Presidential polls: the national vote and the state map.
The popular vote and the Electoral College are different outcomes. Start with a consistently evaluated modern series, then examine one firm’s longer history separately.
National accuracy can hide state misses.
AAPOR’s common method covers seven presidential elections. In 2016, national error averaged 2.1 points while state error averaged 5.7.
President · national popular vote
Average absolute margin error · lower is closer
AAPOR’s revised 2024-report historical series. Final fieldwork 0–13 days before or on Election Day; analytic weights account for duplicate and overlapping tracking estimates. No two-party normalization. Earlier counts are not included in this extracted table.
Inspect the chart data
| Cycle | Mean absolute error | Mean signed error | Estimates | Contests |
|---|---|---|---|---|
| 2000 | 3.90 | −3.3 | — | — |
| 2004 | 2.30 | +1.4 | — | — |
| 2008 | 2.20 | −0.1 | — | — |
| 2012 | 3.10 | −3.0 | — | — |
| 2016 | 2.10 | +1.0 | — | — |
| 2020 | 4.10 | +3.7 | — | — |
| 2024 | 2.60 | +2.5 | 60 | 1 |
What the national vote actually was.
Margins below use FEC candidate counts divided by all counted presidential votes, keeping third parties in the denominator.
| Election | Democratic candidate | Republican candidate | Actual D–R margin | Reconstructed mean polled margin* |
|---|---|---|---|---|
| 2000 ↗ | Al Gore | George W. Bush | D +0.5 | R +2.8 |
| 2004 ↗ | John Kerry | George W. Bush | R +2.5 | R +1.1 |
| 2008 ↗ | Barack Obama | John McCain | D +7.3 | D +7.2 |
| 2012 ↗ | Barack Obama | Mitt Romney | D +3.9 | D +0.9 |
| 2016 ↗ | Hillary Clinton | Donald Trump | D +2.1 | D +3.1 |
| 2020 ↗ | Joe Biden | Donald Trump | D +4.5 | D +8.2 |
| 2024 ↗ | Kamala Harris | Donald Trump | R +1.5 | D +1.0 |
*Approximate reconstruction: actual national margin plus AAPOR’s rounded mean signed error. This is not a contemporaneously published polling average. Gore (2000) and Clinton (2016) led the popular vote while losing the presidency.
Gallup’s final estimates, back to 1936.
One polling organization, changing methods, and varying fieldwork windows. This is historical context—not a 90-year industry average.
Absolute error in the two-candidate-normalized margin
For this historical chart only, both the poll and outcome are normalized to D + R = 100%. Historical estimates often allocated undecided voters; earlier outcome shares are rounded. Normalizing denominators does not make sampling methods or final-poll timing uniform.
Inspect all 20 historical estimates and their definitions
| Year | Poll D | Poll R | Actual D | Actual R | Two-party margin miss |
|---|---|---|---|---|---|
| 1936 | 55.7 | 44.3 | 60.8 | 36.5 | 13.57 |
| 1940 | 52.0 | 48.0 | 54.7 | 44.8 | 5.95 |
| 1944 | 51.5 | 48.5 | 53.4 | 45.9 | 4.55 |
| 1948 | 44.5 | 49.5 | 49.5 | 45.1 | 9.97 |
| 1952 | 49.0 | 51.0 | 44.4 | 55.1 | 8.75 |
| 1956 | 40.5 | 59.5 | 42.0 | 57.4 | 3.51 |
| 1960 | 51.0 | 49.0 | 49.7 | 49.5 | 1.80 |
| 1964 | 64.0 | 36.0 | 61.1 | 38.5 | 5.31 |
| 1968 | 42.0 | 43.0 | 42.7 | 43.4 | 0.36 |
| 1972 | 38.0 | 62.0 | 37.5 | 60.7 | 0.37 |
| 1976 | 48.0 | 49.0 | 50.1 | 48.0 | 3.17 |
| 1980 | 44.0 | 47.0 | 41.0 | 50.7 | 7.28 |
| 1984 | 41.0 | 59.0 | 40.6 | 58.8 | 0.31 |
| 1988 | 44.0 | 56.0 | 45.6 | 53.4 | 4.12 |
| 1992 | 49.0 | 37.0 | 43.0 | 37.4 | 6.99 |
| 1996 | 52.0 | 41.0 | 49.2 | 40.7 | 2.37 |
| 2000 | 46.0 | 48.0 | 48.4 | 47.9 | 2.66 |
| 2004 | 49.0 | 49.0 | 48.3 | 50.7 | 2.49 |
| 2008 | 55.0 | 44.0 | 52.9 | 45.7 | 3.73 |
| 2012 | 49.0 | 50.0 | 51.1 | 47.2 | 4.94 |
Does waiting make the picture clearer?
Different windows can contain different pollsters and races. These are descriptive comparisons, not a causal experiment.
Mean absolute margin error, by days to election
The interactive timing chart loads with JavaScript. Every value is also in the downloadable dataset.
Latest per firm/race within each non-overlapping midpoint window.
Read the case files.
1936: millions of returns, the wrong electorate
A huge voluntary return cannot guarantee a representative sample.
Read the evidence and analysis →1948: the Dewey lead that disappeared
The sampling and timing of a poll matter more than confidence in its winner call.
Read the evidence and analysis →1980: the right winner, a much larger victory
Calling the winner correctly can conceal a substantial error in the victory margin.
Read the evidence and analysis →2016: small state margins, consequential polling misses
A reasonably close national popular-vote estimate can coexist with consequential errors in the states that decide the presidency.
Read the evidence and analysis →Iowa 2024: a striking miss, not a settled explanation
One prominent outlier can dominate the election story. Evaluate it against the result and other polls without pretending that its cause has been established.
Read the evidence and analysis →