Overview
Pollsmax tracks publicly released political polls for U.S. Senate, governor, presidential primary, and national indicator races. Each poll is entered with the topline results reported by the pollster, along with field dates, sample type, and sample size when available. Polling averages on race pages are designed to summarize recent survey evidence—not to predict outcomes on their own.
Poll collection
We add polls when a reputable survey is released with citable toplines. Polls are dated by the end of the field period in Eastern Time. When a release covers multiple days of interviewing, the poll is placed on the last field date. Individual poll rows on a race page show exactly what was reported; the trend line reflects our weighted average, not a pollster’s internal smoothing.
Pollster names and metadata
Pollster names are normalized to a canonical form so the same firm does not appear under multiple spellings across pages. Sample sizes and electorate types (for example likely voters or registered voters) are displayed as reported. Margin-of-error figures are shown when the original release includes them. Polls flagged for exclusion from averaging remain visible in tables but do not contribute to the trend line.
Race-page polling averages
For Senate and governor general-election pages, the polling average is a recency-weighted mean of all polls dated on or before each chart day. More recent surveys receive higher weight; on state race pages the recency curve adapts to how many polls exist in the series so sparse races do not overreact to a single new survey. Democratic and Republican lines are computed separately, but when both candidates appear in the same poll the pair is treated jointly for outlier detection.
Outlier downweighting
When at least three polls with paired Democratic and Republican toplines are available, we compute a robust consensus margin (median Democratic minus Republican) and measure each poll’s distance from that consensus using a median absolute deviation (MAD) scale. Polls far from the pack receive lower weight but are not deleted. On primary pages and some multi-candidate fields, outlier downweighting is disabled so every valid reading counts equally aside from recency.
Partisan pollster adjustment
Polls from pollsters labeled with a party sponsor tag may receive a small margin adjustment before averaging when the sponsored side’s position is plausible. The adjustment is skipped when the sponsoring party is trailing by a wide margin in that survey, and it is not applied in races where all polls are treated uniformly. This step reduces pollster house-effect skew without removing partisan polls from the dataset.
National indicators
Trump approval and the generic congressional ballot use national poll series. Each day’s value is a sample-size-weighted mean of polls in the field on or before that date, with older surveys downweighted exponentially (approximately a two-week effective window). A rolling window fallback is used when decay weighting alone would leave gaps. These series move slightly day to day even between new releases because older polls gradually age out of the average.
Presidential primary averages
2028 presidential primary charts use a fixed lookback window (50 days of field dates) and weight polls by reported sample size only—no exponential decay within the window. Each candidate’s average is calculated independently; a candidate missing from a particular poll does not affect other candidates’ averages for that day.
What we show on each page
Poll tables list individual releases in reverse chronological order. Charts plot each poll as a point and overlay the weighted average as a trend line. Latest values in race summaries and hub maps come from the most recent day on that trend line with valid data for both major candidates (or all listed candidates on primary pages). Poll averages feed forecast models separately; those projections use additional inputs beyond polling alone.