How much a survey result could swing, in either direction, from sampling alone.
This result updates live from the fields on the left; nothing is saved or transmitted.
Margin of error is the plus-or-minus figure attached to a survey result: "52%, margin of error 3 points" means the true figure in the full population is probably somewhere between 49% and 55%. It comes from the formula z times the square root of p times one minus p, divided by n. Bigger samples shrink it because n sits under a square root; the confidence level you pick sets how cautious the z multiplier is.
The relationship is not a straight line. Doubling your sample does not halve your margin of error; you need to quadruple it. Here is the margin at p = 50% and 95% confidence across a few common survey sizes, from this calculator's own formula:
The 1,000-respondent row is close to why so many national polls settle on roughly that size: it lands the margin near ±3 points, a rough industry standard, without the cost of chasing a much bigger sample for a small further gain. See the sample size calculator to work this table backward from a target margin.
No, margin of error is half the width of the interval. Add and subtract it from your sample proportion and you get the confidence interval; this page reports both so you don't have to do that step by hand.
Because p times one minus p is largest when p equals 50%, which makes the margin of error largest too. Quoting the 50% figure is the conservative, worst-case number, valid no matter which way the actual result leans.
Usually a small sample size. Margin of error shrinks with the square root of n, so quadrupling your sample only halves the margin; this calculator does not apply a finite-population correction, so it will overstate the margin slightly for a small, fully-surveyed population.
No. This figure only reflects random sampling error, the noise you would expect from a truly random sample of that size. It has nothing to say about nonresponse bias, wording effects, or a sample that was not actually random, all of which can matter more than the number itself.