Skip to content
MasterMath

Sample Size Calculator

How many people to survey for a given margin of error and confidence level, with the finite population correction for when the group is not enormous.

Sample needed

—

Sample needed—
Before the population correction—
Critical value—
Share of the population—
Margin requested—
Confidence requested—

How this was worked out

    The formula

    n = z² · p · (1 − p) / e², corrected for a finite population

    What it means

    Sample size follows from three decisions: how much confidence you want, how much margin of error you accept and what proportion you expect to find. With no idea of the proportion, 50 % is used, which demands the largest sample and is therefore the safe choice. The finite population correction kicks in when the whole group is not much bigger than the sample: surveying four hundred people in a town of a thousand is not the same as in a country.

    How to work it out by hand

    1. Pick the confidence level and look up its critical value: 1.96 for 95 %
    2. Decide the margin of error you accept, as a fraction
    3. Apply n = z² · p · (1 − p) ÷ e², with p = 0.5 if the proportion is unknown
    4. If the population is small, correct by dividing by 1 + (n − 1) ÷ N

    What is worth knowing

    This number is the mathematical minimum for the margin you asked for, and it assumes simple random sampling and a full response rate. In practice you have to approach more people than the formula says because some will not answer, and if the sampling is clustered or stratified the calculation changes. And the famous figure of 385 people for "any country" is real and rightly baffling: past a certain size, the population stops mattering almost entirely.

    Frequently asked questions

    Why does it say 385 for a whole country?

    Because past a certain size the population stops mattering: what drives the number is the margin of error, not how many people there are.

    What proportion do I enter if I do not know it?

    50 %. It demands the largest sample, so it keeps you on the safe side.

    When does population size matter?

    When the sample approaches the total: below a few thousand, the finite correction cuts the number considerably.

    Is surveying that many enough?

    It is the minimum assuming everyone answers. Since some will not, you need to approach rather more.

    Does it work for any sampling method?

    For simple random sampling. Clustered or stratified designs need a different calculation and usually a larger sample.