← Math & statistics

Statistics Planning

Sample Size Calculator

Other languages English 日本語 Español 简体中文

Plan sample size for survey proportions, mean estimates, and balanced A/B rate comparisons. Set a confidence level, choose a target margin of error, and review the required n before you collect data.

This page is for planning precision, not power. If your question is test sensitivity, move to power analysis. After data collection, switch to t-test, chi-square test, or the CI & hypothesis test wizard when you are ready to analyze results.

How to use

  1. Choose the planning mode that matches your decision: one proportion, one mean, or a balanced two-proportion comparison.
  2. Enter a confidence level and the largest margin of error you can tolerate.
  3. Use a realistic planning rate or sigma, then round up and treat the result as the minimum sample to collect.

Wave 2 statistics expansion

Three sample-size planning workflows

Use this when you need a survey-style sample size for one rate or share, such as approval, defect, or conversion prevalence.

Inputs

Plan a survey or audit when the output is one proportion such as approval, defect, prevalence, or response share.

What the page computes

  1. The selected mode sets the planning formula for one proportion, one mean, or two balanced proportions.
  2. The page uses the chosen confidence level to compute the matching normal critical value.
  3. The required sample size is rounded up so the reported precision target is met or slightly exceeded.

How to choose the right mode

Survey vs experiment

Survey sizing and A/B sizing often sound similar, but they solve different planning problems. Survey mode estimates one rate such as approval or prevalence. Two-proportion mode estimates the gap between two rates under balanced groups. Neither mode on this page answers the power-analysis question of how likely you are to detect a minimum effect under a test.

If your stakeholders ask, “How many responses do we need so the estimate is within ±3 points?”, this page is the right start. If they ask, “How many users do we need to detect a +2 point lift with 80% power?”, that is a power-analysis problem and should be handled separately.

What finite population correction means

Finite population correction matters when your sample is a meaningful fraction of the full population. For example, auditing 180 records out of 400 is different from surveying 180 people out of a city of millions. In the small-population case, each observed unit removes more uncertainty, so the required sample can be smaller than the infinite-population formula suggests.

Turn finite population correction on for surveys, audits, and inventories with a known bounded population. Leave it off for large populations, streams of future traffic, or A/B tests where the practical population is not fixed in the same way.

What to do after sizing

After collection, move to a confidence interval or chi-square workflow if you want inference instead of planning.

FAQ

When should I use this page instead of a power analysis tool?

Use this page when your goal is precision: you want a confidence interval or rate estimate to stay within a chosen margin of error. Use a power analysis tool when the goal is to detect a minimum effect with a specified power.

Why does 50% produce the largest survey sample size?

For a single proportion, variance is largest near p = 0.50. That makes the required sample size conservative. If you have prior evidence that the true rate is farther from 50%, the required sample can be smaller.

What does finite population correction change?

Finite population correction reduces the required sample when you are drawing from a limited population and your planned sample is a noticeable fraction of that population. It matters most for surveys or audits of small populations.

Does the A/B mode calculate power?

No. The A/B mode on this page plans a balanced two-proportion sample size for a target confidence interval width. It does not include hypothesis-test power or minimum detectable effect workflows.

Does the share URL include my inputs?

No. The share URL stores only lightweight settings such as mode, confidence level, and whether finite population correction is enabled. Entered numeric values stay in your browser.

Comments (optional)

To reduce load, comments are fetched only when needed.