Polling Methodology

How logged polls are selected and how the displayed averages are calculated. Dataset date: November 3, 2026.

Coverage and information types

The site tracks Senate and governor races listed in its race directory. National approval and generic-ballot polls are separate indicators. The site does not provide individual House-race polling. “No recent public polls” means no qualifying poll is logged in the current window; it does not mean that no poll exists elsewhere.

At the dataset date, 71 Senate and governor races are tracked and 26 have a poll ending in the current 14-day window. A race can be tracked without recent public polling.

Individual poll results are reported figures. Averages are calculations from stored poll rows. Race context is explanatory content. Historical tracker figures are attributed context from other publishers. AI-assisted briefings and reader submissions are labeled separately. A polling average is not a forecast or probability of winning.

The 14-day average calculation

Race and national daily averages use the dataset date as their end date. A poll is in the window when its recorded field end date is on or before that date and strictly later than the date 14 days earlier. The window includes both endpoints as calendar dates. Field start dates, release dates and the day a visitor opens the page do not move the window.

For each candidate, the arithmetic mean is the sum of included candidate percentages divided by the number of included polls. A race margin is the first displayed candidate’s unrounded mean minus the Republican candidate’s unrounded mean. National generic-ballot averages use the stored Democratic-minus-Republican margin reported for each included poll.

Each included poll has equal weight. Sample size, pollster, recency within the window, population, sponsorship, mode and past accuracy do not change its weight. The calculation does not adjust for partisan bias, undecided voters or third-party support.

Calculations keep full precision. Candidate averages and calculated margins are displayed to one decimal place; raw poll results retain their stored precision. Rounding applies only to display, so subtracting two rounded candidate figures can differ slightly from the displayed margin.

Race averages require complete candidate results between 0 and 100. With no qualifying poll in the window, the current average is unavailable. Older polls remain in the table with their dates but are not carried forward. Historical chart points represent recorded polls, not a forecast.

Populations, samples and overlapping releases

Senate and governor averages accept verified likely-voter (LV) and registered-voter (RV) results; LV is preferred when the same release reports both. Adult-only race samples are excluded. National approval averages prefer adults, then registered voters, then likely voters. Generic-ballot averages accept LV and RV and prefer LV. The original population labels are retained.

Sample size is shown only when present in the stored record and does not affect weighting. A missing sample size is not treated as zero and is not estimated.

A poll qualifies only when its original source, full field dates, analytic sample size, population, geography, pollster identity and result are verified and recorded. Race and generic-ballot rows must use a general-election question. Alternate or composite ballot questions, ranked-choice final rounds, adult-only race samples, incomplete results and unverified records are excluded. Approval questions are averaged separately.

The same sample/release counts once within a race or national series. When one release reports several eligible populations, the population preference above selects one. For overlapping field periods from the same pollster in the same race, only the release with the newest field end date is retained. Non-overlapping releases count separately.

Sources and incomplete records

When an original-source URL is stored, it appears beside the individual poll. Rows with incomplete or unverified details remain visible in the poll list and are labeled as excluded from averages. The data record identifies missing or exclusion reasons.

0 race-poll records do not yet have a stored original-source link. Missing source links and other incomplete fields remain visible and are not filled with estimates.

For national approval, approve and disapprove are averaged separately using available values, so their poll counts can differ. Unknown results remain missing. The site does not infer disapproval as 100 minus approval or invent party percentages from a reported national margin.

Campaign or party-sponsored race releases are excluded as selectively released internal polls. Media-sponsored and openly crowdfunded releases must meet the same source and metadata requirements. Alternate ballot wording is not combined with general-ballot results.

Updates and freshness

“Dataset updated” refers to the date stored in the polling dataset (November 3, 2026), not the current time or the date of every poll. The repository stores a date rather than a verified update time. Rebuilding publishes the stored records and recomputes averages; the code does not establish that new polls are collected automatically each morning.

Daily describes the dated average series, not a guarantee of new polls or fresh publication each day. The newest poll end date, dataset date, briefing date and live-results update time are separate timestamps. Check the date beside each figure.

Uncertainty and historical context

Polls are snapshots and can disagree because of sampling and other sources of error. The site does not calculate a combined margin of error or confidence interval for averages. A small lead alone does not establish a statistical tie. The number of included polls is shown so sparse coverage is visible.

Rating context

Cook Political Report and Sabato’s Crystal Ball ratings are attributed judgments from those organizations, not poll results or calculations by this site. Where a rating’s original publication date or source link is missing, that information remains unverified. Long-term national snapshots are historical context, not a continuous average calculated by this site.

Reader poll and prediction tools

The reader poll is an unscientific, self-selected poll of visitors and does not represent the U.S. electorate. The chart uses rolling seven-day totals and displays a reader percentage only when that window contains at least 20 votes. National surveys remain separately labeled.

Interactive election predictions are choices made by users. They are not an editorial forecast, public opinion survey or published election result.