How to Read an Election Poll: Margin of Error, Likely Voters and Averages
By the Midterm Results Live team · · Guide
Every week brings dozens of new polls, and headlines often treat each one as news. A few basic ideas make it much easier to tell a real shift from normal noise.
1. The margin of error is about each candidate, not the lead
A poll of about 800 voters typically reports a margin of error near ±3.5 points. That figure applies to each candidate's share. The uncertainty on the gap between two candidates is larger, roughly 1.7 to 2 times as big. So a 3-point lead in a poll with a ±3.5-point margin of error is not a statistically clear lead. That is why our race pages label margins under about four points as "within the margin of error."
2. Sampling error is not the only error
The published margin of error covers only random sampling. It does not cover people who refuse to answer, voters who change their minds, or mistakes in guessing who will actually vote. In practice, the total error in a single poll is usually larger than the printed number. Historically, final Senate polls have missed the result by several points on average.
3. Who was asked matters: adults, registered or likely voters
Polls of all adults include people who cannot or will not vote. Registered-voter (RV) polls are narrower, and likely-voter (LV) polls try to estimate who will turn out. The same poll can show different results by screen. In a recent Fox News Texas poll, for example, James Talarico led Ken Paxton 53% to 46% among registered voters but 51% to 49% among likely voters. When a pollster reports both, we log the likely-voter figure.
4. Check who paid for it
Campaigns and party committees release "internal" polls, usually when the results help them. These are not necessarily wrong, but they are released selectively. Some independent pollsters also lean consistently toward one party compared with the average, an effect called "house effect." We label each poll with the pollster name so you can judge for yourself.
5. Averages beat any single poll
Individual polls bounce around. Averaging several recent polls cancels out much of the random noise and some of the house effects. Our averages use every public poll for a race that ended within 14 days of the newest one. When a race has only one or two recent polls, the average is fragile, and we show the number of polls next to it.
6. Polls are snapshots, not forecasts
A poll tells you about opinion during the days it was in the field. Late-deciding voters, turnout and news events can still move a race. Treat a 2-point lead in October as "close," not as a prediction.
Quick checklist
- Is the lead bigger than about twice the margin of error?
- Likely voters or registered voters?
- Independent pollster or a campaign internal?
- Does it agree with the other recent polls, or is it an outlier?
- How many polls are in the average?
You can apply this checklist to any race on our latest polls page.