Overview

How many people do you need to survey? This works out the sample size for estimating a proportion to a given margin of error and confidence level.

Enter a confidence code (1 for 90%, 2 for 95%, 3 for 99%), the margin of error, your best guess of the proportion (50% if unsure) and the population size, or 0 for a very large population.

Worked examples

InputsResult
Confidence level95%Margin of Error (%)5Estimated Proportion (%)50Population Size (0 for large/infinite)0Required sample size: 385Infinite-pop approximation: 385
Show 2 more examples
Confidence level95%Margin of Error (%)3Estimated Proportion (%)50Population Size (0 for large/infinite)0Required sample size: 1,068Infinite-pop approximation: 1,068
Confidence level99%Margin of Error (%)5Estimated Proportion (%)30Population Size (0 for large/infinite)2,000Required sample size: 437Infinite-pop approximation: 558

Required sample size by margin of error

With the calculator's starting values, the required sample size is 385. The table shows how it changes with the margin of error.

Margin of Error (%)Required sample size
3%1,068
4%601
5%starting value385
6%267
7%196

Other inputs: Estimated Proportion: 50% · Population Size: 0 0 for large/infinite. Change them in the calculator above for your own numbers.

How it works

The basic sample size is z² × p(1 − p) ÷ E². For a known, smaller population, a finite population correction reduces it. Using 50% for p gives the largest, safest size. It assumes a simple random sample.

The formula

n₀ = z² × p(1 − p) ÷ E²
z1.645, 1.96 or 2.576
EMargin of error as a decimal

Practical tips

  • 95% confidence with ±5% needs 385 responses.
  • ±3% needs 1,068, and ±1% needs 9,604.
  • Plan for non-response. If you expect 25% to reply, invite 4 times the sample.

Frequently asked questions

Why does population size matter so little?

Once the population is large, precision depends mainly on the sample size.

Why use 50%?

It gives the largest sample, so you're covered whatever the true proportion is.

Does this work for means?

No. For means, n = (z × σ ÷ E)².