Sample Size Calculator
Enter your confidence level and the margin of error you can live with. You get the number of completed responses required, and — accounting for your response rate — how many people you actually need to invite.
The two things you are choosing between
Margin of error is how wide your answer is. A result of 62% with a ±5% margin means the truth is somewhere between 57% and 67%.
Confidence level is how often that range would actually contain the truth if you ran the survey many times. 95% is the near-universal default, and it is what news polls mean when they report a margin without explaining it.
You pick both, and the sample size follows. Neither has a "correct" value — they are statements about how much uncertainty you are willing to accept.
Precision gets expensive very quickly
Sample size scales with the square of the margin of error, which produces a curve most people find counterintuitive.
At 95% confidence, ±5% needs about 385 responses. ±2.5% needs about 1,537 — four times as many for twice the precision. ±1% needs about 9,600.
Look at the three margin outputs on this page. Going from 100 responses to 400 halves your margin, which is a substantial gain. Going from 400 to 1,000 improves it much less. Somewhere between 400 and 1,000 is where most surveys should stop, because past that you are paying a lot for very little additional certainty.
Population size matters less than you expect
The most common misconception about surveys is that a bigger population requires a proportionally bigger sample. It does not.
Polling a city of 50,000 and a country of 300 million both need roughly 385 responses for ±5% at 95% confidence. Sample size depends on the variability you are measuring, not on how many people exist.
The finite population correction only makes a real difference when your sample would be a large fraction of the whole population — surveying 300 people out of 500, say. Below a few percent it changes almost nothing, which is why you can leave the population field at 0 for anything national.
Expected proportion, and why 50% is the safe answer
The formula uses p(1−p), which is largest at p = 0.5. A 50/50 split is the most uncertain outcome, so assuming it gives the largest — and therefore safest — sample size.
Only lower it if you have real prior data. If you already know roughly 10% of customers churn, using 10% instead of 50% cuts the required sample by about 64%. Guessing wrong in that direction leaves you with a margin wider than you planned.
What this does not fix
A big sample does not repair a biased one. If your invitations only reach people who already like you, 10,000 responses will confirm your assumptions very precisely and tell you nothing true.
The margin of error only describes sampling error — the randomness of who happened to answer. It says nothing about leading questions, people who ignored the survey being systematically different from those who replied, or a contact list that never represented your population. Those errors do not shrink with sample size.
Frequently asked questions
How many survey responses do I need?
About 385 completed responses gives a ±5% margin of error at 95% confidence for any large population. Tightening to ±2.5% requires roughly 1,537, because sample size scales with the square of the margin. Most surveys land somewhere between 400 and 1,000.
Does a bigger population need a bigger sample?
Barely. Polling a city of 50,000 and a country of 300 million both need roughly 385 responses for the same precision. Population size only matters when your sample would be a large fraction of it — like surveying 300 people out of 500.
What confidence level should I use?
95 percent is the standard for business and academic work and is what news polls report by default. Use 99 percent when a wrong conclusion is costly, such as medical or safety research — it requires about 73 percent more responses. 90 percent is acceptable for exploratory work.
What should I enter for expected response proportion?
Leave it at 50 percent unless you have prior data. That value produces the largest and therefore safest sample size. Lowering it shrinks the sample considerably — using 10 percent cuts it by around 64 percent — but if your estimate is wrong your real margin will be wider than planned.
Does a large sample fix a biased survey?
No. Margin of error only describes sampling randomness. It says nothing about leading questions, a contact list that never represented your population, or non-respondents differing systematically from respondents. Those errors do not shrink as the sample grows.