Poll example
A poll of 1,000 voters finds 52% support. At 95% confidence: ME = 1.96 × √(0.52 × 0.48 ÷ 1000) ≈ 0.031, so ±3.1 percentage points. The true support is plausibly anywhere from 48.9% to 55.1%.
Margin of error shrinks with the square root of the sample: to halve it you need four times as many respondents.
What it does not cover
The margin of error only accounts for random sampling variation. It does not cover biased samples, leading questions, non-response or people not telling the truth, which can be much larger sources of error.
Frequently asked questions
Why use 50% when I don’t know the proportion?
p(1 − p) is largest at p = 0.5, so it gives the most conservative (largest) margin of error.
When should I use the finite population correction?
When your sample is more than about 5% of the whole population, for example surveying 200 of a school’s 800 students.