Governors run in different years and political environments. Some cycles have few eligible polls. These historical outcomes remain provisional archive values. Missing coverage is shown as unavailable, never as a zero-error year.
Select a cycle to inspect its error and the number of eligible estimates. The changing sample is part of the story.
Governor
Original archive calculation: final 1–14 days by median field date; latest estimate per firm/race, same-date ties averaged. Includes retained general, special and runoff contests within source cycles; not AAPOR’s methodology. Outcomes and exclusions are documented below.
| Cycle | Mean absolute error | Mean signed error | Estimates | Contests |
|---|---|---|---|---|
| 1998 | 7.49 | −4.7 | 66 | 25 |
| 1999 | Unavailable | — | 0 | 0 |
| 2000 | 3.97 | −1.3 | 22 | 9 |
| 2001 | 3.76 | +2.5 | 10 | 2 |
| 2002 | 4.92 | +3.5 | 98 | 32 |
| 2003 | 5.60 | +1.7 | 18 | 4 |
| 2004 | 2.31 | +0.8 | 14 | 7 |
| 2005 | 3.67 | −3.3 | 13 | 2 |
| 2006 | 4.81 | −1.0 | 116 | 29 |
| 2007 | 3.53 | +1.9 | 3 | 1 |
| 2008 | 4.07 | −1.6 | 25 | 9 |
| 2009 | 4.13 | +3.4 | 19 | 2 |
| 2010 | 3.83 | −0.3 | 120 | 31 |
| 2011 | 4.68 | +3.7 | 3 | 3 |
| 2012 | 4.66 | −2.1 | 32 | 9 |
| 2013 | 5.27 | +2.5 | 17 | 2 |
| 2014 | 4.20 | +2.5 | 104 | 26 |
| 2015 | 7.66 | +2.1 | 9 | 3 |
| 2016 | 6.04 | +4.3 | 41 | 12 |
| 2017 | 4.41 | −3.9 | 24 | 2 |
| 2018 | 4.62 | +0.6 | 110 | 28 |
| 2019 | 2.42 | −2.1 | 8 | 2 |
| 2020 | 7.44 | +7.0 | 24 | 6 |
| 2021 | 2.97 | +2.0 | 20 | 2 |
| 2022 | 4.37 | −0.5 | 139 | 22 |
Mean absolute margin error was 7.49 points in 1998 and 4.37 in 2022, compared with 2.97 in 2021. Across the covered cycles, the smallest observed miss was 2.31 in 2004; the largest was 7.66 in 2015.
The latest cycle contains 139 selected firm/race estimates across 22 contests. Its signed error, −0.5 points, overstated Republican margins. Changing races and pollsters limit comparisons; these governor outcomes remain provisional.
AAPOR’s weighted analysis · October 23–November 5 · raw D–R margin error
House district polls cover only 20 contests; this is not an estimate for all 435 districts. Presidential congressional-district polls concern electoral votes and are distinct from House races.
Only with substantial qualifications. The set of states and firms changes, odd-year contests differ from presidential-year races, and state turnout can matter greatly. These are observed errors in covered contests, not a controlled test of polling technology.
Federal series can be cross-checked against FEC and election-research returns. Governor outcomes remain provisional archive-supplied values, without independent count reconciliation. All New York and Connecticut governor records and Maine 2010 are excluded; the methodology page records exactly what was corrected or excluded.
Different windows can contain different pollsters and races. These are descriptive comparisons, not a causal experiment.
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.
