Pollster Accuracy Rankings: Who Got 2024 Right

2024 was a good year for polling, and the receipts say so

The AAPOR Task Force on 2024 Pre-Election Polling analyzed 116 general election polls conducted between October 23 and November 5, 2024, and concluded that public polls painted an essentially accurate picture of an extraordinarily close contest. By the task force's accounting, state-level polls were the most accurate since 1944, and national polls performed about average by historical standards (AAPOR, 2025). After the documented misses of 2016 and 2020, that is a measured, sourced rehabilitation.

One caveat survived: polls again understated the Republican vote share relative to Democrats, for the third consecutive presidential cycle, though by less than in 2016 or 2020 (AAPOR, 2025). Split Ticket's post-election review put it bluntly: there does not appear to be a single state where the polls underestimated Harris; Trump consistently outran his numbers (Split Ticket, 2024).

Who actually topped the error tables

The Silver Bulletin's 2024 review computed simple average error for pollsters with at least five qualifying late polls. The leaders: OnMessage Inc. at 1.2 points, AtlasIntel at 1.5, and Patriot Polling at 1.5, with Mitchell Research, Echelon Insights, TIPP Insights, and The Washington Post all near 2 points (Silver Bulletin, 2024). OnMessage and AtlasIntel also posted the lowest bias among qualifying firms, overestimating Democrats by just 0.1 points on average (Silver Bulletin, 2024).

The ActiVote MVP rankings, using a weighted score across 10 national polls, 70 swing-state presidential polls, non-swing-state polls, 29 Senate polls, and 5 governor polls, reached a compatible answer by a different road: AtlasIntel first with a 1.38 percent average error, InsiderAdvantage second, OnMessage third, with Trafalgar, Rasmussen, and Patriot Polling in the top performers and Emerson seventh (ActiVote, 2025). Two independent scoring systems, one overlapping leader board. That convergence is the strongest evidence in this report.

The Selzer miss, because honesty requires it

The most famous poll of 2024 was also its largest documented miss. J. Ann Selzer's final Des Moines Register Iowa poll showed Harris ahead 47 to 44 in a state Trump had won by 8 points in 2020; Trump carried Iowa by about 13 points, a miss on the margin of roughly 16 points. Split Ticket called it one of the worst polls of the cycle (Split Ticket, 2024). Selzer had been among the most decorated state pollsters in America for decades.

The lesson is not that Selzer was secretly bad. The lesson is that a single poll, even from an excellent pollster, carries irreducible error, and that a single cycle, even a clean one, cannot separate skill from luck for firms with thin portfolios. This report ranks by documented 2024 accuracy because that is its assignment, and it flags the uncertainty because that is the truth.

Single-cycle accuracy versus long-run quality

The Silver Bulletin maintains two different measurements, and they disagree instructively. Its 2024 single-cycle error table crowns OnMessage and AtlasIntel. Its overall pollster ratings, updated January 2026 and blending accuracy across many cycles with transparency and methodology, rank The Washington Post, Marquette University Law School, and New York Times/Siena College at the top (Silver Bulletin, January 2026). Both tables are correct. They answer different questions.

ActiVote's data adds a structural finding: the number one pollster of 2024 used non-probability methods, and no probability-based pollster reached its top ten, while no major media outlet made the top ten either (ActiVote, 2025; Denver Gazette, 2025). Whether that reflects a durable methodological shift or one cycle's conditions is exactly the kind of question the 2026 midterms will test. The framework here scores the record, not the reputation. If that ordering produces discomfort, the discomfort belongs to the reader, not the data.