National

Trusting the polls

How far can we trust political polls?

Information checked against its sources on October 21, 2026.

First drafted by Claude, an AI model. Edited and reviewed for balance by Leonard Dalenberg. How this site is written

Side A

Good polls are reliable, if read correctly.

Supporters of this side say that well-run polls rest on sound statistics and usually land close to the result. The problems, they argue, come mostly from treating a single poll as certain and ignoring its stated uncertainty.

2024 showed it. The profession's own review found public polls largely accurate in 2024. State presidential polls were the most accurate in any presidential year since 1944.
Midterms come close. In 2018 and 2022 the average poll leaned toward one party by less than a point.
The math works. A truly random sample of about 1,000 people describes the whole country to within about 3 points, 95 times out of 100.
Fewer answers need not mean wrong answers. Pew studies in 1997, 2003, 2012 and 2016 found little link between how many people respond and how accurate a poll is.

On this view the fix is better reading of polls, not less polling: look at averages, respect the margin of error, and favor pollsters who show their methods.

Side B

Polls are too shaky to lean on.

Supporters of this side say polling has grown harder and less dependable, and that the confident numbers reported every day hide how much guesswork lies behind them.

The same miss, three times. Polls underestimated Republicans in 2016, 2020 and 2024. In 2020 the national error was the highest in 40 years.
Hardly anyone answers. The response rate for a typical phone survey fell from 36 percent in 1997 to 6 percent in 2018.
Many polls are not random. Online polls that anyone can join attract people who answer for rewards. In one, 12 percent of adults under 30 claimed to be licensed to run a nuclear submarine.
The real error is bigger than stated. The margin of error covers only chance. It leaves out who refuses to answer, how questions are worded and who is guessed to vote.

On this view polls are treated as news far beyond what they can bear, and may shape the very races they claim to measure.

Other views

Trust the average, not the poll. Many analysts say a single survey means little, but an average of many good ones is far steadier.
Judge the pollster. Others point to ratings that score firms on past accuracy and openness about methods, and would ignore polls from firms that publish neither.
Good on issues, weak on close races. Some say polls are useful for knowing what the public broadly thinks, but cannot reliably call an election decided by a point or two.
Hold back late polls. Others would bar publishing new polls in the final days before an election, so that they cannot sway turnout.

Factual review

What is established

How sample size works. A simple random sample of 1,067 people gives a margin of error of about 3 points either way for each candidate. The margin on the gap between two candidates is about twice as large, so a 5-point lead could really be anything from −1 to +11. A group making up 15 percent of adults would have only about 160 people in such a poll, with a margin of about 8 points, and 16 on the gap.¹
What the margin leaves out. The stated margin of error covers only the luck of the draw. Weighting, people who refuse to take part, question wording and guesses about who will vote add further error that is hard to measure and rarely reported.¹
Response rates. Pew's typical telephone survey reached a 36 percent response rate in 1997, 9 percent in 2012 and 6 percent in 2018. Pew found weighted phone surveys still accurate on most political measures, with one known exception: people who take surveys are more civically active than those who do not.^2,3
The record. Across 611 polls in the last two weeks of the 2024 campaign, the average miss on the margin between the two parties was 3.3 points, down from 5.3 in 2020 and 5.2 in 2016. National presidential polls missed by 2.6 points and state presidential polls by 3.0. Polls overstated the Democratic margin by 2.7 points in 2024, 4.6 in 2020 and 3.1 in 2016, but by little in the midterms: they overstated Republicans by 0.6 points in 2022 and Democrats by 0.1 in 2018. Since the 1930s, each party has been underestimated about equally often.⁴
2020. The national popular-vote error was the highest in 40 years, and state errors the highest in at least 20. The profession's review found the cause was not late deciders, not a failure to account for education, and not Trump voters hiding their views. It pointed to differences between the people who took polls and those who did not.⁵
Opt-in online polls. In Pew's tests, opt-in respondents gave implausible answers far more often than people chosen at random. In one opt-in poll, 20 percent of adults under 30 agreed that the Holocaust is a myth; Pew's random-sample survey with the same question found 3 percent.⁶
Pollster ratings. The Silver Bulletin ratings, the successor to the FiveThirtyEight ratings, scored more than 500 pollsters on accuracy and openness in January 2026. The Washington Post, Marquette University and the New York Times/Siena College were at the top.⁷

What is disputed

Why do polls keep missing Republicans? The 2024 review found three groups were hard to capture: Republicans in Republican-leaning areas, Hispanic voters, whose Democratic support was overstated, and people who had not voted in 2020, who leaned Republican and turned out in larger numbers than polls expected. Weighting by how people voted last time can help, the review found, but doing it well is not straightforward.⁴
Does method matter? Pew's tests show opt-in panels produce more nonsense answers. Yet the 2024 review found that most major choices of method showed little relationship to how far a poll missed, and that no single recipe guarantees accuracy.^4,6
Were partisan pollsters better? Republican-affiliated pollsters were slightly more accurate in 2024. The review cautioned that this may reflect a habit of releasing Republican-friendly numbers, which helps only in years when the average poll leans Democratic.⁴
Do polls change the outcome? Critics say a candidate shown far ahead or behind can lose donors and volunteers, or discourage turnout. Others reply that voters see many conflicting polls and decide for their own reasons.

Where it comes down to values

Snapshot or forecast. Should a poll be reported as a picture of opinion today, or as a prediction of the result?
The press's duty. How much uncertainty should reporters spell out before a poll becomes a headline?
Who may poll. Should polls paid for by campaigns or parties be treated the same as independent ones?
Timing. Is publishing polls up to election day part of a free press, or an influence on the vote that should be limited?

Sources

Pew Research Center, “Understanding the margin of error.” pewresearch.org
Pew Research Center, “Assessing the Representativeness of Public Opinion Surveys,” May 15, 2012. pewresearch.org
Pew Research Center, “Response rates in telephone surveys have resumed their decline,” Feb. 27, 2019. pewresearch.org
American Association for Public Opinion Research, Task Force on 2024 Pre-Election Polling, executive summary, 2025. aapor.org
Pew Research Center, “A conversation about U.S. election polling problems in 2020,” July 21, 2021, with the chair of the profession's 2020 review. pewresearch.org
Pew Research Center, “Online opt-in polls can produce misleading results, especially for young people and Hispanic adults,” March 5, 2024. pewresearch.org
Marquette Today, “Marquette Law School Poll again ranked No. 2 in annual national pollster ratings,” Feb. 9, 2026 (Silver Bulletin ratings, January 2026). marquette.edu

← All issues

Reader comments

No comments yet. Respond to either side, add a view the page missed, or challenge a fact with a source.

Commenting is open to account holders. Accounts are approved by the editor, and comments appear under your real name.

Request an account Sign in to comment