Answers · Voters and issues

What does margin of error mean in a poll?

By Mert Yerekapan, maker of Midterms 2026 · Updated 2026-10-28

The margin of error is how far a poll's number could be from the truth because of random sampling alone. If a poll says 50% support something with a margin of error of plus or minus 3 points, the real figure is probably between 47% and 53%. "Probably" usually means 95% of the time.

Why there's a margin at all

A poll talks to a sample, not everyone. Pick a different random sample and you'd get slightly different numbers. The margin of error measures that wobble. Bigger samples wobble less, which is why a poll of 2,000 people has a smaller margin than a poll of 500.

Rough rule of thumb: a sample of about 1,000 people gives a margin near 3 points. To cut the margin in half, you need about four times the sample.

The part people miss: it applies to each number

Say a poll shows Candidate A at 48% and Candidate B at 45%, with a margin of 3 points. That looks like a 3-point lead. But each number has its own wobble, and the gap between them can wobble about twice as much as a single number. So a 3-point lead with a 3-point margin is, in plain terms, too close to call.

I think this is the single most useful thing to know when reading headlines. "Leads by 2" in a poll with a 3-point margin is mostly a tie.

What the margin does not cover

The margin of error only describes random sampling error. It doesn't capture the other ways a poll can miss:

So the real uncertainty in a poll is usually larger than the published margin. Think of it as a minimum, not a guarantee.

Subgroups are noisier

If a poll surveys 1,000 people but only 100 are young voters, the margin for young voters alone is far bigger, often around 10 points. A headline about one small group in a national poll deserves extra caution.

A quick way to read polls

Where this shows up in the game

In Midterms 2026, my free iPhone election strategy game, you get a live "House if the vote was today" projection. It isn't a poll and has no margin of error, but the lesson carries over: a state that's barely leaning your way isn't safe. I'd spend my actions on the close states first, because that's where a small push flips a result.

Related questions