact?

For the t-tests, yes. They use the exact noncentral t distribution, and the answer matches R and Python to many decimals. Proportions and correlation use standard normal approximations, which are accurate for moderate samples. The page says which method each result uses.

How accurate is the Excel formula?

Excel has no noncentral t function, so the check uses the normal approximation n = 2 × ((z for alpha ÷ 2 + z for power) ÷ d)². For d = 0.5 that gives 62.79, which rounds up to 63. The exact answer is 64, so treat the Excel line as approximate.

Does the calculator save my numbers?

No. It runs in your browser with no account, and nothing you type is stored or sent. Press Copy results to keep a record, including the methods sentence.

Easy Stat Calc · Results checked against Python, R and Excel · Last updated November 18, 2026 · Methodology