ExplainerMargin of errorQuayside Ledger data desk

Why 1,000 people can speak for a whole city

Picture a city where exactly 52% of adults want schools to start later. Nobody can see that number, so 100 pollsters each phone a random sample and report what they heard. Every dot below is one poll. Drag the sample size and watch them.

Fig. 1 · 100 polls of one citytrue share: 52%

1001,00010,000

Margin of error

±3.1 points

1.96 × √(0.5 × 0.5 ÷ 1,000) = 0.0310

52% ± the marginlanded insidemissed

01What the ± promises

Every one of those polls was done properly. They disagree only because each drew different people. The shaded band is the margin of error: the distance from the truth that about 95 polls in 100 stay within. That is all "plus or minus 3 points" under a poll of 1,000 means.

A real pollster runs one poll, not a hundred, and can't tell whether theirs is one of the unlucky few.

02Why pollsters stop at about 1,000

Ask more people and the band narrows, quickly at first and then hardly at all. The margin follows the square root of the sample, so cutting it in half takes four times as many interviews.

Fig. 2 · Margin of error by sample size95% confidence

From 250 to 1,000 to 4,000 people the margin goes 6.2, 3.1, 1.5. The first step turns a vague poll into a usable one. The second costs 3,000 more interviews and buys a point and a half.

03The size of the city barely matters

Nothing so far has mentioned how many people live in the city. Drag the population from a town of 20,000 to a country of a billion and watch the margin.

Fig. 3 · Same sample, bigger cityn = 1,000

20,0001 billion

Margin of error

±3.10 points

1 adult in 500 asked

A cook checks a pot of soup with one spoonful, however big the pot, as long as the soup is stirred. Picking people at random is the stirring. The city's size only starts to count when the sample is a real slice of it: 1,000 out of 20,000 is one adult in 20.

04What the margin leaves out

  1. A

    Who never got asked. The formula assumes every adult had the same chance of being picked. If the people who answer polls differ from the people who don't, a poll can miss by far more than its margin, and the margin won't show it.

  2. B

    The gap between two candidates. A margin belongs to one number. In a race at 48 to 45 with 1,000 people asked, the margin on the 3-point gap is about ±6.0, nearly double.

  3. C

    Small groups inside the poll. A fifth of the sample, voters under 30 for instance, is 200 people. Their answers carry a margin of ±6.9.

05Say it back

A margin of error says how far a properly random poll is likely to land from the truth: about 95 times in 100, no further than that.

It shrinks with the square root of the sample, so 1,000 people buy about ±3 points and 4,000 buy about ±1.5.

It says nothing about who was missed or how the question was put.

Method: 95% confidence, a simple random sample, and the worst case of an even split, so margin = 1.96 × √(0.25 ÷ n). For a city of N adults, multiply by √((N − n) ÷ (N − 1)). For a gap between shares a and b, use 1.96 × √((a + b − (a − b)²) ÷ n). The 100 polls are simulated in your browser: every answer is a yes with a 52% chance. Written by Rania Mansour for reporters at the Quayside Ledger, November 2026. The newsroom, the writer and the city are invented; the arithmetic is real.