Methodology & Corrections

Where the numbers come from

We record the top-line results of publicly released national and state polls from named pollsters (for example Quinnipiac, Marist, Emerson, Reuters/Ipsos, Economist/YouGov), taken from each pollster's own release or from news reports of it, and link to the source. Poll results are facts; we credit the pollster every time.

We do not copy other sites' polling averages, charts, tables, articles or photos. Every average and chart on this site is calculated and drawn by us.

Daily averages

Each day we average all polls that finished in the previous 14 days (the battle board uses 14 days, the head-to-head view 30). On days with no recent poll the line pauses. These are averages, not forecasts. The "Since 2025" approval view shows long-term context as reported by the named trackers.

Reader poll

Our reader poll is an unscientific straw poll. Only votes from U.S. IP addresses are counted in the results, one vote per question per device per day, with a bot check. The chart shows a 7-day total and hides days with fewer than 20 votes. It shows how our readers feel, not how all Americans feel, and we always show it next to the national poll average for comparison.

Ratings

Race ratings are the published ratings of the Cook Political Report and Sabato's Crystal Ball, credited by name.

AI use

The daily briefing is drafted with AI from the numbers in our dataset, then checked against that data. AI never creates poll numbers, ratings or candidate information.

Sampling error

Every poll has sampling error, usually ±3 to ±5 points. A lead smaller than that is a statistical tie.

Corrections

Found an error? Send the race, the figure and a source to hello@midtermresultslive.com. We correct confirmed errors promptly. Corrections are free for everyone, including campaigns.