How close each polling institute’s final polls came to the result. Errors are in percentage points; lower is better.
| Rank | Institute | Elections | Polls | Mean error | Recent error | Margin error |
|---|---|---|---|---|---|---|
| 1 | Emerson CollegeUnited States | 5 | 5 | 3.1 | 3.1 | 4.5 |
| 2 | InsiderAdvantage (R)United States | 3 | 3 | 4.9 | 4.9 | 4.8 |
| 3 | Quantus Insights (R)United States | 5 | 5 | 6.0 | 6.0 | 10.3 |
| 4 | JMC Analytics (R)United States | 4 | 4 | 7.5 | 7.5 | 9.7 |
| 5 | Public Policy Polling (D)United States | 6 | 9 | 7.5 | 7.5 | 6.4 |
| 6 | University of New HampshireUnited States | 7 | 7 | 10.0 | 10.0 | 14.1 |
Fewer than 3 elections analysed: shown for reference, not ranked.
| Institute | Elections | Polls | Mean error | Recent error | Margin error |
|---|---|---|---|---|---|
| FM3 Research (D)United States | 1 | 1 | 0.8 | 0.8 | 2.1 |
| Cygnal (R)United States | 2 | 2 | 2.3 | 2.3 | 2.6 |
| David J WolfsonUnited States | 1 | 1 | 2.4 | 2.4 | 6.6 |
| GBAO (D)United States | 1 | 1 | 2.6 | 2.6 | 3.6 |
| University of Texas/Texas Politics ProjectUnited States | 1 | 1 | 3.4 | 3.4 | 15.4 |
| Trafalgar Group (R)United States | 1 | 3 | 3.6 | 3.6 | 3.2 |
| Fabrizio, Lee & Associates (R)United States | 1 | 1 | 3.7 | 3.7 | 6.3 |
| Meeting Street Insights (R)United States | 1 | 1 | 3.9 | 3.9 | 6.4 |
| UMass Amherst/YouGovUnited States | 1 | 1 | 3.9 | 3.9 | 1.7 |
| Tavern Research (D)United States | 2 | 2 | 4.1 | 4.1 | 4.8 |
| Mitchell ResearchUnited States | 1 | 1 | 4.1 | 4.1 | 13.2 |
| YouGovUnited States | 2 | 2 | 4.2 | 4.2 | 7.4 |
| Blueprint Polling (D)United States | 2 | 2 | 4.2 | 4.2 | 10.8 |
| Mitchell Research & CommunicationsUnited States | 1 | 1 | 4.3 | 4.3 | 13.0 |
| Impact Research (D)United States | 1 | 1 | 4.4 | 4.4 | 12.5 |
| OnMessage Inc. (R)United States | 1 | 1 | 4.4 | 4.4 | 6.7 |
| SurveyUSAUnited States | 1 | 1 | 4.5 | 4.5 | 2.5 |
| Deep Root Analytics (R)United States | 1 | 1 | 4.6 | 4.6 | 1.7 |
| WickUnited States | 1 | 1 | 4.9 | 4.9 | 8.3 |
| Victory Phones (R)United States | 1 | 1 | 5.0 | 5.0 | 10.2 |
| The CitadelUnited States | 1 | 1 | 5.0 | 5.0 | 1.7 |
| State NavigateUnited States | 1 | 2 | 5.3 | 5.3 | 26.0 |
| Saint Anselm CollegeUnited States | 1 | 1 | 5.5 | 5.5 | 1.4 |
| UT TylerUnited States | 2 | 2 | 5.5 | 5.5 | 9.2 |
| Data for Progress (D)United States | 1 | 1 | 6.0 | 6.0 | 12.0 |
| Remington Research Group (R)United States | 2 | 2 | 6.1 | 6.1 | 8.7 |
| co/efficient (R)United States | 2 | 2 | 6.4 | 6.4 | 3.1 |
| KTSP/SurveyUSAUnited States | 1 | 1 | 7.0 | 7.0 | 16.9 |
| Pulse Decision Science (R)United States | 1 | 1 | 7.0 | 7.0 | 10.2 |
| The Public Sentiment InstituteUnited States | 2 | 2 | 7.2 | 7.2 | 11.6 |
| McLaughlin & Associates (R)United States | 2 | 2 | 7.5 | 7.5 | 9.5 |
| University of GeorgiaUnited States | 1 | 1 | 7.7 | 7.7 | 6.6 |
| Tulchin Research (D)United States | 2 | 2 | 7.8 | 7.8 | 16.5 |
| Suffolk UniversityUnited States | 1 | 1 | 7.8 | 7.8 | 15.7 |
| SoCal Strategies (R)United States | 1 | 1 | 8.0 | 8.0 | 16.0 |
| Bowling Green State University/YouGovUnited States | 1 | 1 | 8.0 | 8.0 | 0.9 |
| Zenith ResearchUnited States | 1 | 1 | 8.5 | 8.5 | 17.0 |
| UpONE Insights (R)United States | 1 | 1 | 9.2 | 9.2 | 10.7 |
| Change Research (D)United States | 2 | 2 | 9.8 | 9.8 | 15.8 |
| Public Opinion Strategies (R)United States | 2 | 2 | 10.2 | 10.2 | 8.9 |
| Sacred Heart University/WFSB/CT InsiderUnited States | 1 | 1 | 10.5 | 10.5 | 0.6 |
| High Point UniversityUnited States | 1 | 1 | 11.3 | 11.3 | 23.0 |
| Marquette UniversityUnited States | 1 | 1 | 11.3 | 11.3 | 35.5 |
| GQR (D)United States | 1 | 1 | 11.8 | 11.8 | 23.6 |
| Emerson College/WPRI-TVUnited States | 1 | 1 | 12.5 | 12.5 | 9.9 |
| Harper Polling (R)United States | 1 | 1 | 12.6 | 12.6 | 19.0 |
For every past election with a stored result, each institute’s polls whose fieldwork ended (or, without fieldwork dates, that were published) in the 21 days before election day are compared with the result. A poll’s error is the mean absolute difference, over the parties or candidates it shares with the result, between its figure and the result; its margin error is how far its gap between the result’s top two was from the real gap. Each institute’s errors are averaged over its polls in an election, then over elections. The recent error covers elections held in the last 2 years.
Institutes are ranked by mean error once they have polled at least 3 elections. Ranked institutes count somewhat more or less in the Pollytic average; see the methodology. Polls measure opinion during fieldwork, not on election day, so part of every error is real late movement.