The biggest miss in 40 years: Biden's shrinking lead in 2020
Abstract. The final averages had Joe Biden 7.9 points ahead and he won the popular vote by 4.5, the largest national polling error in 40 years by AAPOR's count, with state polls missing by even more. AAPOR traced at least part of it to nonresponse: the people who answered polls were more Democratic than the people who voted, in ways demographic weighting could not fix.
Summary
The final national polling averages put Joe Biden 7.9 points ahead of Donald Trump. He won the popular vote by 4.5 points (51.3% to 46.9%), so the averages overstated his lead by 3.4 points, and AAPOR's task force found the error in national polls was the largest in 40 years. Biden still won, 306 electoral votes to 232, but state polls missed by even more: Wisconsin by 7 points on our numbers, and Florida and North Carolina went to Trump after polls showed Biden ahead. The cause was not late deciders or missing education weights but nonresponse: the people who took polls were more Democratic than the people who voted, in ways that ordinary weighting did not fix.
What the polls said
Almost every national poll showed Biden ahead by between 4 and 11 points. The final polls from the larger pollsters in our database:
| Pollster | Fieldwork end | Biden | Trump | Lead |
|---|---|---|---|---|
| Ipsos/Reuters | 2 Nov | 52 | 45 | Biden +7 |
| YouGov/Economist | 2 Nov | 53 | 43 | Biden +10 |
| Research Co. | 2 Nov | 50 | 42 | Biden +8 |
| IBD/TIPP | 2 Nov | 50 | 46 | Biden +4 |
| Quinnipiac | 1 Nov | 50 | 39 | Biden +11 |
| Rasmussen | 1 Nov | 48 | 47 | Biden +1 |
| SurveyMonkey/Axios | 1 Nov | 52 | 47 | Biden +5 |
| Morning Consult | 31 Oct | 52 | 44 | Biden +8 |
| NBC News/Wall Street Journal | 31 Oct | 52 | 42 | Biden +10 |
| Fox News | 29 Oct | 52 | 44 | Biden +8 |
The election-eve averages were Biden +7.2 at RealClearPolitics (51.2 to 44.0), +8.0 at 270toWin (51.1 to 43.1) and +8.4 at FiveThirtyEight (51.8 to 43.4). Their mean was Biden 51.4, Trump 43.5, a lead of 7.9 points.
The swing-state averages had Biden ahead in every state Hillary Clinton lost narrowly in 2016, and also in Florida, North Carolina, Arizona and Georgia. On our final-poll averages, he led by 8.0 points in Wisconsin, 7.7 in Michigan and 5.1 in Pennsylvania.
What happened
The Federal Election Commission's official totals gave Biden 81,268,924 votes (51.31%) and Trump 74,216,154 (46.86%), out of 158.4 million cast. Biden won the Electoral College 306 to 232.
Biden's share was almost exactly what the averages said: 51.4 in the polls, 51.3 in the count. Trump's was understated by 3.4 points, 43.5 against 46.9. The whole error was on one side. Undecided voters were few (AAPOR counted only 4% of state-poll respondents giving an answer other than Biden or Trump), so this was not a question of how undecideds split. The polls had too few Trump voters.
In the states, Biden won Wisconsin by 0.6 points, Pennsylvania by 1.2, Arizona by 0.3 and Georgia by 0.2. Trump won Florida by 3.4 and North Carolina by 1.3, both states where the polls had Biden ahead.
Why it happened: what the pollsters found
AAPOR's Executive Council set up its Task Force on 2020 Pre-Election Polling in October 2019, with 19 members chaired by Josh Clinton of Vanderbilt University. It collected more than 2,800 public polls and reported in July 2021. Its findings:
- The error was historically large. It was "the highest in 40 years for the national popular vote and the highest in at least 20 years for state-level estimates." In the final two weeks, national polls missed the margin by 4.5 points on average and state presidential polls by 5.1.
- It ran one way. The signed error was 3.9 points toward Biden in national polls and 4.3 in state polls. In Senate and governor races it was worse, 6.0 points toward Democrats.
- It did not fade. The error was the same for polls in the last two weeks, the last week and the last three days.
- It was larger in Trump country. In states Trump won by more than 5 points in 2016, polls overstated Biden's margin by 5.3 points; in states Clinton won by more than 5, by 3.5.
The task force ruled several things out: late deciders (they split evenly in the exit polls), a lack of education weighting (92% of state polls that disclosed their method weighted on it), the demographic mix of the electorate, "shy" Trump voters lying to interviewers (the error was bigger in races without Trump on the ballot), and the balance of early and Election Day voters.
What was left was unit nonresponse. Reweighting pollsters' data to the actual result needed more Republicans and more 2016 Trump voters, and barely changed the demographic weights. Even weighting to partisanship and 2016 vote did not fix it, which points to the Republicans who answered polls being more Democratic-leaning than the Republicans who did not. The task force suggested that Trump's attacks on polls as "fake" may have turned answering a survey into a political act, but said it could not prove this without knowing how non-respondents voted. New voters, of whom there were many in 2020, were a second possible source.
What our data shows
Our database holds 38 national pollsters' final polls from the last two weeks. Their average put Biden 8.1 points ahead, 3.7 points above the certified margin, and 33 of the 38 overstated his lead. The few that came close, such as Rasmussen and IBD/TIPP, had also been among the most Trump-friendly national polls in 2016. This was an industry-wide miss, not a few bad firms.
The state errors were broad. Taking each pollster's final poll in each state over the last three weeks, polls in 47 states overstated Biden's margin by 4.7 points on average, with an average absolute error of 5.1. Polls overstated the Democrat in 42 of the 47.
| State | Final-poll average | Result | Error on lead |
|---|---|---|---|
| Wisconsin | Biden +8.0 | Biden +0.6 | 7.3 |
| Iowa | Trump +1.1 | Trump +8.2 | 7.1 |
| Ohio | Trump +1.9 | Trump +8.0 | 6.1 |
| Florida | Biden +2.2 | Trump +3.4 | 5.6 |
| Michigan | Biden +7.7 | Biden +2.8 | 4.9 |
| Texas | Trump +1.3 | Trump +5.6 | 4.2 |
| Pennsylvania | Biden +5.1 | Biden +1.2 | 4.0 |
| North Carolina | Biden +1.6 | Trump +1.3 | 3.0 |
| Arizona | Biden +2.7 | Biden +0.3 | 2.4 |
| Georgia | Biden +1.6 | Biden +0.2 | 1.4 |
Compared with 2016, the Midwestern pattern repeated almost exactly: Wisconsin was off by 7.2 points in 2016 and 7.3 in 2020, Iowa by 7.0 and 7.1, Ohio by 5.5 and 6.1. Education weighting had been widely adopted in between, and the same states missed by the same amounts. The 2016 explanation was at most part of the story. Something about who answers polls in those places had not been fixed.
What the pollsters changed
After 2020, pollsters moved to two kinds of change: political weighting and new ways of reaching people.
Weighting on party and past vote. Because demographic weights did not catch the problem, more firms began to weight on self-reported party identification or on how respondents said they voted in 2020. In AAPOR's survey of 39 firms for its 2024 report, 68% said they always used a political variable (party, past vote or turnout history) in their weights, 32% always weighted on past vote and another 42% sometimes did. Pew Research Center, which had weighted its American Trends Panel on party since 2014, added the 2024 presidential vote to its weights from June 2025, saying its polls had underestimated Trump in 2016, 2020 and 2024 even after the party adjustment.
Sampling and mode. Pew's 2023 study of 78 polling organizations found that 37% changed how they sampled or interviewed people between 2020 and 2022, against 26% after 2016. The share using several methods rose to 39%, nine pollsters (13% of those active) began texting respondents, and 23 used probability-based panels, up from 7 in 2016. More firms drew samples from voter files, which let them check respondents' turnout history and modeled party.
What it means for our forecast
2020 is the strongest evidence in our US history for a systematic lean, and the largest single entry in the miss record our model replays. We correct polls by half the average past miss, and keep that correction only if it would have improved past forecasts made for each election without knowing its result. After 2016, 2020 and 2024, that rule moves our polling averages a little toward Republicans. It moves them only a little: polls overstated the Republican in 2012 and, slightly, in 2022, and a full correction fitted to the latest cycles would have made those forecasts worse.
The more important lesson is the size of the tail. A 3- to 4-point national miss, shared by almost every state, is not rare in our simulations: it happened in 2020, four years after the industry had diagnosed and fixed its last miss. Our 5,000 simulated elections draw on the actual pattern of past state errors, so a year in which Wisconsin, Michigan and Pennsylvania all miss toward the Republican together is built in. The method is in our US simulation paper, and each firm's 2020 record feeds our US pollster ratings.
Sources
- AAPOR: Task Force on 2020 Pre-Election Polling: An Evaluation of the 2020 General Election Polls (2021)
- AAPOR: Task Force on 2024 Pre-Election Polling (October 2025)
- Federal Election Commission: Official 2020 Presidential General Election Results
- Pew Research Center: A conversation about U.S. election polling problems in 2020 (21 July 2021)
- Pew Research Center: How Public Polling Has Changed in the 21st Century (19 April 2023)
- Pew Research Center: Why and how we're weighting surveys for past presidential vote (23 July 2025)
- Wikipedia: Nationwide opinion polling for the 2020 United States presidential election (aggregator table, cross-check)
How to cite
Polling Forecast (2026). The biggest miss in 40 years: Biden's shrinking lead in 2020. Retrieved 2 November 2026 from https://www.pollingforecast.com/research/missed-us-2020. Live figures change as new polls arrive; the date above is when the text was last revised.