United States Presidential election7 Nov 2028in 717 days Senate D 99% House D 58% Calendar Track record Research

Since 2014, Democrats lost 19 of 51 three-week poll leads

Republicans with leads of the same size lost once in 32 races. Eleven Democrats hold leads in that range in our averages today.

Written in house from our poll data and our forecast.

Michigan State Capitol rotunda 2025d.jpg
Photo: Antony-22, CC BY-SA 4.0

Since 2014, Democratic Senate and governor candidates who led by 2 to 10 points three weeks before election day lost 19 times out of 51, or 37%. Republicans with leads of the same size lost once in 32 races, or 3%. Before 2014 the record ran the other way. From 2002 to 2012, Democrats lost 3 of 25 such leads and Republicans 4 of 19.

That matters now, because election day is three weeks away and our averages show eleven Democrats leading by 2 to 10 points. Four are in Senate races: Abdul El-Sayed in Michigan, Sherrod Brown in the Ohio special election, Peggy Flanagan in Minnesota and Chris Pappas in New Hampshire. Seven are running for governor: Amy Acton in Ohio, David Crowley in Wisconsin, Deb Haaland in New Mexico, Jocelyn Benson in Michigan, Katie Hobbs in Arizona, Rob Sand in Iowa and Amy Klobuchar in Minnesota.

What we counted

Our archive holds the final poll each pollster published in every Senate and governor race since 2000, matched to the certified result. For this piece we took the independent polls whose fieldwork ended 15 to 28 days before election day, averaged them for each race, and kept the races where one side led by 2 to 10 points. Then we asked a plain question: did the leader win. Odd-year governor races in Virginia, New Jersey, Kentucky and elsewhere are included.

The gap is too large to be luck. With 51 Democratic leads and 32 Republican ones, a split of 19 losses against 1 would arise by chance about 4 times in 10,000.

It is not that Democratic leads were thinner. Since 2014 the results ended 3.4 points more Republican than the three-week average whichever party led. The same miss that turns a narrow Democratic lead into a defeat simply widens a Republican win. From 2002 to 2012 the results ended 1.0 points more Democratic than the average where Democrats led, and 1.4 points more Republican where Republicans led. Our accuracy page tells the same story by cycle: statewide polls overstated the Democrats in 2014, 2016, 2020, 2022 and 2024, and understated them in 2010 and 2012.

The races that got away

The list of lost Democratic leads is a tour of the last decade's disappointments for the party. In Wisconsin in 2016, Russ Feingold led Ron Johnson by 8.5 points across six polls three weeks out and lost by 3.4. In Iowa in 2018, a University of Iowa poll put Fred Hubbell 8.6 points ahead of Kim Reynolds; she won by 2.7. In Maine in 2020, a Pan Atlantic Research poll had Sara Gideon 7 points ahead of Susan Collins, who won by 8.6. In Pennsylvania in 2024, two polls three weeks out had Bob Casey ahead by 7.5; Dave McCormick won by 0.2. Brown himself is on the list: a Bowling Green State University poll had him 2.2 points ahead in 2024, and he lost by 3.6.

The single Republican lead lost since 2014 was in Nevada in 2016, where Joe Heck led Catherine Cortez Masto by 2.4 points across four polls. She won by 2.4.

The smallest leads were the most fragile. Since 2014, Democrats leading by 2 to 4 points lost 7 of 17 races. Those leading by 6 to 10 lost 7 of 24, still more than one in four.

The exception and the earlier years

The pattern is not a law. In 2022 none of the six Democrats who led by 2 to 10 points three weeks out lost: John Fetterman in Pennsylvania, Catherine Cortez Masto in Nevada and Raphael Warnock in Georgia among them, along with governors Tim Walz, Kathy Hochul and Tina Kotek. Our accuracy page puts the average statewide miss that year at 1.1 points toward the Democrats, the smallest of the recent pro-Democratic misses. Scott Keeter, senior survey adviser at Pew Research Center, told Axios on 10 October that midterm polls have been more accurate than presidential ones, though some underestimated Republicans in 2022.

The earlier record matters too. From 2002 to 2012, Republican leads were the ones that slipped: George Allen's 2-point edge in Virginia in 2006, Ken Buck's in Colorado and Denny Rehberg's in Montana, and Rick Berg's 8-point lead in North Dakota in 2012, which Heidi Heitkamp overturned. The Democratic leads lost in those years included Betty Castor in Florida in 2004 and Jon Corzine in New Jersey in 2009. So this is not a fixed property of American polling. It has run in one direction for a decade, after running mildly in the other.

How firm is it

We checked the obvious weak points. One race in the analyst's first pass was miscoded: the 2004 Washington governor's race, which Christine Gregoire won by 129 votes, sat in the data as a dead heat and was scored as a Democratic loss. We corrected it. The 2017 Virginia governor's race was flagged for a Hampton University poll that had Ed Gillespie 8 points ahead of Ralph Northam 16 days out; WTVR reported that poll at the time, and Northam won by 9. It is a real poll, and at a lead of 10.8 points on our scale it falls just outside the range counted. A 2010 Massachusetts poll showing Charlie Baker 7 points ahead of Deval Patrick could not be checked against its source. It came from the Hart Research and Public Opinion Strategies partnership, which our archive now files under Public Opinion Strategies, a firm that works mainly for Republicans, so it is no longer counted among the independent polls. Counting it would make the 2002 to 2012 Republican record 5 losses in 20.

The result survives other cuts. Senate races alone: Democrats lost 10 of 26 leads since 2014, Republicans 1 of 22. Governors alone: 9 of 25 against none of 10. Counting only polls whose figures we checked against the pollster's own release, Democrats lost 7 of 27 and Republicans 1 of 22. Using our comeback calculator's poll averages 30 days out instead of single final polls, Democrats lost 16 of 67 leads since 2014 and Republicans none of 37. Adding presidential races by state gives 25 lost Democratic leads in 71 against 1 in 46.

There are limits. The archive keeps each pollster's last poll, so a race's three-week reading comes from pollsters who stopped early, not from everything published at that point. The counts are small; six cycles of midterm and presidential races is not a long record.

What it means for 3 November

Our averages already take 0.7 points off every Democratic margin, half the average statewide miss, and remove each pollster's usual house effect. Even so, our forecast gives these eleven Democrats chances of 64% to 94%, which adds up to about two expected defeats. The 2014 to 2024 record, applied to the same eleven races, would mean about four.

The five closest are where the history bites hardest. Crowley leads by 2.8 points in our average, El-Sayed by 3.0, Acton by 3.5, Brown by 3.6 and Flanagan by 3.8. Fox News had El-Sayed ahead by a point in early October; CNN and SSRS had Brown 6 points up, Forbes reported. Our forecast gives them 64% to 77%.

Others have looked at the same problem from another side. CNN's Harry Enten has pointed out that Republicans outperformed their September polls in Iowa, Ohio and Michigan by at least 5 points in each of the last few cycles, and the Republican analyst Patrick Ruffini found about 93% of public polls in those states overstated Democrats a month out, the Washington Examiner reported. What our count adds is the conversion rate: how often a lead of a given size actually turned into a defeat, and how lopsided that has been between the parties since 2014.

Newsweek noted that pollsters have adjusted their methods since past misses, and the 2022 midterms show a cycle can break the pattern. If 2026 looks like 2022, all eleven leads could hold. If it looks like 2014, 2016, 2020 or 2024, several will not.

Polling Forecast reads every poll in ten countries, corrects for how pollsters have missed before, and turns them into seats. Our forecasts, the polls and our methods are free to explore.

Our projection and every poll →

Elsewhere on Polling Forecast

Figures as of . Published by Polling Forecast.