Transparency
Methodology
Which polls we include
Full weight: high-quality, transparent pollsters, most of them AAPOR Transparency Initiative members with good past accuracy. These include NYT/Siena, Marquette Law, Marist, Emerson, YouGov (Economist and CBS), Quinnipiac, Fox News (Beacon/Shaw), CNN/SSRS, Washington Post/ABC/Ipsos, Pew, AtlasIntel, and state university polls such as Suffolk, UNH, Muhlenberg and Roanoke.
Included with adjustments: partisan or sponsored polls (campaign internals and party groups) are moved 3 points toward the other side. High-volume online or IVR trackers, such as Rasmussen, Trafalgar, InsiderAdvantage, Quantus and ActiVote, are capped. Only their 2 newest polls count, and no single pollster can make up more than 20% of a race's average.
Left out: opt-in website or social media polls, polls that don't publish field dates, sample size or question wording, and firms caught fabricating data.
How the average is calculated
Each poll's weight = pollster rating × sample size factor (square root of n, capped at 3,000) × recency decay (weight halves every 14 days, measured from the middle of the field dates) × population factor (likely voters 1.0, registered voters 0.9, all adults 0.75).
Polls older than 90 days drop out, or after 365 days in races with little polling. Each pollster's estimated house effect (its typical lean compared with other pollsters) is subtracted before averaging.
Sources
Pollster ratings: Silver Bulletin. FiveThirtyEight's ratings stopped when ABC shut that site down in 2025.
Poll listings: imported daily from VoteHub's public data, plus toplines entered directly from each pollster's own release. Every poll links to its original source.
Results: official state results, compared against our final pre-election average on the Results page.