How to read a political poll
Polls are snapshots, not predictions. Here's how to tell a real shift from statistical noise — in about five minutes.
Updated
What a poll measures
A poll interviews a sample — usually somewhere between 500 and a few thousand people — and uses it to estimate what a much larger group thinks. Pollsters weight their samples so that groups such as age, education, race and region match the population. Done well, a sample of about a thousand people can describe the views of millions.
The margin of error
A reported margin of error of ±3 percentage points means that if the survey were repeated many times, roughly 95 out of 100 results would fall within 3 points of the true value. Three things trip people up:
- Leads have a bigger margin. The margin on the gap between two candidates is roughly double the margin for each one. With ±3 per candidate, a lead must be about 6 points before it is clearly beyond sampling error. A smaller lead is what reporters call a "statistical tie."
- Changes between two polls are noisy. A 3-point move between two polls can easily be noise; Pew Research Center's example puts the margin on such a change at about ±8 points. A trend across several polls is far more convincing.
- Subgroups are much less precise. A subgroup of about 160 people in a 1,000-person poll carries a margin of roughly ±8 points — and ±16 on the gap between candidates.
Who was asked: A, RV and LV
Every poll on this site is labeled with its population:
- A — all adults. Everyone 18 and over. Common for presidential approval.
- RV — registered voters. People who say they are registered to vote.
- LV — likely voters. Registered voters the pollster expects to actually vote, identified with questions about their intention to vote, interest in the campaign and past voting. Because more people say they will vote than actually do, pollsters don't rely on stated intention alone.
Likely-voter screens get more accurate as Election Day approaches; months out, RV and LV results often look similar.
Approval, favorability and the generic ballot
- Approval — do you approve or disapprove of the way someone is handling their job?
- Favorability — do you have a favorable or unfavorable opinion of a person? It works for people in or out of office.
- Generic ballot — which party's candidate would you vote for in your congressional district, without naming candidates? It's a broad read of the national mood.
Outliers happen — on purpose
At a 95% confidence level, about 1 poll in 20 should land outside its margin of error by pure chance. A surprising result isn't necessarily wrong, but it should be confirmed by later polls before you treat it as a shift.
Errors the margin doesn't cover
The margin of error only measures random sampling error. Real polls also face:
- Coverage error — some people can't be reached by a given method (for example, people without internet access in an online poll).
- Non-response — the people who answer can differ from those who don't.
- Mode and wording effects — answers can change with a live interviewer, a different question order or slightly different wording.
- Likely-voter models — guessing who will turn out adds its own uncertainty.
Why averages beat single polls
Averaging many polls smooths out sampling noise and the house style of any one pollster. Our methodology uses a transparent 21-day rolling average of every poll that ended in the window — no weighting, no secret adjustments — and shows every individual poll as a dot so you can see the spread.
Polls vs. prediction markets
Polls ask voters what they think. Prediction markets such as Kalshi and Polymarket let traders buy contracts that pay $1 if an outcome happens, so a contract trading at 64¢ implies about a 64% chance. Markets move within minutes of news and fold in information polls can't (early-vote data, candidate quality, money), but they reflect traders' beliefs, can be thinly traded and have been wrong in big ways before. A useful habit: when the market and the poll average disagree, ask what the market knows that the polls don't — or whether it's just running ahead of the evidence.
Reading election-night results
- "Percent counted" is an estimate. It compares ballots counted with the total expected, and the expected total is itself a guess until every ballot is in.
- Watch which votes are left. States count early, mail and election-day ballots in different orders, so early leads can shrink or grow as the remaining batches come in.
- A race is "called" when the uncounted ballots can't mathematically change the outcome — not when counting is finished. Official certification comes weeks later.
- Some races go to extra rounds: Maine and Alaska use ranked-choice voting, and Georgia holds a runoff when no Senate candidate wins a majority.
A 30-second checklist
- Who conducted it, and who paid for it? Check the internal poll flag for campaign-sponsored polls.
- When was it in the field?
- Who was asked — A, RV or LV — and how many?
- Is the lead bigger than about twice the margin of error?
- Does the trend across many polls agree?
Frequently asked questions
How are betting-market odds different from polls?
A poll measures what voters say today. A prediction-market price (on Kalshi or Polymarket, for example) is what traders will pay for a contract that pays out if an outcome happens, so a 64¢ price reads as roughly a 64% chance. Markets react quickly to news but can be thinly traded; polls are slower but measure voters directly. They answer different questions, so this site shows both side by side.
Why do election results change after election night?
Ballots are counted in batches, and the order differs by state: some count early and mail votes first, others last. Mail ballots that arrive late or need signature checks can take days. That's why a leader at midnight can lose ground — and why outlets call a race only when the remaining uncounted votes can't change the outcome.
What does a margin of error of ±3 points mean?
It describes sampling error only. If the same survey were repeated many times, about 95 out of 100 results would land within 3 points of the true value for the whole population. It does not cover other errors such as who chose to respond or how questions were worded.
Is a 2-point lead outside the margin of error?
Usually not. The margin of error on the gap between two candidates is roughly double the margin for each candidate, so with ±3 points per candidate a lead needs to be about 6 points before it is clearly beyond sampling error.
What is the difference between likely voters and registered voters?
Registered-voter (RV) polls include everyone who says they are registered. Likely-voter (LV) polls narrow that to people a pollster expects to actually vote, using questions about intention, interest in the campaign and past voting. LV estimates usually matter most close to an election.
Why do two polls taken at the same time disagree?
Different samples, populations (adults, registered or likely voters), interview methods, question wording and weighting all move results. Random sampling error alone means about 1 poll in 20 will land outside its stated margin of error.
What is a polling average and why use one?
An average combines many recent polls so that the quirks of any single poll matter less. Ours is a simple 21-day rolling average of every poll that ended in the window, with no weighting or adjustments.
What is the generic congressional ballot?
A question asking which party's candidate people would vote for in their district's U.S. House race, without naming candidates. It is a broad gauge of the national partisan mood ahead of midterm and general elections.
What is the difference between approval and favorability?
Approval asks how someone is doing their job (for example, the president's job performance). Favorability asks whether people have a favorable or unfavorable opinion of a person, whether or not they hold office.
Sources and further reading
- Pew Research Center — Understanding the margin of error in election polls
- Pew Research Center — Determining who is a likely voter
- Pew Research Center — 5 tips for writing about polls
Polls are only as useful as our picture of who is answering them. See political myths and facts: what Americans get wrong about Republicans, Democrats, progressives and democratic socialists, with sources.