Why we removed 48 polls from our tracker, and how we decide what to publish
Dated editorial analysis; figures in this article are not recalculated automatically. Visit the linked race pages for current data.
On October 7, we removed 48 polls from the Midterm Results Live dataset. None of them were necessarily wrong. We removed them because we could not link each one to an original public release from the pollster or from a named media organization that sponsored it.
The result was a smaller tracker. Before the change, 26 races had at least one recent poll. After it, 16 did. Some race pages on this site now say "no verified public poll" where they used to show numbers.
We think that trade-off is the right one, and this piece explains why, along with the rules we follow for every poll we publish.
Our rule: every poll needs a source you can open
A poll goes into our ledger only if we can link it to one of the following:
- The pollster's own release, such as a press release, a topline PDF or a results page on the pollster's website.
- A report from a named news organization that sponsored the poll or published its results directly.
For each poll, we record the pollster, the end date of the fieldwork, the candidates and their percentages, and the link to the source. You can see the link next to every poll on our latest polls page.
Why we are strict about it
Poll numbers spread quickly. A result can be repeated on social media, then in a blog post, then in an aggregator, each time losing a little context. By the third or fourth copy, it can be hard to tell who conducted the poll, when it was taken, how many people were surveyed, or whether the numbers were reported correctly.
Some "polls" that circulate online are not polls at all. They can be internal campaign surveys released selectively, online click-in surveys with no sampling method, or simple errors copied from one site to another. If we cannot find the original, we cannot rule any of those out.
Requiring a source protects readers in three ways:
- You can check us. If we have made a typing error, the original release will show it.
- You can judge the pollster. Knowing who ran a poll lets you consider its track record and methods.
- You can see the details we do not show. Sample size, field dates, question wording and margin of error are usually in the original release.
Why some races show no polls
Many of the 35 Senate and 36 governor races on the ballot this year are not expected to be competitive. Pollsters and news organizations spend their budgets on close races, so a race rated Solid Republican or Solid Democratic may not get a single public poll all year.
When that happens, our race page says so plainly, instead of showing an estimate or a number from a source we cannot verify. For those races, the forecaster ratings from the Cook Political Report and Sabato's Crystal Ball are often the best available guide.
How we handle polls that disagree
We do not adjust or "correct" any pollster's numbers. When two polls of the same race disagree, we show both, each with its own source link.
When a poll is shown on the site but left out of an average, we mark it "not in averages" so readers can see the difference.
Corrections
When we find a mistake, we fix it and update the date on the affected page. If you spot something that looks wrong, please tell us through our contact page and include the link to the original source if you have it.
How to check any poll yourself
Whether you read a poll here or anywhere else, these five questions will tell you most of what you need to know:
- Who conducted it, and who paid for it? A poll paid for by a campaign or party should be read with more caution.
- When was it taken? A poll from three weeks ago may not reflect the race today.
- Who was surveyed? Adults, registered voters and likely voters can give different results.
- How many people were surveyed? Smaller samples mean larger margins of error.
- Where is the original? If no one can point you to it, treat the number as unconfirmed.
Our full approach is on the methodology page, and our editorial principles are on our independence page.