Sample Size Calculator β survey sample size
Find how many survey responses you need from confidence level, margin of error and population size, with finite-population correction.
π Runs entirely in your browser. Your input and output are never sent to a server or stored β fully private.
Formula: nβ = zΒ²Β·p(1βp)/eΒ², with finite-population correction n = nβ/(1+(nββ1)/N) when a population size is given. 50% distribution is the safest (largest) choice. Click a result to copy.
About this tool
This sample size calculator tells you how many survey responses you need for a result you can trust. Choose a confidence level (usually 95%), the margin of error you can live with (say Β±5%), the expected response split (50% is the safe default) and, if your group is small, the total population. It applies Cochran's formula with the finite-population correction and shows the required sample, plus a quick table for tighter or looser margins. It runs entirely in your browser β no data, no sign-up.
Frequently asked questions
Why default the response distribution to 50%?
Because 50/50 is the most cautious assumption β it produces the largest required sample. The formula uses p Γ (1 β p), which peaks at p = 0.5, so if you have no prior idea how people will answer, 50% guarantees your sample is big enough whatever the true split turns out to be. If you genuinely expect a lopsided result (say 90/10 from past data), entering that lets you get away with a smaller sample.
What does the finite-population correction do?
For a huge population, the required sample barely depends on the total β polling a whole country or a whole city needs about the same number. But when your population is small (a class, a company, a club), you do not need as many responses, and the correction shrinks the requirement accordingly. For example Β±5% at 95% needs about 385 responses from a very large group, but only around 278 if the whole population is 1,000.
Is this all computed locally?
Yes β the whole calculation is JavaScript in your browser, so nothing you enter is uploaded, and it works offline once loaded. One caveat the number can't capture: it assumes a random, representative sample. If your respondents self-select or you can't reach part of the population, no sample size fixes that bias β coverage and randomness matter as much as raw numbers.