3 elections where the final polls got the result badly wrong. Each deep dive sets out what the polls said, what happened, what the official inquiries and the pollsters found, what our own data shows, and what the pollsters changed afterwards. Every one of these misses is in the record our forecasts learn from.
How far out the final polls were
The error on the lead between the top two: the final polls' average against the result, in points.
Harris 48.5, Trump 47.7: Harris +0.7 (mean of six final aggregator averages)
Result
Trump 49.8, Harris 48.3: Trump +1.5
The final averages had Kamala Harris 0.7 points ahead and Donald Trump won the popular vote by 1.5, a 2.2-point miss that was smaller than in 2020 but the third presidential election running in which polls understated Trump. Weighting on party and recalled 2020 vote tightened the polls, but they still missed Republicans in Republican areas, the swing of Hispanic voters to Trump and new voters.
Biden 51.4, Trump 43.5: Biden +7.9 (mean of three final aggregator averages)
Result
Biden 51.3, Trump 46.9: Biden +4.5
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.
Clinton 47.3, Trump 43.3: Clinton +4.0 (mean of six final aggregator averages)
Result
Clinton 48.2, Trump 46.1: Clinton +2.1
National polls had Hillary Clinton about 4 points ahead and she won the popular vote by 2.1, a normal-sized miss, but state polls overstated her by 5 to 7 points in Wisconsin, Michigan and Pennsylvania, all of which Donald Trump won. AAPOR blamed late deciders breaking for Trump and state polls that did not weight by education.
For each country we measure how far the final polls missed at past elections. We take half the average miss off today's polls only where that would have made past forecasts better, testing each election without knowing its result; elsewhere the polls stand as published. Every forecast then runs thousands of simulated elections with polling errors the size of these, so a miss like the ones above is inside our ranges, not a surprise. Polls against results, every election · Our forecast record · Pollster leaderboard
United States · 5 November 2024
2.2 pts