How do I know if a poll is reliable?

Check the sampling mode (probability samples and live phone beat opt-in panels), sample size, whether weighting variables and crosstabs are disclosed, the pollster's public track record, and whether the sponsor is disclosed. A brand-new pollster with one prior poll, no crosstabs and an undisclosed sponsor should be treated with caution.

Poll Credibility Checker

Credibility scorer

Enter what the pollster actually discloses. The worked example below is prefilled with an n=800 online opt-in panel from a pollster with one prior public poll.

—

Fill in the form and press "Score this poll" to see a 0-100 credibility score with factor-by-factor reasoning.

Live margin-of-error calculator

MOE = 1.96 · √(p(1−p)/n) at 95% confidence. Watch repeated random samples pile into a sampling distribution.

Sample size: 800Proportion p: 50%

Stated margin of error

±3.5 pts

Opt-in online panels are not random samples, so their true error is typically 1.5-2x the stated MOE. Treat the published number as a floor, not a ceiling.

How polling actually works

Four concepts that separate a serious survey from a press release with numbers.

Online opt-in panels vs probability sampling

Probability sampling starts from the full population and randomly selects respondents, so everyone has a known chance of inclusion. Opt-in panels recruit volunteers online — the people who join are systematically different from those who do not, and weighting can only partially fix that.

Weighting

Pollsters adjust raw responses so the sample matches known population benchmarks — age, race, education, and often recalled past vote. Education weighting became critical after 2016; a pollster that skips it (or hides its weights) is a warning sign.

House effects

A house effect is a pollster's consistent lean relative to the polling average — from panel composition, likely-voter models, or weighting choices. With only one released poll, you cannot estimate a new pollster's house effect at all.

Herding

Herding is when pollsters nudge results toward the existing average to avoid standing out. Late-cycle polls that all cluster suspiciously tightly are less independent — and less informative — than they appear.

Red flags for brand-new pollsters

None of these alone proves a poll is fake — together they say "wait for corroboration."

Only one prior poll. There is no track record to measure accuracy or house effects against. Wedgwood's single Louisiana runoff release is exactly this case.
No crosstabs. If you cannot see subgroup results, you cannot sanity-check whether the internals are plausible.
Undisclosed sponsor. Anonymous funding is a classic vehicle for campaign-aligned narrative polls.
Appears right before an election. New "pollsters" often materialize in the final weeks to flood averages or shape coverage, then vanish.
A slick website is not evidence. Legit-looking branding costs a weekend; a methodology statement and released data cost credibility.