Methods

Published by Onward Media LLC · Questions: support@peoplespoll.com

---

1. What we publish

For each Senate and governor race we track, and for presidential job approval and the generic congressional ballot, we publish:

• a polling average built from published polls,
• a trend line showing the average day by day over up to the last 180 days, with a dot for each poll, and
• the list of every poll in the average, with its pollster, sponsor (if any), field dates, sample size, population and a link to the pollster's release.

We don't publish forecasts, win probabilities or odds.

House races. For House races we list every published 2026 general-election poll, district by district, on the House page. We don't average them, because most districts have only one or two polls. Each listing shows the pollster's numbers for the named candidates, the field dates, sample size, population, sponsor (marked (D) or (R)) and a link to the release. Undecided and other answers are left out. Where a poll asked more than one matchup, each is shown. A House poll is listed if its dates are real and not in the future, it names a pollster, and its sample size, if reported, is between 100 and 100,000.

2. Where the polls come from

Poll data comes from published polls compiled by The New York Times (nytimes.com/interactive/polls), used under the Creative Commons Attribution 4.0 license. The averages are ours, not the Times's. We check the Times's poll files every three hours.

A new poll goes into the average automatically if it passes every check below. Otherwise it's held for a person to review:

• it matches a race we track,
• its dates are real and not in the future,
• it names a pollster and links to an https source,
• its sample size, if reported, is between 100 and 100,000,
• results are present for every candidate or answer, and they add up to between 60% and 101%.

If the same pollster publishes the same poll twice (same end date and population), it counts once.

3. Senate and governor averages

Each race's average is a weighted mean of all of its published polls. Older polls fade quickly, so in practice recent polls drive the average. A poll's weight is the product of three things:

1. Recency. A poll's weight halves every 14 days after its fieldwork ends. If a race has fewer than three polls, the half-life is 28 days instead.
2. Sample size. Weight grows with the square root of the sample size, capped at 3,000 respondents, so a very large poll can't dominate. A poll that doesn't report its sample size is treated as 400.
3. Unusual results. When a race has at least three polls, any poll whose margin is more than two median absolute deviations from the median margin counts at 30% of its normal weight.

Sponsored polls. A poll paid for by a party or a campaign is shifted 2 percentage points toward the other side before averaging (the Democratic candidate's share down 2 and the Republican's up 2 for a Democratic-sponsored poll, and the reverse for a Republican-sponsored one). Published research finds sponsored polls tend to favor their sponsor. These polls are marked (D) or (R) everywhere they appear.

4. Approval and generic-ballot trackers

The national trackers use a 28-day rolling average. Each day's value is the sample-size-weighted mean (square root of the sample size, capped at 3,000) of the polls whose fieldwork ended in the previous 28 days, with the same treatment of unusual results as above. For the generic ballot, sponsored polls are shifted 2 points as described in section 3. Approval polls are not shifted, because approval questions have no party side.

5. Dates on the site

"Updated" shows the date the averages were last recalculated. "Newest poll" shows the end date of the most recent poll in our data. The averages are recalculated every day, so a poll's weight keeps declining as it ages, even when no new poll arrives.

6. Corrections and links

We check every poll's source link every few days. If a pollster has taken a page down, we mark it "Link unavailable" rather than sending you to an error. If you spot a mistake, email support@peoplespoll.com with the pollster's name and a link. We correct errors promptly. If a correction changes an average, the page updates the next time the averages are recalculated.

7. People's Poll (coming soon)

When we open the People's Poll, readers will be able to answer the same questions the pollsters ask. Those answers will be clearly labeled as a self-selected, unweighted group of readers, not a representative sample. Before that happens, we'll publish the full methods for collecting, counting and displaying them here.

8. Independence

People's Poll isn't affiliated with any party, campaign, pollster, news organization or government agency. We don't accept payment from campaigns, parties or political action committees.